JOURNAL OF SEMANTICS AN INTERNATIONAL JoURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE
MANAGING EDITOR : PETER BoscH ( IBM Scientific Centre, Heidelberg and University of Osnabriick) AS SOC lATE EDITORS: MANFRED KRIFKA ( University of Texas, Austin) R o VAN DER SANDT ( University ofNijmegen) REVIEW EDITOR: TIBOR K1ss (IBM Scientific Centre, Heidelberg) ASSISTANT EDITOR: BIANKA BusCHBECK -WOI.F (IBM Scientific Cenrre, Heidelberg)
o
EDITORIAL BOARD : N. AsHER ( University of Texas, Austin) R. BARTSCH ( University of Amsterdam) M. BIERWISCH (MPG and Humboldt University Berlin) B. BoGURAEV (Apple Computer Inc) M. BoRILLO ( University of Toulouse) G. BROWN (University of Cambridge) K. BROWN (University of Essex) G. CHIERCHIA ( University of Milan) 0. DAHL (University of Stockholm) S. C. GARROD ( University of Glasgow) B. GEURTS (University of O snabriick) M. HERWEG (IBM Scientific Centre, Heidelberg) L. R. HoRN (Yale University) J. jACOBS ( University of Wuppertal) P. N. JoHNSON-LAIRD ( Princeton University) M. KAMEYAMA (SRI lnrernational, Menlo Park)
H. KAMP ( University of Stuttgart) S. LEVINSON (MPI Nijmegen) S. UiBNER ( University of Dusseldorf) SIR JoHN LYONS ( University of Cambridge) A. MANASTER-RAMER (Wayne State University) J. M C A LE ( University of Chicago) M. MOENS ( University of Edinburgh) F. J. PELLETIER ( University of Alberta) M. PINKAL ( University ofSaarbriicken) T. SANFORD ( University of Glasgow) R. ScHA (University of Amsterdam) A. VON STECHOW ( University of Tiibingen) M. STEEDMAN ( University of Pennsylvania) W. WAHLSTER (DFKI, Saarbriicken) B. WEBBER ( University of Pennsylvania) H. ZEEVAT ( University of Amsterdam) T. E. ZIMMERMANN ( University of Stuttgart)
c W Y
EDITORIAL ADDRESS: Journal of Semantics, c/ o Dr P. Bosch, IBM Germany Scientific Cenrre, Vangerowstr. 18, D-69115 Heidelberg, Germany. Phone: (49-6.1..1.1-) 59-4.1.51/4483. Telefax: (49-6.1..1.1-) 59-J.I.OO. Email:
[email protected],
[email protected] ©Oxford University Press 1995 All rights reserved; no part of this publication may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise without either the prior written
permission of the Publishers, or a licence permitting restricted copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1 P
Congress Street, Salem, Mass o 1 970.
9HE,
or in the USA by the Copyright Clearance Center,
27
The Journal of Semantics is published quarterly in February, April, August and November by Oxford University Press. Subscription is S I33·
$120 per
year. Second class postage paid at Newark NJ and at additional mailing offices. ISSN
0167-
POSTMASTER: send address corrections to The Journal of Semantics, c/o Virgin Mailing and Distribution, Cargo Atlantic,
10 Camptown Road,
Irvington NJ
07111-IIO\. USA.
For Subscriptions information please see inside back cover
JOURNAL OF SEMANTICS Volume
12
Number
2
SPECIAL ISSUE: LEXICAL SEMA NTICS PART
2
CONTENTS GEOFFREY NUNBERG
Transfers of Meaning
109
jAMES PusTEJOVSKY AND PIERRETTE BoUILLON Aspectual Coercion and Logical Polysemy
I
33
NICHOLAS AsHER AND PIERRE SABLAYROLLES
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
I63
Book Review
2I I
Linguistics and Philosophy
A Journal of Natural Language Syntax, Semantics, Logic, Pragmatics, and Processing Editor-in-Chief:
Journal
Gregory N. Carlson
Highlight
Associate Editors:
Jeroen Groenendijk, Martin Stokhof, Laurence R. Hom, Pauline Jacobson, William C. Rounds, Erhard Hinrichs, Mark Richard, Scott Soames
Executive Board:
Richmond H. Thomason, Barbara Partee, Robin Cooper
Linguistics and Philosophy is a journal for studies focused on
natural language, and is of interest to practitioners in the disci plines covered in the title. Although the field thus described is so extensive that a complete listing of relevant topics is precluded, at least the following specifically fall within it: - tradional areas in the philosophy of language such as meaning, truth, reference, description, entailment, speech acts - tradi tional areas of linguistics such as syntax, semantics and prag matics (when the studies are of sufficient explicitness and generality to be also of philosophical interest) - aspects of artificial intelligence concerned with language such as computational linguistics and natural language processing - systems of logic with strong connections to natural language: modal logic, tense logic, epistemic logic, intensional logic - philosophical questions raised by linguistics as a science: lin guistics methodology, the status of linguistics theories, the nature of linguistic universals - philosophically interesting problems at the intersection of lin guistics and other disciplines: language acquisition, language and perception, language as a social convention.
Subscription Information 1995, Volume 18 (6 issues) Subscription Rate: NLG 488.00
Private Subscription Rate: NLG
including postage and handling.
ISSN 0165-0157 279.00. 195.00 I USD 85.00. Prices are
I USD
P.O. Box 322, 3300 AH Dordrecht, The Netherlands P.O. Box 358. Accord Station, Hingham,
MA 02018-0358, U.S.A.
Kluwer academic publishers
C
o
m p
u
t
a
t
1 o
_11� a
I
� � r1 g 1.:! 1
s t
i
_
c
s_
Computational linguistics is the foremost journal devoted exclusively to computational analyses of natural language.
It encompasses AI research in language, linguistics, and the psychology of
Julia Hirschberg Editor II
language processing and performance.
1995 lnstirurional Rate: $90.00. Add
With each issue providing applied and
$16.00 posagc and handling ouuide
theoretical papers, book reviews, technical correspondence, and letters to
U.S.A. Canadians add additional 7% GST. Prices subject to change without notice. Prepayment is required. Send
the editor, the journal presents a
chcdt dfllwn
stimulating forum for the exchange of
fund.., MC, VISA, or AMEX number
ideas and trends in computational
MIT Pras Joaraab 55
language.
Computational Linguistics is the
againsr a U.S. bank in U.S. ro:
Hayword Succt
Cambridgr. MA02142-1399
TEL: 617-253-2889
fA)(, 617-258-6779
[email protected] journal serving a rapidly expanding interdisciplinary field and should be part of every research library colleccion. Recommend CL to your librarian today.
Individual subscriptions arc available only through mcmbcrlhip in
rhc
Associuion for CompurarionaJ Linguistics.
Please con�:aet: Priscilla Rasmussen Associarion for Computarional Linguisrics (ACL)
Published quarterly by The MIT Press for me AJsociarion for Computacional Linguistia ISSN 0891-2017. Volume 21 forlhcoming.
P.O.Box 6090
Some!SCI, NJ 08875 USA
Td: 908-873-3898
[email protected] Journal of Logic, Language and Information Editor:
Peter Giirdenfors
Dept. ofPhilosophy, Lund University, Sweden Section Editors:
D. Gabbay Daniele Mundici Stanley Peters, R. Stalnaker, Johan van Benthem
Journal Highlight
This is the official publication of the European Association for Logic, Language, and Information. The scope of the journal is the logical and computational foundations of natural, formal, and programming languages, as well as the different forms of human and mechanized inference. It covers the logical, linguistic, and information-theoretic parts of the cognitive sciences. Examples of main subareas are Intensional Logics including Dynamic Logic; Nonmonotonic Logic and Belief Revision; Constructive Logics; Complexity Issues in Logic and Linguistics; Theoretical Problems of Logic Programming and Resolution; Categorial Grammar and Type Theory; Generalized Quantification; Information-Oriented Theories of Semantic Structure like Situation Semantics, Discourse Representa tion Theory, and Dynamic Semantics; Connectionist Models of Logical and Linguistic Structures. The emphasis is on the theoretical aspects of these areas. The purpose of the journal is to act as a forum for researchers interested in the theoretical foundations of the above subjects and their interdisciplinary connections, with an emphasis on general ideas increasing coherence. Journal of Logic, Language and Information is abstracted/Indexed in Mathematical Reviews; Current Mathematical Publications; MathSci; lnspec Database; Zentralblatt fOr Mathematik; Sociological Abstracts; Unguistic and Language Behavior Abstracts; MLA International Bibli ography Subscription
Information
ISSN 0925-8531
1995, Volume 4 (4 issues)
Subscription Rate: NLG 435.00 I USD 249.00, including postage and handling. Special rate for members of the European Foundation for Logic, Language and Information available. For more information, please write to the FoLU Bureau, Muidergracht 24, 1018
TV
FWI,
University of Amsterdam, Plantage
Amsterdam, The Netherlands
P.O. Box 322, 3300 AH Dordrecht, The Netherlands P.O. Box 358, Accord Station, Hingham,
MA 02018-0358,
U.S.A.
Kluwer academic publishers
Highly respected for its sustained excellence in quality of scholarship, Linguistic Inquiry leads the field in research on current topics in linguistic theory. Issue after issue, LI keeps you informed of new theoretical developments by presenting the latest in international research. Sc:Ject Forthcoming Articles 26:1 Feature Geometry and Feature Spreading
Morris Hnllt
26:2 "Somethmg Else" for the Bmding Theory Pctrr Culiclltrtr aud Rnyfacktmlll{f 26:3 Marking Statements, Complexity and Simplirication Procedures A11drta Calnbrcst 26:4 NC. licensing and Underspecification in Optimality Theory Ju11ko /lc1, Annm Mtslcr. aud Jove Pad�ell Unguistic /uquiryWorld-Wide Web (WWW) Home Pase URL: http://www·mitprcss.mit.edu/irnl.s UCB/CSD 9 I /6 1 8), Computer Science
polysemy in lexicology and lexicography',
Division, University of California, Berke
in K. V. Tolonen
ley.
H amm Tommola, T. Salmi & J. Schopp (eds), Proceedings of Eurale."l." II, Tampere, Finland.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Natural Language and
Syntax and Semantics,
Atkins, B. ( 1 99 1 ), ' Predictable
Pustejovsky, J. ( 1 993), Type coercion and
Jackendoff, R. ( 1 992), 'Mme. Tussaud meets
Linguistic Theory ,
&
meaning shift: some linguistic properties of lexical implication rules', in J. Puste
Pustejovsky, J. ( 1 99 1 ), The generative lexi
Press, Cambridge, MA.
the binding theory',
Ostler, N.
Journal ofSemantics
12: I J J-I62
©Oxford Universiry Press I 995
Aspectual Coercion and Logical Polysemy J A M E S P U S T EJ O V S K Y Computer Science Department, Brandeis University P I E RRE TTE B O U I L L O N ISSCO, University ofGeneva
Abstract
1
P O LY S E M Y A N D S E M A N T I C E X P RE S S I VE NE S S
Recently, work in computational semantics and lexical semantics has made an interesting shift. Motivated by a concern for lexical organization and global coherence in the structure of the language lexicon, some researchers have moved towards more expressive semantic descriptions, as well as more powerful methods of composition. 1 There has been some concern expressed, however, as to the general applicability of type-changing operations such as coercion, as well as concerning the power of generative mechanisms operating in the lexicon and semantics. In this paper, we address these concerns directly, and show that, upon closer examination, these critiques are not substantiated by the linguistic data. Without a proper notion of constraints on coercion, however, there can indeed be overgeneration of interpretations in the semantics, and in fact, the notion of conditions on coercion has always been integral to the basic spirit of generative lexicons. In particular, we examine the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Recent work in computational semantics and lexical semantics has made an interesting shift. Motivated by a concern for lexical organization and global coherence in the structure of the lexicon, some researchers have moved towards more expressive semantic descriptions, as well as more powerful methods of composition. There has been some concern expressed, however, as to the general applicability of type-changing operations such as coercion, as well as concerning the power of generative mechanisms operating in the lexicon and semantics. In this paper, we address these concerns directly, and show that, upon closer examination, these critiques are not substantiated by the linguistic data. Without a proper notion of constraints on coercion, however, there can indeed be overgeneration of interpretations in the semantics, and in fact the notion of conditions on coercion has always been integral to the basic spirit of generative lexicons. In particular, we examine the consrraints on type coercion in complement constructions of aspectual predicates in English and French. What we discover is a natural explanation for the behavior of coercion that makes reference to different types of event selection while also addressing the polysemous nature of aspectual verbs.
134
Aspecrual Coercion and Logical Polysemy
Argument structure: Specification of number and type of logical arguments. 2. Event structure: Definition of the event type of a lexical item or phrase. Sorts include STATE, PROCESS, and TRANSITION. J. Qualia structure: Composed of FORMAL, CONSTITUTIVE, TELIC and AGENTIVE roles.2 4· Lexical inheritance structure: Identification of how a lexical structure is related to other structures in the type lattice. r.
A set of generative devices connects these four levels, providing for the compositional interpretation of words in context. The exact nature of these devices will determine the polymorphic expressiveness of the semantics in fairly definite ways. The best-studied illustration of this is the phenomenon of type coercion , but it is by no means the only one. Copestake & Briscoe ( 1 992) model the mechanisms of generative lexicon
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
constraints on type coercion in complement constructions of aspectual predicates in English and French. What we discover is a natural explanation for the behavior of coercion that makes reference to different types of event selection while also addressing the polysemous na.ture of aspectual verbs. We will assume some general familiarity with the framework of a generative lexicon (GL), as outlined in Pustejovsky ( 1 99 1 a, 1993), and Copestake ( 1 993). We feel it is important, however, to clarify the motivating principles and general methodology behind this work, since it is crucial to the analysis taken in this paper. The empirical study of the range and limits of type change and co composition operations in natural language is an essential part of research in formal semantics. The advantages accompanying generative mechanisms and the characterization oflanguages as polymorphic in well-defined ways allow us to overcome the explanatory inadequacies inherent in traditional approaches to lexical design and semantic projection, what Pustejovsky & Boguraev ( 1 993) call word sense enumeration approaches. In order to help characterize the generative power of natural languages in terms of semantic expressiveness, it is natural to think in terms of semantic systems with increasing functional power. Furthermore, a natural way of capturing this might be in terms of the type system which the grammar refers to for its interpretation. There are reasons for describing semantic systems as falling on a hierarchy of increasing expressive power and it seems clear at this point that the current enumerative techniques for lexical description are too impoverished adequately to describe the richness of semantic data, much less to explain either how word senses relate to one another or the creative use of words in novel contexts. A generative lexicon of the type we assume can be characterized as a system involving at least the following four levels of representations:
James Pustejovsky and Pierrette Bouillon
r35
theory as a type system for a lexical knowledge base. Pustejovsky & Boguraev ( I 993) extend this view into the compositional semantics by having the operations make direct reference to the types within the system. The qualia structure along with the other representational devices (event structure and argument structure) can be seen as providing the building blocks for possible object types. The typing information mentioned above comes together in the lexical representation for verbs as well. 2 L I N G U I S T I C E V I D E N CE F O R C OERC I O N
(I) a. John began to read the book. (vP[+ INF] ) b. John began reading the book. (vP[+ PRG]) c. John began the book. (NP) (2) a. Jean a commence a lire le livre. (vP[+INF]) b. Jean a commence le livre. (NP) This deep type is able to project to one of three possible surface forms in English, and two forms in French, depending on which coercion rule applies. There is, however, only one semantic type being selected for, and the clustering of the particular syntactic forms appearing as surface complement types in (I) and (2) are systematically projected by virtue of this semantic type. That is, a verb such as begin or commencer, selecting for an event, will paradigmatically allow for the expression of the grammatical forms shown above, assuming surface syntactic constraints are satisfied. For this reason, the structuring of this kind of linguistic knowledge, where this event type has syntactic expression as any one of the surface types in (I) or (2) is called a lexical conceptualparadigm ( lcp ) (c£ Pustejovsky & Anick, I 98 8). In this view, the NPs, the book in ( I c) and le livre in (2b), are coerced to the appropriate type required by its governing verb, in this case an event. What makes coercion possible in these cases is the availability of the selected type, given as part of the NP's qualia structure , indicating, for example, that the TELIC role for book is the event of reading, while the AGENTIVE role is an event of writing. The result of applying this coercion operator to an NP is effectively to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The phenomenon of multiple subcategorization has motivated much of the type changing literature. The approach taken in generative lexicon theory builds on the ideas developed by Partee & Rooth ( I 98 3) and Klein & Sag ( I 98 5 ), while attempting to derive the syntactic expression of a verb's complement from a deep semantic type assignment, together with syntactic constraints. For example, in the well-studied case of aspectual verb complementation in (I) and (2) below, the verbs begin and commencer carry a 'deep type' selecting for an event in complement position.
1 36
Aspecrual Coercion and Logical Polysemy
create an extension of the NP meaning, called a metonymic reconstruction . In the case of NP, the book , for example, the coercion operators provide two event interpretations, namely, reading the book and writing the book . This interpretation is produced by virtue of the type of the selected complement and the availability of such types in the qualia structure of the complement itsel£ It is important to point out that this is a semantic 'reconstruction' rather than a syntactic one.
3
PROBLEMS W I T H U N C O N ST R A I N E D C OE R C I O N
complement coercion. The first apparent counterexamples to the general application of type changing operations argue that commencer does not universally allow NP complements with a coerced interpretation (c£ Goddard & Jayez, 1 993). For example, the NPs in (3) below do not appear to allow the analogous coerced readings that the sentences in (I) and (2) allow. Although the qualia for highway and dictionary presumably make reference to the events of driving and referencing, respectively, these interpretations are not available for the sentences in (3) and (4). Thus, the system would appear to overgenerate interpretations when no constraints on the application of type coercton operations are imposed. NP
(3 ) a. b. (4) a. b.
*Mary began the highway. (driving on ) *John began the dictionary. (referencing) *Marie a commence l'autoroute. *Jean a commence Ia dictionnaire.
What these data show is that the acceptability of coercion with aspectual predicates such as commencer and begin is conditioned largely by the telicity of the event which is metonymically reconstructed in the complement position. Essentially, these verbs select for an event of the sort TRANSITION, ruling out the coerced complement interpretations of drive on the highway for (3b) and (4b), and consult the dictionary for (3c) and (4c), which are both PROCESS events.J
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
There are several phenomena discussed in the literature which would apparently suggest that type coercion is not a general interpretative strategy for compositional semantics. In this section, we review these apparent counter examples and discuss each briefly to show why they are nonproblematic. We will concentrate, however, on the selectional properties of aspectual verbs such as commencer and begin , in order to show very clearly that sense enumerative approaches, such as that taken by Godard & Jayez ( 1 993) are unable to capture linguistic and computational generalizations, with respect to how the lexicon contributes to the compositional semantics of natural language.
James Pustejovsky and Pierrette Bouillon
1 37
Observe that the sentences in (3) and (4) do have legitimate coerced readings with transitional 'create' interpretations, such as building a highway and compiling a dictionary. Constraints due to 'boundedness' of the predicate (sentences (sa) vs. (sb) and (6a) vs. (6b)) are also consistent with conditions on coercion. Observe that mass noun and bare plural NPs in complement position are not acceptable as possible coercions. (s) a. *John began the cheese (eating) I the book (reading). b. John began cheese (eating) I books (reading). (6) a. Jean a commence le fromage I le livre. b. *Jean a commence du fromage I des livres.
(7) a. *The acid is beginning the marble. (corroding)
b. *John is beginning the noise. (being annoyed by) (8) a. *L'acide commence le marbre. b. *Jean commence le bruit.
In the discussion below, we show that the sentences in ( 7) and (8) are actually raising constructions rather than control structures, and that such constructions do not allow coercion at all. Experiencer predicates. A second argument against coercion might appear to come from the selectional properties of experiencer predicates. Consider the sentences in (9) below. (9) a. Books bore me. b. The movie frightened Mary. c. John's face scared me. d. Listening to Mary bores Alice. While it seems straightforward to admit that verbs like begin select an activity of some sort in all their subcategorization forms, what common selectional
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In these sentences, the homomorphic relation between the NP type in (sb) and (sa) (mass vs. count) and the event type gives rise to process and transition interpretations of event structures, respectively (corresponding roughly to the amorphous and bounded readings within Krifka's 1 992 analysis). That is, the mass NPs in (sa) and (6a) (cheese , books , du fromage , and des livres ), when metonyrnically extended with their qualia, emerge as PROCESS or amorphous event types (i.e. 'eating cheese' and 'reading/writing books'). The count NPs in (sb) and (6b), on the other hand, emerge as TRANSITIONS or bounded events (i.e. 'eating the cheese' and 'reading/writing the book'). Finally, there would appear to be constraints on coercion which suggest that agentivity may play an important role in licensing the operation. For example, as pointed out by Godard & Jayez (1993), sentences such as (7) and (8) are ill formed:
- --------
1 38
- -- �-
-------
Aspecrual Coercion and Logical Polysemy
( 10) D'Vx'Vy'V e [anger(e, y, x )] - 3 e13 e23Exp [Exp (e1, x, y ) 1\ -. angry (e1, x) 1\ angry (e2, x) 1\ ....., e2 < et] This states that a verb such as anger involves someone directly experiencing something, and as a result becomes angry. What is interesting about examples such as (9a) and (9b) is that the semantics of the NP in surface subject position contributes information to the interpretation of what kind of experiencing event is involved. That is, the qualia structure projected by the NP books contributes to the particular manner in which I became bored in (9a), namely the NP's TELIC role of reading. Similarly, our knowledge of movies as something that we watch and experience in a particular manner is encoded in the TELIC role of movie in sentence (9b). The event projected from the noun movie , viz. watch , in tum satisfies the selectional requirements of the verb anger on its subject. Adjectival selection. A third argument against coercive operations involves examples such as a long novel and a bright bulb . If adjectives such as long in this case are analyzed as event predicates, as suggested in Pustejovsky & Anick ( I 988), where long modifies the activity of reading a novel, then there would appear to be a problem with selection in sentences such as (9), where the verb acheter (buy) selects for an individual while long selects for an event. The question is, how can the head of the NP possibly denote both an event and an individual, since such sentences are perfectly natural? ( I I ) Jean a achete un long roman. John bought a long novel.'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
property would relate the different subject types in (9)? The answer is in fact very similar to that for verbs such as begin . We can view these sentences as involving a metonymic reconstruction of the subject to an event and, in particular, to an experiencing event between the surface object and the surface subject. That is, in (9), it is (my reading) books which bores me, (Mary's watching) the movie which frightened her, (my seeing) John's foce which scared me, and (Alice's ) listening to Mary which bores her.4 Thus, it seems that the linguisric evidence supports an underlying semantic type of an event as the subject, which would directly explain what the connection between the subject and object of the experiencing relation is. In Pustejovsky ( I 99 I b), it is suggested that the underlying semantics of psychological predicates such as bore, anger, and frighten is a causative structure where the surface subject is the logical object of an experiencing event. On this view, the lexical representation for the verb anger has something like the following form, where Exp (x, y) is a sortally restricted relation of experiencing (e.g. hearing, seeing, watching, etc.), and < is a strict partial order of temporal precedence.
James Pustejovsky and Pierrette Bouillon
1 39
What these adjectives demonstrate is not a violation or puzzle for coercion and selection; rather, they serve to illustrate the selectional properties of different classes of adjectives, as modifying different facets or qualia of the head. Modification by an adjective such as long, rapide (fast), or brillant {bright), can be seen as event predication, submodifying the appropriate quale of the head.s The adjectives in these cases modify a distinguished event predicate (i.e. the TELIC quale) associated with the head,
long book
read
for
book , and illuminate for bulb . Thus, a bright bulb is a
is interpreted as one taking a long time to read, while a
bulb which shines brightly when illuminated. These adjectives, on the other hand, should be contrasted with modifiers such as expensive in an expensive book and
opaque
in
an opaque bulb ,
both of which refer to the physical object rather
discussion, it should not be surprising that an entire NP can appear in an environment typed for an individual (e.g. as the object of buy), while its head is modified by an event predicate within the NP (e.g. as modified by
that the lexical semantics for the noun
( 1 2)
novel
is that given in ( 10) below:
.Ax (novel(x) 1\ CONST = narrative '(x) 1\ FORM = book '(x) 1\ TELIC = Ay, eT(read '(x) (y)(eT)] 1\ AGENTIVE = Ay , eT(write '(x)(y)(eT)]] a.
The analysis of adjectives such as
long
and
bright
overall type of the NP, as illustrated in ( I I ) below:
( I 2)
long). Assume
b.
does not change or shift the
.Ax (novel(x) . . . I\ [Telic = Aw , eT(read (x)(w)(eT) 1\ long(eT)]] . . .]
The reading in ( I I ) specifies that the event of reading is modified by the event predicate
long .
The resulting compositional structure is still the type of the
whole NP, and has no effect whatsoever on selection by an outside governor such as acheter (buy) as in ( I I ). That is, verbal selection and event modification are operating within different predicative domains. While the matrix verb
acheter
NP, an long can submodify the TELIC event associated with the object (namely, reading) while not affecting the overall type of the NP. selects for an individual such as a physical object, within the
event predicate such as
In what follows, we demonstrate how the apparent violations of the coercive behavior of aspectual predicates actually reveal a much deeper semantic distinction between two logically related senses of the verb in all the complement forms they select, and not just NP complement cases, which can be applied mutatis mutandis to commencer. We will show that this distinction is due to the event structure of the complements selected by the aspectual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
than an activity or state associated with the object. These adjectives can be seen as modifying the FORMAL role of the qualia for these nouns. Given this
1 40
Aspectual Coercion and Logical Polysemy
predicates. We also demonstrate why commencer andfinir(finish) allow coercion while cesser (cease) and a"eter (stop) do not.
4 THE RELEVANCE O F EVENT S T R U C T U RE
An event tree structure can be defined in terms of (at least) one of three relations: ( I ) that of'exhaustive ordered part of', , giving e [Honda :s:; vehicle] : Honda .... vehicle .20 Similarly, in (36b), Tractatus :s:; book :s:; text defines a relation between the type selected by the verb read and the actual individual. This is an instance of the more general subtyping coercion operation, illustrated below:
1 so
Aspectual Coercion and Logical Polysemy
marble , for example, in (26a), is (e, eP), then, FC(begin , VP) = A.9' [begin (corrode(9' , the-marble)]. Following Pustejovsky (1994b) we can view the basic composition of the sentence in (26a) as type inference, where E9 represents type application as inference according to the typing judgements, G , in the grammar. The type inference tree for this construction is shown in (39). FC (commencer : c 0 -
,J: wrrodrr lr marbrr: e - £'1)
L'acide : e E9 commencrra corrodrr le marbrt : r- ,T (39) L 'acide commencer a corroder le marbre : eT
(4o)
a : a $ Q[a, r] : a - r Q[a, r](a ) : r
This says that, given an expression a of type a, there is a coercion possible between a and r, which changes the type of a in this composition, from a to r. We will illustrate the further application of this coercion operation below, as used in the commencer examples. In (4 1 a) we see how the aspectual verb commencer selects the complement VP, and how in (4 rb) an NP is coerced into an event interpretation. Both sentences involve left-headed event structures, resulting in a control interpretation for the verb commencer. (4 1) a. Marie a commence a lire le livre. (VP) Mary began to read the book. b. Marie a commence le livre. (NP) Mary began the book. The type inference tree for (4 r a) is given below: commencer : (e - cT) - (e - cT) G)
;, lire /e livre :
e- ,T
commrnma lirt le livre : e - cT (42) Marie : e $ Marie commencer a lire le livre : fT
For the derivation of(41 b), reference is made to the qualia structure of the noun book , as shown below, where ARGSTR refers to the argument structure of the nominal book, treated as a type of implicit relational noun (c£ Nunberg 1979 and Pustejovsky I 994b):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In Pustejovsky & Boguraev (1993) a general mechanism is defined which makes the appropriate type available for a coercion operation. The qualia can be seen as partial functions, returning the value of a particular quale for an NP. The combined set of qualia provide a set of type aliases for the expression containing them. One particular mechanism, type pumping, has been explored as a means to generating the alias set, and we will make use of this device below. Thus for example, the type available to an expression a with quale Q of type r, can be seen as the following type inference:
James Pustejovsky and Pierrene Bouillon
book ARGSTR =
(43) QUALIA =
[
J
1s1
information . y : phys obJ infonnation-phys obj-container-lcp FORM = hofd(y , x) TELIC = read (eT, W , X ) AGENT = write(eT, V, y) ARGI = x :
ARG2 =
_
[
_
]
c A.eT[commencer'(A.x , e [lire(le livre)(x)(e)])(Marie)(eT)) => Marie{A.x, eT[commencer'(A.x, e [lire(le livre)(x)(e)](x •))(x•)(eT))} => Marie{A.x, eT[commencer'(A.e [lire(le livre)(x •)(e )])(x•)(eT))} => 3eT[commencer'(3e [lire (le livre)(Marie)(e )])(Marie)(el)]
The syntactic structure associated with this sentence can be illustrated in (46):
(46) Control (left-headed transition):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This representation illustrates how the qualia make reference to two events associated with the noun, reading and writing. As illustrated below in (44), coercion applies to the complement NP, where reconstruction with either the TELIC or AGENTIVE qualia will result in the appropriate type selected by the verb. We illustrate the derivation with the TELIC role selected.21
152
Aspecrual Coercion and Logical Polysemy
[NP ------[J �] [VP s
VP
lex Marie TYPE ind
� �
�
AUX
lex a TYPE aux
.
v
ex commence A.x k 3P 1
�
I
A.x A£, 3 e2 F
[
F
=
A*
=
]
L�NP J . J] L J
P(x)(e2) R (e2)(x)(e ) 1
=
l
P ' (le-livre)(x)(e2)
�����;�
Let us now return to the discussion of raising constructions and coercion in such structures. We observed in sentence (6a) that type coercion is unacceptable with this predicate: (6) a. *L'acide commence le mabre. (corroding) *The acid is beginning the marble. There would appear to be a possible derivation involving type coercion in this sentence where we choose the raising sense of commencer, imposing the type I!' on the complement. But notice that coercion will be successful only if the appropriate type exists in the alias set of the complement. Metonymic reconstruction on the complement in (6a) returns an eventual function of type (e, e'f) rather than the type selected by the verb, 1!'. Since function composition is an operation at the level of the VP, there is no point in the derivation at which the appropriate type is available for the rule to apply, and the sentence is not semantically well-formed. As we saw above, this is not the case with control verbs. Having outlined the basic mechanism of coercion under constraints, we can explain now why examples like (47) are ungrammatical. (47) a. *Marie a commence l'autoroute. (driving on ) *Mary began the highway. b. *Jean a commence le dictionnaire. (consulting) *John began the dictionary. c. *Jean a commence le sommet. (reaching) "'John began the top of the mountain. d. *Jean a commence la symphonie. (listening to ) "'John began the symphony.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
l
lex commence le livre
James Pustejovsky and Pierrene Bouillon 1 5 3
e. *Jean commence le livre. (destroying) *John began the book £ *Jean commence le desen de Gobi. (going through ) *John began the Gobi desert.
sonate ARGSTR -
QUALIA =
[
[
- e : event . ARG2 - x : abstract_obJ performance-lcp ARGI
J
music (x ) TELIC = perform (e, w , x ) , listen (e', z , e) AGENT = compose (e", y, x ) FORM -
]
The first thing to notice is that the lexical item directly denotes an event, as well as an individual sortally restricted to music . As pointed out by psychologists such as Miller (I 99 I ), social artefacts are very different from simple physical anefacts, in that their function is defined in a more complex manner. For example, in defining the TELIC role for an event object such as symphony, one cannot ignore the role of the listener (the experiencer). That is, music is performed for an audience. This must be reflected in the qualia structure as a conjunction of relational values, i.e. perform and listen . Recovering the event listen in the metonymic reconstruction due to coercion without also recovering perform is similar to binding a variable with a partial value; that is, listen is a
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The coercion in (47a) and (47b) is impossible for reasons already discussed; namely, if these sentences are examples of raising constructions, then they are ruled out according to our previous discussion. If they are control senses, then the metonymic reconstruction on the NP in each case produces a type (PROCEss) that is incoherent with that selected by the verb, namely an eventual function with a TRANSITION. In (47e), the coercion is impossible as the qualia does not stipulate how you destroy an object, but rather its bringing about or what you do with it. Similar remarks hold for (47c) and (47f). One might argue that (47d) should be possible with a control interpretation (the event in the complement is controllable and bounded), and yet coercion is not possible. This example points to a somewhat different phenomenon, one involving the semantics of the nominal itsel£ The qualia representation of symphonie appears to make reference to both an event and an individual reading. It suggests that the semantic distinction between objects such as books and tapes on the one hand, and symphonies and sonatas on the other is responsible for the unavailability of such coerced readings. The qualia structure for event-objects (using Dowty's 1979 terminology) such as sonate and symphonie can be schematically given as follows:
1 54
Aspectual Coercion and Logical Polysemy
dependent event while perform is independent, being projectable through coercion by itsel£ We can see how, with a raising sense, as in the sentence La symphonie a commence, nouns such as symphonie and sonate are directly selected by the verb 2 since they are able to denote events; hence, no coercion is involved (c£ (49)): 3
(49)
'Raising' (right-headed transition): s
[:
]
Ia symphonie TYPE event
(I s) a. b. ( I 6) a. b. ( I 7) a. b. c. I 8. a. b. c.
He finished reading the book. (VP[+PRG]) He finished the book. (NP) II a fini de lire son livre. (VP[+INF]) II a fini le livre. (NP) *I finished the symphony. (listening) I finished the chocolate. (eating) * I finished chocolate. (eating) *]'a fini la symphonie. J'ai fini le chocolat. *J'ai fini du chocolat.
The ungrammaticality of (I 7a) and ( I Sa, c) will follow from the same analysis given for begin above. Notice, however, that finish differs from commencer and begin in that it is not logically polysemous, having only a control sense. That is, the raising examples in (so) are ungrammatical. (so) a. *It has finished raining. b. *The sun has finished shining in my eyes. c. *The acid finished corroding the marble. This would suggest thatfinish is not lexically underspecified with respect to headedness, as is begin , but is already specified with a head. It is this lexical specification which gives rise to the control reading only. Interpreting Dowry's (I 979) analysis o(finish within our framework, we can analyze this verb as making reference to two events: ( I ) that subevent which brings about,
sine qua non , the culmination of the event as a whole; and (2) an
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us now turn briefly to the semantics of the verbs finish and finir. These verb patterns are the same as commencer and begin with respect to coercion (c£ ( 1 s )-(I 8) repeated below),
------
--- -
- ---
James Pustejovsky and Pierrette Bouillon
--
1ss
assertion that the entire event has occurred. As Dowty makes clear, this presupposes that the event has a natural division into two subparts. This would seem to indicate that finish is an aspectualizer which type-shifts the complement event into an achievement (i.e. a right-headed transition). It does this, however, by preserving the integrity of the complement event, for notice how finish -sentences behave in many respects as both accomplishments and achievements: (51) a. Mary finished building the house in 3 months. b. Mary finished building the house at 3:oo pm today.
(52) V'e1'1e2 [cul(e1) = e2
++
-.3e [e2 :S e1 1\ e2 < e 1\ e :5 e1]]
we can build this relation directly into the event structure itself, in which case it would be a relation on event trees. This should essentially be a logical culmination relation between events; cula:(e1 , e2). Now we can express the semantics offinish as a right-headed transition, where the subevent standing in the culminating relation with the larger event is seen as the AGENTIVE of the overall aspectual event. Furthermore, the FORMAL or result of the aspectual event is the assertability of the entire transition, of which the AGENTIVE is a part. (53) h-A.ef:le13R3P (finish: ES = e1 cufa: e2 1\ FORMAL = (P(x)(e?)]] 1\ AGENTIVE = [R (e1, x)]] Consider now the sentences in (54) and (5 5), which appear to be raising constructions and are grammatical in French and marginally acceptable in English.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Hence, even lexical accomplishments (left-headed transitions) can be interpreted as achievements (right-headed transitions) when complements of finish . In order to capture this intuition while still satisfying Dowty's fundamental interpretation offinish , we define a general relation of logical culmination, cui, between an event and one of its subevenr:
( 54) a. ?The leaves have finished falling. b. ?The paint has finished drying.
( 5 5 ) a. Les feuilles finissent de tomber. b.
Le
peinture a fini de secher.
These data would suggest that raising construction is possible with finish and finir with some nominals. But in fact, the sentences in ( 54) and ( 5 5) are best analyzed as pseudo-control cases, and they are restricted to a certain well defmed class of norninals. In general, these verbs do not pass the standard rais ing tests, bur nouns such as paint and leaves are exceptions because they carry qualia information indicating a kind of 'autonomy of behavior' relative to
-,
1 56
Aspecrual Coercion and Logical Polysemy
certain predicates. Hence, paint , for example, is consmted as a pseudo-agent in the control relation because of this property; i.e. it can dry on its own. It is interesting to observe that even in English an intransitive consmtction is possible; namely, if the event nominal in subject position has an agentive component (c£ (s6b) and (s6c)), then a control interpretation is possible in what would otherwise appear to be an intransitive (i.e. raising) consmtction. We will refer to these as intransitive control consmtctions. (56) a. b. c. (57) a.
While classes and talks have an apparent agentivity and controllable component to them, parties are less controllable, resulting in the less acceptable (s6a). Since rain is completely uncontrollable, it is ungrammatical in an intransitive control construction with the verb finish (c£ (57a)). The verb stop , as we see below, allows a raising interpretation and permits the intransitive raising consmtction in (s7b). The semantics of arreter and stop are interesting because they have both control and raising senses, yet do not allow complement coercion at all (c£ ( I 9) and (2o) repeated below).
( I 9)
Madame arrete de verser le the. ?Madame is stopping steeping the tea. b. *Madame arrete le the. *Madame is stopping the tea. (2o) a. ]'arrete de lire le livre. ?I am stopping reading the book. b. *]'arrete le livre. *I am stopping the book. a.
Why then should coercion not be possible with what would appear to be a verbal form almost identical to that of commencer and begin ? The answer to this question emerges from a closer examination of the data. Observe that stop appears in the sentences in (s8) with a non-control consmtction, assuming the sense of 'prevent': (58) a. John stopped Mary from smoking in his house. b. Mary stopped the man from hitting her. c. John stopped the bomb from exploding. In fact, there is a kind of coercion possible in complement position with stop , essentially reconsmtcting an ellipsed predicate, as in (59).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
??The party finishes at midnight. Class will finish at 2:oo pm. The talk will finish by noon. *The rain will finish by noon. b. The rain will stop by noon.
James Pustejovsky and Pierrette Bouillon
1s7
(59) a. John stopped the car. (from moving) b. The referee stopped the clock. (from moving) c. Mary stopped the record. (from playing/moving) While constructions such as (59) are possible in French, those in (5 8) are not. But there are data suggesting that arreter does allow non-control complementa tion, as in (6o), with the sense of empecher (prevent). (6o) a. ]'arrete la bombe avant qu'elle explose. I am stopping the bomb before it explodes. b. ]'arrete le moteur avant qu'il ne chauffe. I am stopping the car before it overheats.
7
CONCLUSION
In this paper, we have attempted to illustrate how coercion operations are constrained by typing judgements and the structure of lexical semantic representations. Constraints on generative operations such as coercion are an integral component of the approach to semantics we have presented. In the process of this discussion, we have reiterated the advantages of a generative lexicon in the context of the larger theoretical and methodological issues in lexical semantics. More specifically, we have shown how begin and commencer exhibit both raising and control behaviour, and that this is an instance of the larger alternation class between causative and inchoative verbs, itself an example of logical polysemy. We have further shown why coercion is possible only with the control sense of commencer and begin and illustrated both the type inference involved and the semantic derivation of these constructions. We have also examined the behavior of two other classes of aspectual verbs, arreter and stop and finir and finish , and have shown whyfinir and finish are unambiguous but do exhibit complement coercion, and do appear to allow raising construc tions. Finally, we explained why arreter and stop do not allow coercion, even though they have control readings. We believe that the advantages accom panying generative mechanisms and the characterization of languages as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What these data suggest is that the complement type of arreter and stop is not an eventual function, as with commencer and begin , but rather simply an event, where the type ofverb is (�. (e, ET)). That is, these verbs are not strict obligatory control verbs, such as try and begin , but impose 'available controller' binding, as with verbs such as want (c£ Chomsky 1 98 1 ; Dowry 1 98 5; and Farkas 1 988). Given that the complement of both arreter and stop is �. it is clear that coercion is not possible since this is not among the type aliases for the NP complements given in (r 9b) and (2ob).
I
s 8 Aspectual Coercion and Logical Polysemy
polymorphic in well-defined ways allow us to overcome the explanatory inadequacies inherent in word sense enumeration approaches to lexical semantics. Although some of the details of the analysis have been omitted in order to concentrate on the general strategy of lexical analysis employed here, we have hopefully made clearer what some of the specific theoretical advantages of this approach are. Acknowledgements We would like to thank Peter Bosch, Bob Ingria, Noam Chomsky, Federica Busa, Michael Johnston, Evelyne Viegas, Graham Russell, and two anonymous reviewers for their helpful comments and suggestions. All responsibility for errors are ofcourse our own. Received: I 8.07.94 Revised version received: 05.I2.94
Computer Science Department Brandeis University Waltham, MA 02254 USA
[email protected] PIERRETTE BOUILLON
ISSCO University ofGeneva 54 route des Acacias Ch-t zz7 Geneva Switzerland
[email protected] NOTES
=-
I See, for example, the work reported in Buicelaar & Mineur (I 994), which attempts to unifY some of the notions from Generative Lexicon Theory wii:h Categorial Grammar. Busa & Dini ( I994) attempt to import the notions of coercion with qualia srructure (see below) into HPSG for the handling of control phenomena, while Bouillon & Viegas (I994) handle cross-linguistic pheno mena of adjective-noun collocations. Both Copestake (I993) and Sanfilippo (I 990) are also interesting in how they model the projection of lexical semantic information to the syntax.
2 Qualia structure can be seen as providing the 'modes of explanation' for a concept, as lexicalized in a particular word. This is not strictly rrue, as we shall see in Section 5 below, where we introduce the notion of an eventual function. 4 As one reviewer points out, we experience objects in any number of ways. That is, one need not read books in order to be bored by them. One can be bored by looking at them, shopping for them, writing them, or chinking about them. This is not in any way inconsistent with the GL approach. As discussed in Puste jovsky (I99Ia) and elaborated on in
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
JAMES PUSTEJOVSKY
James Pustejovsky and Pierrette Bouillon 1 5 9 Pustejovsky ( 1 995). the qualia determine two types ofinformation in the context of coercion: i. type and sort information which the qualia must satisfy; ii. specific qualia values which are the explanatory modes in understanding a word.
1I
I2
I3
14
1s
16
17 18
---
--
-- -- ---
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
For words such asfilm and book, the TELIC quale role value of watch and read respectively are not optional in any sense, but are part of the semantics of the words. When an NP enters into a coercive environment, such as here with experien cer verbs, the qualia values act only to determine the default assignment for how the type environment is recon srructed. Strictly speaking, in terms of type saris faction either TELIC or AGENTIVE is a possible target of the modification. Hence, it can also be interpreted as e.g. 'a book taking a long rime to write'. 6 We follow Landman (199 1 ) in making the distinction between the temporal relations in an event structure and the intensional relation between event parts. 7 Formally, the head is interpreted as a focus srructure over the domain ofevents. This approach is explored in Pustejovsky (1995)· 8 We will continue referring to the binding relation between the matrix subject and an argument position in the complement position as semantic control, although the analysis does not necessarily assume a syntactic reflex for this binding relatiotL 9 There are at least two other major properties of raising predicates that should be mentioned, but that are not as important to our discussion. These are: (a) the inheritance property, which ensures that any syntactic restrictions imposed by the embedded VP on the subject are inherited or reflected in the 'raised' position; and (b) the narrow scope interpretation of the raised NP relative to the raising verb. See Jacobson (1990) and
10
Di Sciullo & Williams (1987) for discus sion. Further discussion of the syntactic patterning associated with aspectual predicates in English can be found in Freed ( 1979) and Rudanko ( I 989). Other verbs in this natural class include grope for, reach for, and other conative verbs as well. On this view, Tense is analyzed as a function from events to propositions, viz of rype (ff', t). We simplify this analysis here for ease of presentation. In fact, an eventual function is a function from individuals to sets of events: ((o, r)r)D. With this typing, begin as a conrrol verb has the following type: ((e, (er, t)), ((e, (eT))). Furthermore, under this analysis, Tense is rreated as a generalized quanti fier: ((£0, t), t). For details see Pustejovsky (1995). We will also refer to the typing assign ments in (3 I ) as (e - eT) - (e - eT) and (ro - eT), respectively, for use in type inference trees below. We follow Asher & Morreau ( 199 1 ) and Asher & Lascarides ( 1 993) in the use of the defeasible conditional > for specifying default lexical inferences. Unless otherwise stated, the default event srructure (Es) associated with the qualia for a TRANSITION is: ES - [,T eP , where A > B means 'if A then normally B' and represents a defeasible or generic rule. The semantics of this language is given in Asher & Morreau (I 991) and allows us to encode defaults about lexical entries. We will shortly be reasoning about the types and properties of eventualities that verbs and verb complexes introduce within our semantic representations. As we intend ultimately to talk of discourse structure using the framework of Asher et a / . ( I 993), our semantic representations for clauses will be DRSs and hence the representational elements that verbs will introduce will be predicative DRSs that abstract over discourse referents. The types of eventualities that we shall discuss are in fact types of discourse referents. Predicative DRSs are functions from tuples of discourse referents of the appropriate type to conditions in a DRS. Our axioms will then apply these predicative DRSs to a sequence of discourse referents of particular types that occur as arguments and spell out the consequences in terms of spatiotemporal conditions on those discourse referents. We will also generalize on the types of conditions; thus, we will have axioms that apply conditions of a certain type derived from those verbs and prepositions. These axioms should be understood as rules for expanding a representational structure-either a DRS or a structure that in the DRS construction procedure we favor will eventually yield a DRS when combined with other elements. This is similar to the approach taken by Kamp & RoGdeutscher in a number of papers (see e.g. Kamp & RoBdeutscher I 992). We will assume the existence of certain functional dependencies between eventualities and other discourse entities, e.g. Source(e), SIP(e), Goal(e), Mobile(e)-the moving entity-and Lre�e)-the location of reference. 1 1 We will forgo the box notation of DRT, since for us nothing hangs on it at the level of the spatiotemporal information we address here, though we shall construct DRSs when we apply our theory. 1 2 We give below the properties for each group of verb.
Nicholas Asher and P�erre Sablayrolles
1 73
Definition 5 Verbs ofchange oflocation 1 . A verb of change oflocation, entails that the moving entity changes location during the process ex: entrer to enter, arriver to arrive, atterrir to land, s'approcher to approach . . . 2. change-oflocation (e) -+ SIP(e) # {(Source(e)))
One may also wish to require that for changes of position, the moving entity at least by default stays in the same location during the whole process, but we will not do this here. So it is possible that a change of position is also a change of location. Definition 7 Verbs ofinertial change of position 1 . By defoult a verb of inertial change of position implies a change ofposition for the moving entity ex: courir to run, danser to dance, voler tofly . . . 2. inertial-change-ofposition (e) > Path-position(e) # {(!nit-position (e))} Definition 8 Verbs ofchange of posture 1 . Verbs of change of posture entails that the moving entity stays inside the same location and at the same position during the whole process, but also changes of posture during the process ex: se pencher to lean over/towards(forward!back, s'asseoir to sit down, se baisser to bend down . . . 2. change-ofposture(e) -+ Path-posture (e) # {(Init-posture (e))) change-of-posture (e) - Path-position (e) = {(!nit-position (e))} Note that a change of posture is not any change of posture, but just one that is expressed by a certain verb. Changes of posture in our sense, those that are expressed by verbs in French (and we suppose for other languages too), do not entail changes of position-and, hence, changes oflocation. 3 ·4
Embedding classes
For each subclass of motion verb defined in Section J.J, we associate truth conditional properties expressed in our formal language. These properties
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Definition 6 Verbs of change of position 1 . A verb ofchange of position entails that the movement entity changes position during the process ex: se deplacer to move around, circular to circulate, descendre to go down, s'elever to go up . . . 2. change-ofposition (e) - Path-position (e) # {(!nit-position (e))}
1 74 A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
.
1 S· Courir sur place To run on the spot 16. Courir autour du stade To run around the stadium 17. Courir vers Ia maison To run towards the house
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
reflect the minimal intrinsic core meaning common to all the verbs belonging to the class. The three first subclasses are not exclusive; when we take each class to be defined by the eventualities they describe, they are in fact strictly ordered by the subset relation. That is, the spatial properties of the minimal core meaning of verbs of change of location are more constrained than those of verbs of change of position, which, in their turn, are more constrained than those of verbs of inertial change of position. Our typology is based on intuition and our definitions of location and position: one cannot change location if one does not change position during the process of changing of location; and one cannot change position if one does not inertially change position during the process. Our taxonomy allows for a monotonic calculation of constraints for a verb complex. The combination of a verb and its PP or other arguments (direct object, for example) can only add to the spatiotemporal properties that affect the Source, Goal and Path of the targets' posture, position, or location. There is an evident advantage in doing this. Non-monotonic constraint propagation would require us to assume the existence of something like an intensional operator that might block the truth conditional content of the verb itself-in the sense that Believe (¢ ) fails to entail ¢ -or to rephrase all our constraints at the default level. The latter makes lexical semantics very complex (for a discussion see Lascarides & Asher 1 993), and we do not see our way through. The former seems semantically unmotivated. With monotonic constraint propagation, we need resort to neither alternative.U Take, as an illustration, the following verbs: courir to run , voyager to travel , and entrer to enter, which are verbs of inertial change of position, of change of posi tion, and ofchange oflocation, respectively. The composition of to run with some other elements can either result in a complex having the same constraints (such as in ( 1 s)), or more constraints (such as in ( 16)), or even more constraints (such as in (17)). In the case of the verb to travel, the compositionality can either result in a complex having the same constraints on the movements of the moving entity (such as in (r 8}}, or more constraints (such as in (19)), but never fewer constraints (as (2o) shows). In the case of the verb to enter , which already belongs to the class placing the most constraints on the movement of the moving entity (ofthe three classes considered here), only a complex having the same constraints can result from compositional calculation (c£ for example (2 1 ))
Nicholas Asher and Pierre Sablayrolles
175
8. Voyager en Europe To travel inside Europe 19. Voyager de Toulouse a Austin, Texas To travelfrom Toulouse to Austin, Texas 20. *Voyager sur place *To travel in place 2 1 . Entrer dans une maison To enter a house 22. S'asseoir sur une chaise To sit on a chair 23. S'asseoir par terre To sit down 1
4 F R O M LEX I C A L S E M A N T I C S TO D I S C O U RSE U NDERSTANDI N G 4.1
The verbs ofchange oflocation
In this section, we focus on verbs of change of location and ignore three other groups we have defined in Section 3·3· Verbs of change of location are propitious for a detailed analysis of how the space is organized in and around the location with respect to which the displacement takes sense. They describe displacement going from the outside of a location to the inside, or the reverse displacement. Nevertheless, a coarse organization of the space into only two 'zones', the interior and the exterior of the location, quickly proves insufficient. Compare, for example, (24) with (25) and (26) with (27). 24. Abby est sortie de la maison Abby has gone out ofthe house 25. Abby est partie de la maison Abby hasgone awayfrom the house 26. L'avion a atterri sur la piste 4 The plane has touched down on runway 4 27. L'avoin s'est approche de la piste 4 The plane has approached runway 4
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The composition ofverbs ofchange of posture with any PP (cf (22) and (23)) will always result in a complex having the same constraints as those the verb introduces. If one does not change position or posture during the process described by a verb or verbal complex, then this 'process' can only be a static one.
1 76
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
In both
(24)
and
(25), Abby has
gone from the inside of
the house
to the
outside ofit. But if sortir in (24) only describes this sort of displacement, partir in
(25) forces Abby to continue her displacement till she is away from the house, at a sufficient distance. Under a certain critical distance, the verb partir could not be used. For example, ifAbby took two minutes to talk in the garden with her neighbour Peggy, and if someone asked AI, who is inside the house 'Abby est
elle partie?' 'Has Abby gone away?', AI will surely answer no, if he can hear her
talking or see her through a window. This critical distance is a function of many
parameters-not only the size of the moving entity and of the location, but also the presence and the size of the other referents (locations and/or objects) in the
difficult to circumscribe precisely this critical distance. But we do believe it is pan and parcel of our natural language metaphysics and classification of movements. We believe that the contextual factors relevant to describing this critical distance must be handled non-monotonically. But we shall do no more here than assume that there is some contextually given and pragmatically constrained Ininimal distance. With
(26)
and
(27),
the plane stays during its whole motion outside the
location which here is runway 4· Nevertheless, the displacement is not the same in the two sentences. In (26), the plane touches down and thus finally comes in contact with the runway, whereas in touching it.
(27)
it comes near the runway without
In order to be able to take into account and to represent formally these differences, we propose seven relations (cf Section
4.2).
Four of these relations
encoding how the space is structured by verbs of change oflocation are sirnilar to the ones presented byJackendoff& Landau ( 1 992) which encode the space as it is structured by spatial prepositions. For example, the preposition describes the interior of the reference object;
against
inside
describes an external
contact with the reference object; near describes an exterior close by with respect to the reference object; finally for describes a far away exterior with respect to the reference object. As we have come to these distinctions by examining motion verbs, we conclude that language structures space in the same way whatever son of lexical items (motion verbs or spatial prepositions) we examine. We talk about (spatiaVspatiotemporal) relations and not about (spatiaVspatiotemporal) zones1• below for the following reason. Consider, for example,15 our plane and our runway in the sentences (26) and (27). If we want to define a zone of contact between the plane and the runway, we have to construct this zone as the union of all the spatial positions the plane can occupy so long as it is, in one way or another, in contact with the runway. This will result in a zone whose height is equal to the longest dimension of the plane (and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
scene, the perceptual (visual, auditory . . .) capabilities in the case of human beings, and other elements of the context. The critical distance will be different ifwe know that Abby intends to go away by foot or by car, for example. It is very
Nicholas Asher and Pierre Sablayrolles
1 77
so a tone determined by imagining the plane in a vertical position and its nose in contact with the runway). It is then easy to place in such a zone the plane in its canonical position at, let's say, one metre up from the ground. So, the plane is inside the zone of contact but definitely not in a relation of contact with the runway. In conclusion, it would be false to say that the plane is in contact with the runway just because it belongs to the zone of contact. Belonging to the zone of contact is a necessary but not sufficient condition to be in contact. 4.2
Seven spatial relations Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We introduce our relations discursively (Definitions 9. 1-1 5.1) and offer a formal definition (Definitions 9.2- 1 5.2). In the logical formulation, we make use of the mereology of Lesniewski ( 1 92 7-3 1 ), formalized and extended with topological structure by Clarke (198 1 , 1 98 5). 1 6 In Arnague & Vieu ( 1 993), the reader can find an extension of Clarke's topological mereology and an application to the formal representation of the semantics of space and space time in natural language. Definition 9 x is in a relation ofinner-halo with y iff 1 . x is strictly in thefunctionafl1 inside ofy, i.e. without sharing any ofits points with the fro ntier ofy 2. inner-halo {x, y) =def PP(x,fint(y, x}} Definition I O x is in a relation ofcontact with y: 1 . x is in thefunctional outside ofy and is weakly connected toy, i.e. x andy are in contact but share no points (Aurnague and Vieu talk about a 'natural contact' in Aurnague & Vieu (1993) 2. contact (x, y) =tkf PP(x,fext(y, x)) A weak- contact(x, y) Definition I I x is in a relation of outer-halo with y: 1 . x is in thefunctional outside ofy but at a distance lesser than the critical distance, i.e. x is 'in proximity' ofy 2. outer-halo {x, y, C) =def PP(x, prox{y, x, C)) Definition 1 2 x is in a relation ofouter-most with y: 1 . x is in theJunctional outside ofy and at a distancegreater than the critical distance, i.e. x is not 'in proximity' ofy 2. outer-most (x, y, C) =d�J PP(x,fext(y, x}} A -. PP(x, prox{y, x, C)) Definition I3 x is in a relation ofinner-transit with y: 1. x shares at least one ofits points with thefrontier ofy 2. inner-transit(x, y) =dif EC(x, y) --
--1
178
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
Definition 1 4 x is in a relation ofcontact-transit with y: 1 . x shares at least one of its points with the frontier delimiting the foct to be in contact with y and thefoct to be in an outer-halo relation with y 2. contact-transit (x, y) =J4 contact (x, y) 1\ outer-halo(x, y) Definition 1 5 x is in a relation ofouter-transit with y: 1 . x shares at least one ofitspoints with thefrontier delimiting thefoct to be in a relation of outer-halo with y and thefoct to be in a relation ofouter-most with y 2. outer-transit(x, y) =dtf outer-halo(x, y) 1\ outer-most(x, y)
With our relations defined in Section 4.2, we have at our disposal seven generic locations linked to a given referent (see Figure I ). Talk about locations, however, suggests zones and not relations. We have shown why such zones cannot be defined simply relative to the reference location but rather must be defined relative to the referent, the target, and some relation of spatial inclusion.18 These zones must obey certain constraints given by the relations we have introduced. But we can still define the relevant zones by the following axiom scheme I . Let R be any one of our seven generic relations, and Z - R the corresponding zone. Axiom
1
R (x , y) - P (x , Z - R (x, y )) 'proximity limits ' - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - · ' . ' .
Z-contact-transit Z-contact
Reference Location Z-outer-most
Z-inner-halo
Z-inner-transit Z-outer-halo •-
- - -----�
�---- - - - - J .
Z-outer-transit
Figure 1 The seven generic locations
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4·3 A subclassificationfor verbs ofchange oflocation
Nicholas Asher and Pierre Sablayrolles
1 79
Note the use of an implication and not an equivalence relation in this axiom scheme. Now for the definition of the relevant R-zone, we appeal to a notion of physical possibility that is quite restricted; we are interested in the possible positions ofx with respect to y without any deformation or change in the shapes or volumes of x and y. We also suppose that the reference location y then does change. Axiom 2
Z - R (x , y) � U {z : (z
�
STref(x ) 1\ R (x , y ))}
1. Name of the group: a'approcher to approach Logical definition: Approach(e) -+ {P(Source(e),Z-outer-most(cible(e),Lref(e))) 1\ P(SIP(e),Z-outer-transit(cible(e),Lref(e))) 1\ P(Goal(e),Z-outer-halo(cible(e),Lref(e)))} Lexical entries: 6 Examples: s' avancer to moveforward ; accou rir to rush up Explanation: Such verbs describe a motion going from a far away outside to a near outside of a location of reference. 2. Name of the group: Arriver to arrive Logical definition: Arrive(e) -+ [P(Source(e).z-outer-mos«cible(e),Lreqe))) 1\ P(SIP(e).z-outer-halo(cible(e),Lreqe))) 1\ P(Goal(e),Z-inner-halo(cible(e),Lref(e)))} Lexical entries: 2 3 Examples: aller to go ; venir to come Explanation: Such verbs describe a motion going from a far away outside to the inside of a location of reference, via a near outside of this location.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Our axiom scheme and definition of zones yields the picture of their interrelation as depicted in Figure I. In this section we will concentrate on the verbs of change oflocation. The zones we have so constructed can be considered as locations; they have functionality and they can, even if rarely for some, be lexicalized in different ways. We are then able to subclassify these verbs on the basis of which locations the moving entity is inside, at the beginning, the middle, and the end of its motion, respectively. Not all the possibilities are lexicalized in French. We have listed 2 1 6 lexical entries for French intransitive verbs of change of location, which can be dispatched in ten groups. Nine of them correspond to a change of location in the pure sense of location. The tenth one is a little more special and we will return to it at the end of this section.
1 80
3·
4·
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
s. Name of the group: S'eloigner to distance oneselffrom Logical definition: Distance-from(e) -+ {P(Source(e),.Z-outer-halo(cible(e),Lref{e))) I\ P(SIP(e),Z-outer-transi«cible(e),Lref(e))) I\ P(Goal(e),Z-outer-mos«cible(e),Lref(e)))} Lexical entries: 8 Examples: s' isoler to become isolated ; se reculer to move back Explanation: Such verbs describe a motion going from a near outside to a far away outside of a location of reference. 6. Name of the group: Partir to leave Logical definition: Leave(e) -+ {P(Source(e),.Z-inner-halo(cible(e),Lref(e))) I\ P(SIP(e),.Z-outer-halo(cible(e),Lref{e))) I\ P(Goal(e),.Z-outer-mos«cible(e),Lref{e)))J Lexical entries: 44 Examples: s'en aller to go away; deserter to desert Explanation: Such verbs describe a motion going from the inside to a far away outside of a location of reference, via a near outside of this location. ·
7· Name of the group: Sortir to go out Logical definition: Go-ou«e) -+ {P(Source(e),.Z-inner-halo(cible(e),Lref{e))) I\ P(SIP(e),.Z-inner-transi«cible(e),Lref{e))) I\ P(Goal(e),.Z-outer-halo(cible(e),Lref{e)))J
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Name of the group: Entrer to enter Logical definition: Entrer(e) {P(Source(e),.Z-outer-halo(cible(e),Lref(e))) I\ P(SIP(e),.Z-inner-transi«cible(e),Lref{e))) I\ P(Goal(e),.Z-inner-halo(cible(e),Lref{e)))J Lexical entries: 4 3 Examples: s'embarquer to board; pemetrer to penetrate Explanation: Such verbs describe a motion of going from a near outside to the inside of a location of reference, crossing its 'frontier'. Name of the group: Se poser to alight ; to land Logical definition: Land(e) -+ {P(Source(e),Z-outer-halo(cible(e),Lref{e))) I\ P(SIP(e),Z-contact-transi«cible(e),Lref(e))) I\ P(Goal(e),Z-contac«cible(e),Lref(e)))} Lexical entries: 44 Examples: se jucher to perch ;1 se suspendre to hang Explanation: Such verbs describe a motion going from a near outside of a location of reference to an external contact with this location of reference.
Nicholas Asher and Pierre Sablayrolles
I8I
Lexical entries: I 6 Examples: debarquer to land ; jaillir to spring (up) Explanation: Such verbs describe a motion going from the inside to a near outside of a location of reference, crossing its 'frontier'. 8.
9· Name of the group: Passer (par) to go through ; to cross Logical definition: Cross(e) -+ {P(Source(e),Z-outer-halo(cible(e),Lref(e))) 1\ P(SIP(e),Z-inner-halo(cible(e),Lref(e))) 1\ P(Goal(e),Z-outer-halo(cible(e),Lref(e)))} Lexical entries: 4 Examples: couper to cross; repasser (par) to pass by again Explanation: Such verbs describe a motion going from a near outside to a location of reference, entering the location, crossing it, and going outside to a near outside of this location. 1 0.
Name of the group: Devier to deviate Logical definition: Deviate(e) -+ [P(Source(e),Z-inner-halo(cible(e),Lref(e))) 1\ P(SIP(e),Z-inner-transmit(cible(e),Lref(e))) 1\ P(Goal(e),2-outer-halo(cible(e),Lref(e)))} Lexical entries: 1 7 Examples: devier to deviate; bifurquer to turn off Explanation: Such verbs describe a motion going from the inside to a near outside of an ideal trajectory.
Now, as promised, some words about the tenth group, i.e. the group of verbs such as Devier. In contrast with the other groups for which there is a change of location in the pure definition of the term location, here the location considered is a little special. The location does not consist in a 'normal' portion ofspace, but rather in an ideal trajectory, or ideal path which might be captured by means of aXioms using >. This is reminiscent of the English progressive in which we similarly appeal to 'ideal' or 'inertial' paths. With the progressive,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Name of the group: Decoller to take off Logical definition: Take-off(e) -+ [P(Source(e),2-contact(cible(e),Lref(e))) 1\ P(SIP(e),Z-contact-transit(cible(e),Lref(e))) 1\ P(Goal(e),2-outer-halo(cible(e),Lref(e)))} Lexical entries: I I Examples: se decoller to come unstuck ; deconnecter to disconnect Explanation: Such verbs describe a motion going from an external contact with a location of reference to a near outside of this location.
1 82 A Typology and
Discourse Semantics for Motion Verbs and Spatial PPs in French
such paths can be naturally captured in a non-monotonic formalism {Asher 1 992a).19 We propose the schema depicted in Figure 2 for our ten groups of verbs of change oflocation. 'proximity limits ' � - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - ' '
Passes
Entrer to enter
Z-outer-most
S'eloigner to distance oneselffrom
Lref
·
·
·
�
·
·
·
Partir to leave
Sortir
to go out
Z-outer-halo
'ideal trajectory ' �
Devier to deviate , I
Z-inner-halo
Z-outer-most
Figure 2 The 1 0 groups of verbs of change oflocation
As Tal my { 1 98 3) noted, motion complexes, and for us verbs and prepositions, can focus on particular spatial portions of the STref of a motion. We will make use of (especially in Section 4·5) the Initial, Median, and Final polarities defined in Boons ( 1 98 5) and Laur ( 1 99 1 ). Items with an initial polarity focus on the source of the motion, items with a median polarity on the SIP, and items with a final polarity on the goal of the motion. Our groups S'eloigner, Partir, Sortir, Decoller, and Devier have an Initial polarity. Our groups S'approcher, Arriver, Entrer, and Se poser have a Final polarity. Our group Passer has a Median polarity.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
S'approcher to approach . . . ----i>+- . . . . : Arrive to arrive
Nicholas Asher and Pierre Sablayrolles 1 8 3 4·4 A
classificationfor spatial prepositions
Prepositions
inner-halo
contact
outer-balp
outer-most
Positional
chez; dans at; in
sur; conrre on ; against
sous; derriere; a below; behind ; at
loin de Jar awayfrom
Initial directional
de chez from -'s
de sur from on to
de derriere from behind
de dehors from the outside
Medial positional
par through
au fil de
le long de along
au-deJa de beyond
jusque dans up to the inside of
jusque sur up on to
vers towards
pour for
Final positional
FigUre 3
Classification of French spatial prepositions
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As we have said from the outset, we are interested in the compositional determination of the spatiotemporal properties of motion complexes consisting of a verb of change of location and a spatial preposition. In order to formulate rules of combination (c£ Section 4.5), we briefly present in this section a classification for French spatial prepositions. It is based on the classification proposed in Laur ( 1 991), which we have refined using the spatial relations defined in Section 4.2. Under the term 'preposition', following Laur ( 1 991), we consider simple prepositions (such as dans in ) and complex prepositions, i.e. prepositional phrases (such as en face de infront of). We separate prepositions into two main groups: the positional pre positions, which just describe a relation of localization (such as dans in ), and the directional prepositions, which in addition to a relation of localization also suggest a direction of motion (such as de from , which has an Initial polarity, par through , which has a Median polarity, or jusqu'a to , which has a Final polarity). We set out this classification in Figure 3· In all, 1 99 French prepositions have been listed and classified. We nevertheless have to remark that such a classifica tion using the spatial relations dedicated to CoL verbs is not as refined as we would like it to be. For example, sous below and derriere behind could be differentiated considering the intrinsic orientation of the entity; others require a distinction using the vertical dimension (i.e. the axis of gravity). At present, we are investigating such refinements for motion verbs in a decompositional semantic study (Asher & Sablayrolles, forthcoming) and we intend to generalize them for spatial prepositions.
1 84
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French 4· 5
In providing rules for lexical · entries, we must extend our formal language. Roughly, we will be going from a syntactic parse of a motion complex to the semantic conditions introduced in the DRS. We avoid any issues here about incremental parsing, and in any case we will need information both about the verb and its arguments to introduce the appropriate semantic conditions into the DRSs. To expand our formal language, we will use ¢ as a variable for motion complexes from which we may isolate the verb and propositional phrase adjuncts; though the fragment we will codify here only makes use of one prepositional phrase adjunct, our language will be general enough to handle more prepositional phrases. v will range over verbs in these complexes while n will range over prepositions. Polarity will be used to denote the polarity of the verb (Polarity (v)) or of the preposition (Polarity(n)); IZone(v, mobile(e),l), MZone(v,mobile(e),l) and FZone(v,mobile(e),l) will denote the kind of initial, median, and final zone intrinsically implied by the verb, the mobile, i.e. the moving entity, and a location, which may depend on e, the eventuality that it is an argument to the motion verb of the motion complex. Prep Zone(n,mobile(e),l) will denote the kind of zone intrinsically implied by the preposition and its arguments; Ns(n) will denote the location introduced by the noun phrase in the PP of the motion complex; the question mark (?) will denote a location that will have to be matched with some location given by the discourse (in previous or following sentences) by anaphora resolution in DRT (Asher & Wada 1 989). Finally the DR-theoretic lexical entry for an item v will be denoted ii. Figure 4 illustrates, for each of our ten groups of verbs of change of loca tion, the polarity Polarity (v) and the zones in which the moving entity is located at the initial (variable IZone (v,mobile(e),l)) median (variable MZone(v,mobile(e)), and final (variable FZone(v,mobile(e),l)) stage of the groups displacement, respectively.20 These zones will be related to the Source( e), SIP( e) and Goal(e) by axioms later on in this Section. We show in Figure s. for each group of directional prepositions the polarity (Polarity(n)) and the spatial 'zone' it suggests (PrepZone(n,mobile(e),l)). This 'zone' (c£ remark made in Section 4.3) is a function of the moving entity, the location introduced by the preposition and the preposition itself Finally, we show in Figure 6, for each group of positional prepositions the spatial 'zone' (PrepZone(.n,mobile(e),l)) it suggests. Now we propose the following compositional rules (under the form of logical axioms), with some discursive explanations. In Axioms I I and I 2, > denotes the weak conditional of DICE used in Asher et al. ( 1 993), Lascarides & Asher ( 1 993), and Asher ( I 992b). As in the treatment of 'Discourse in Commonsense Entailment' or DICE developed in Lascarides & Asher ( 1 993),
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1-
Compositional rules
Nicholas Asher and Pierre Sablayrolles 1 8 s
Groups of verbs of change of location Decoller Sortir Partir S'eloigner Devier
MZone
FZone
Z-contact Z-inner-halo Z-inner-halo Z-outer-halo Z-inner-halo Z-outer-most Z-outer-most Z-outer-halo Z-outer-halo Z-outer-halo
Z-contact-transit Z-inner-transit Z-outer-halo Z-outer-transit Z-inner-transit Z-outer-transit Z-outer-halo Z-inner-transit Z-contact-transit Z-inner-halo
Z-outer-halo Z-outer-halo Z-outer-most Z-outer-most Z-outer-halo Z-outer-halo Z-inner-halo Z-inner-halo Z-contact Z-outer-halo
I
F F F F M
Data for French verbs of change of location
Figure 4
Directional prepositions
Polarity
PrepZone
I F F F F M M M M
Z-inner-halo Z-contact Z-outer-halo Z-outer-most Z-inner-halo Z-contact Z-outer-halo Z-outer-most Z-inner-halo Z-contact Z-outer-halo Z-outer-most
de chez de sur de derriere de dehors jusque dans jusque sur vers pour par au 6l de le long de au-dela de Figure 5
Data for French directional prepositions
Positional preposition
PrepZone
chez; dans sur; contre sous; derriere loin de
Z-inner-halo Z-contact Z-outer-halo Z-outer-most
Figure 6
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
S'approcher Arriver Entrer Se poser Passer
!Zone
Polarity
Data for French positional prepositions
--
-- ---
-- -
- - - - -- - ----
1
1 86
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
Asher ( r 993), and Lascarides & Asher ( r 99I), we will later label consituents of a discourse representation as having certain properties, and we will reason about lexical properties within Commonsense Entailment. ·
Axiom 3
( Verb (v, � ) A Change-oflocation (v) A ii(e)) .... P(Source(e), IZone (v, Mobile(e), Lref(e))) Axiom 4
( Verb (v, � ) A Change-oflocation (v) A v(e)) .... P(SIP(e), MZone(v, Mobile(e), Lref(e))) ( Verb ( v, � ) A Change-oflocation ( v) A v( e)) .... P(Goal(e), FZone(v, Mobile(e), Lref(e)))
Axioms 3, 4 and 5 say that.the Source, Path, and Goal of a motion are always a part of a zone given by the verb via !Zone, MZone and FZone, and processed with respect to the reference location Lre£ Axiom 6
( Verb (v, � ) A Change-oflocation (v) A v (e) A Preposition (n , � ) A Directional_Prep (n) A Polarity (v) = Polarity(n)) .... Lref(e) = PrepZone (n, Mobile(e), strej(Ns (n))) Axiom 6 says that when the preposition is directional with the same polarity as the verb (e.g. sortir de Ia maison to go out ofthe house) the reference location (Lref(e)) is a zone given via PrepZone (Z-inner-halo, in our example) and processed with respect to the Ns (Ia maison in our example). Compare that, for example, with sortir de derriere Ia maison to go out ofthe back ofthe house , where the reference location (Lre�e)) cannot directly correspond to the location introduced by the noun phrase (here Ia maison), but has to be built using the PrepZone (here Z-outer-halo for the preposition de derriere) to give the zone of the back ofthe house . Axiom 7
( Verb ( v , � ) A Change-of-location ( v) A ii (e) A Preposition (:rc , � ) A Directional_Prep (n) A Polarity(v) � Polarity (n) A Polarity(n) = I) .... (P(Source(e), PrepZone(n, Mobile(e), strej(Ns (n)))) A Lref(e) = ?)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Axiom s
Nicholas Asher and Pierre Sablayrolles
187
Axiom 8
-+
[ Verb ( v, ¢ ) 1\ Change-of-location ( v) 1\ ii( e) 1\ Preposition (n, ¢) 1\ Directional_Prep (:rr ) 1\ Polarity( v) � Polarity(n) 1\ Polarity(n) - M) (P(SIP (e), PrepZone(:rr , Mobile(e ), strej(Ns(:rr )))) 1\ Lref(e) - ?)
Axiom 9
-+
[ Verb ( v, ¢ ) 1\ Change-of-location ( v ) 1\ ii (e) 1\ Preposition (n, ¢ ) 1\ Directional_Prep ( n) 1\ Polarity (v) � Polarity(n) 1\ Polarity(n) - F) (P( Goal (e), PrepZone(n) , Mobile(e ) strej(Ns(:rr )))) 1\ Lref(e) - ?) ,
Axiom I O
-+
[ Verb ( v , ¢ ) 1\ Change-oflocation ( v) 1\ ii (e) 1\ Preposition (n , ¢ ) 1\ Positional_Prep (n) 1\ Polarity ( v) - I) (P( Goal (e), PrepZone(n) , Mobile (e ), strej(Ns (:rr )))) 1\ Lref(e) - ?)
Axiom 1 0 says that when the preposition is positional and the verb Initial (e.g. sortir dans togo out in/into ) the Goal is a part of a zone given via PrepZone, and processed with respect to Ns; the reference location remains unmatched. Axiom I I
[ Verb ( v , ¢ ) 1\ Change-oflocation ( v ) 1\ ii( e) 1\ Preposition (n, ¢) 1\ Positional_Prep (n) 1\ Polarity( v ) - M 1\ :J ((Info (a , {3) 1\ ( r , a , {3) 1\ Lref(e) - PrepZone(n , Mobile(e), strej(Ns (n)))), R (a , {3) 1\ R � Narration 1\ R � Result) 1\ --. :J ((Info (a , {3) 1\ ( r , a , {J)), R (a , b ) 1\ R � Narration 1\ R � Result)) > Lref(e) - PrepZone(n , Mobile (e), stref(Ns (n)))
Axiom I I encodes a complex interaction between lexical inference and discourse context. It is of the sort we have found common in other areas (Lascarides & Asher I 993). When the preposition is positional and the verb medial (e.g. courir dans to run in ), and if the fact that the reference location is the zone given via PrepZone and processed with respect to Ns enables us to deduce
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Axioms 7, 8, and 9 say that when the preposition is directional with a polarity different from the verb (which is Initial (e.g. sortir par to go out through ), Medial or Final) the Source, Path, or Goal, respectively, is a part of a zone given via PrepZone, and processed with respect to the Ns; the reference location remains unmatched.
1 8 8 A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
in CE a discourse relation R, other than Narration or Result, that we cannot deduce without this information, then we infer that the location of reference is so determined. :l encodes in the object language of CE the non-monotonic consequence relation F· Axiom 12
Axiom I 2 is similar to Axiom I I. When the preposition is positional and the verb Medial (e.g. courir dans to run into ), and if the fact that the Goal is a part of a zone given via PrepZone and processed with respect to Ns enables us to deduce in CE a discourse relation R, other than Narration or Result, that we cannot deduce without this information, then we infer that the Goal is so determined. The reference location remains unmatched. Axiom 13
( Verb ( v, �) 1\ Change-oflocation ( v) 1\ ii (e) 1\ Preposition (:rc , �) 1\ Positional_Prep (:rc) 1\ Polarity(v) = F) -+ Lref(e) = PrepZone (:rc , Mobile (e), stref(Ns (:rc))) Axiom I 3 says that when the preposition is positional and the verb Final (e.g. entrer dans togo into ) the reference location is the zone given via PrepZone, and processed with respect to the Ns. Having detailed in axiomatic form the lexical entries for verb classes and classes of prepositions and their combination, we propose to illustrate their application with the following example: (28), (29) and (3o). 28. Leticia est arrivee par le jardin Leticia has a"ived by the garden 29. Eva est sortie de derriere Ia maison Eva has gone out via the back ofthe house 30. Ines est passee sous le porche Ines has passed under the porch We will derive DRSs for these sentences, so we now briefly mention our views on the construction of DRSs and the syntax/semantics interface. We
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( Verb (v, � ) 1\ Change-oflocation (v) 1\ ii(e) 1\ Preposition (:rc, �) 1\ Positional_Prep (:rc) 1\ Polarity ( v) = M 1\ :J ((Info (a . {J) 1\ (T, a , {J) 1\ Lref(e) = ? 1\ P(Goal(e), PrepZone (:rc, Mobile (e), stref(Ns (:rc))))), R (a , fJ) 1\ R i' Na"ation 1\ R i' Result) 1\ -. ((Info (a , {J) 1\ (T, a , {J)), R (a , fJ) 1\ R i' Na"ation 1\ R i' Result)} > (P(Goal(e), PrepZone(:rc, Mobile(e), stref(Ns(:rc)))) 1\ Lref(e) = ?)
Nicholas Asher and Pierre Sablayrolles 1 89
APAQ
X
P(x) Q(x) This structure says that the discourse referent introduced by the partial DRS above can fill in the argument position in a predicative DRS, which is introduced, for example, by the common noun jardin: AU
jardin(u) They combine to yield the following partial DRS for the noun phrase le jardin: AQ
X
jardin(x) Q(x) Before turning to the first example we should stress again that we take Mobile , Goal, Source , and SIP to pick out semantic arguments of a motion verb; we will use functions, e.g. mobile(e), goal(e), to refer to them. Now, as we have seen, sometimes other elements of the motion complex specifY what these various semantic arguments are. We are skeptical that syntax alone can determine which part of the motion complex specifies which argument in every case, especially with regard to spatial prepositional phrases. Certain lexical conditions or elements in the discourse context (some of which we have
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
employ the 'bottom-up' DRS construction procedure from Asher ( 1 993), in which the function argument structure of the syntax can be followed, if we ignore the problem of scopes of noun phrases and of anaphora resolution. With this approach, determiners are relations between dynamic properties, noun phrases a property of dynamic properties, while common noun phrases and intransitive verb phrases introduce dynamic properties. The account resembles Montague's intensional logic, but instead of formulas of intensional logic we use OR-theoretic notation to denote dynamic properties and propositions. A determiner, semantically, introduces a partial DRS, which when combined with the two dynamic properties yields the semantic value of a DRS (a set of embedding functions from worlds into the domains of those worlds); for instance, le introduces the following structure:
1 90 A Typology and Discourse Semantics for Morion Verbs and Spatial PPs in French
codified in the rules above) determine whether, for instance, the spatial prepo sitional phrase determines the goal or not. With this in mind let us consider example (28) within the bottom-up DRS construction procedure. We take a syntax of the form shown in Figure 7 with a VP internal subject (though this is not essential). Now we simply go back up the tree using the expansions given above of the conditions given by the lexical
� .� IP
I'
VP
A �estrA
•e
PP
VP
V'
Leticia
V
P
arrivee
par
I
Figure 7
P'
DP
I
I
ADP
I
D.
le jardin
The bottom-up DRS construction procedure
elements. In (28) arriver implies that Polarity(v) - F 1\ !Zone - Z-outer-most 1\ MZone Z-outer-halo 1\ FZone - Z-inner-halp . Then, using Axioms 3, 4, and 5 , we obtain the following predicative DRS from the verb: =
A.e.?.x P(Source(e), Z-outer-most(STref(mobile(e)),Lref(e))) P(SIP(e), Z-outer-most(STref(mobile(e)),Lre�e))) P(Goal(e), Z-inner-halo(stref(mobile(e)),Lre�e))) mobile(e) - x Then after conversion with the partial DRS derived from the DP Leticia, we following structure:
have the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I
Nicholas Asher and Pierre Sablayrolles 1 9 1
ux Leticia(x) mobile(e) - x P(Source(e), Z-outer-most(x,Lre�e))) P(SIP(e), Z-outer-halo(x,Lre�e))) P(Goal(e), Z-inner-halo(x,Lre�e)))
A.vA.PA.e'
u u = Z-inner-halo(STre�mobile(e ')),v) P(e ') Lre�e ') = ?
which, when combined with the translation of le jardin, yields:
A.PA.e'
uv u = Z-inner-halo(STre�mobile(e ')),STre�v)) P(e ') jardin(v)
Since the structure above has a lambda abstract for a property variable, it may combine with the property from the VP, given above, to give us: ux Leticia(x) mobile(e') = x P(Source(e'), Z-outer-most(x,Lre�e'))) P(SIP(e'), Z-outer-halo(x,Lre�e')))) P(Goal(e'), Z-inner-halo(x,Lre�e'))) u - Z-inner-halo(STre�mobile(e')),STre�v)) jardin(v)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In (28), we also have the directional preposition par which implies that Polarity(n) = M 1\ PrepZone = Z-inner-halo . Because of the directional preposition and Polarity(v) "' Polarity(n) and Polarity(n) = M, we can apply Axiom 8, which gives us: P(SIP(e), Z-inner-halo (stref(mobile(e)), strej(Ns(n)))) 1\ Lref(e) = ?. We will exploit the function argument structure of the syntax and propose the following translation of the preposition in this case:
1 92
A Typology and Discourse Semantics for Motion Verbs and Spatial PPs in French
Now we apply the partial DRS (givenjust below) derived from the inflection node in the syntactic tree that gives us the temporal conditions and introduces an eventuality filling in the event argument place in the partial DRS given just above.
euxn Leticia(x) mobil(e) = x P(Source(e), Z-ou ter-most(x,Lre�e))) P(SIP(e), Z-outer-most-halo(x,Lref(e))) P(Goal(e), Z-inner-halo(x,Lre�e))) u - Z-inner-hal(STre�mobile(e)),STref(v)) jardin(v) e