JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MANA...
5 downloads
317 Views
16MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MANAGING EDITOR: EDITORIAL BOARD:
Peter Bosch Simon C. Garrod Leo G.M. Noordman Pieter A.M. Seuren
REVIEW EDITOR:
Rob A. van der Sandt
ASSISTANT EDITOR:
Bart Geurts
CONSULTING EDITORS:
W. Levelt (Max Planck Inst. Nijmegen),
J. Allwood (Univ. G<>teborg),
J. Lyons (Trinity Hall, Cambridge),
R. Bartsch (Amsterdam Univ.),
W. Marslen-Wilson (Max Planck lnst. Nijmegen),
J. van Benthem (Groningen Univ.), H.E. Brekle (Univ. Regensburg),
J. McCawley (Univ. Chicago),
G. Brown (Univ. of Ess�x),
B. Richards (Edinburgh Univ.),
H.H. Oark (Stanford Univ.),
R. Rommetveit (Oslo Univ.),
H.-J. Eikmeyer (Univ. Bielefeld),
H. Schnelle (Ruhr Univ. Bochum),
G. Fauconnier (Univ. de Vincennes),
J. Searle (Univ. Cal. Berkeley),
P. Gochet (Univ. de Liege),
A. von Stechow (Univ. Konstanz),
J. Hintikka (Florida State Univ.),
M. Steedman (Edinburgh Univ.),
St. Isard (Sussex Univ.),
G. Sundholm (Nijmegen Univ.),
D. Israel (SRI, Stanford)
Ch. Travis (Tilburg Univ.),
P.N. Johnson-Laird (MRC Appl. Psych.
B. van Fraassen (Princeton Univ.),
Z. Vendler (UCSD),
Unit, ,Cambridge), E. Keenan (UCLA),
Y. Wilks (New Mexico State Univ.),
S. Kuno (Harvard Univ.),
D. Wilson (UCL).
EDITORIAL ADDRESS: Journal of Semantics, P.O. Box 1454, 6501 BL Nijmegen, The Netherlands
Published by Faris Publications,
P.O. Box 509,
3300 AM
Foundation 0167·5133 the Netherlands by ICG
Dordrecht, The Netherlands.
Cl N.I.S.
Printed in
lSSN
Printing
ISSN
0167-5133
JOURNAL OF
SEMANTICS VOLUME 4
(1985)
FORIS PUBLICATIONS Dordrecht
JOURNAL OF SEMANTICS AN INTERNATIONAL JOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MANAGING EDITOR:
Peter Bosch
EDITORIAL BOARD:
S1mon C. Garrod Leo G.M. Noordman Pieter A.M. Seuren
REVIEW EDITOR:
Rob A. van der Sandt
ASSISTANT EDITOR:
Bart Geurts W. Levelt (Max P lanck Inst. Nijmegen},
CONSULTING EDITORS: J. Allwood (Univ. Goteborg),
J. Lyons (Trinity Hall, Cambridge),
R. Bartsch (Amsterdam Univ.),
W. Marslen-Wilson (Max Planck Inst.
J. van Benthem (Groningen Un1v.),
NiJmegen), J. McCawley (Univ. Chicago},
H.E. Brekle (Univ. Regensburg}, G. Brown (Univ. of Essex),
B. Richards (Edinburgh Univ.},
H.H. Oark (Stanford Univ.},
R. Rommetveit (Oslo Univ.),
H.-J. Eikmeyer (Univ. Bielefeld},
H. Schnelle (Ruhr Univ. Bochum},
G. Fauconnier (Univ. de Vmcennes),
J. Searle (Univ. Cal. Berkeley},
P. Gochet (Univ. de Liege),
A. von Stechow (Univ. Konstanz),
J. Hintikka (Flor!da State Univ.),
M. Steedman (Edinburgh Univ.),
St. Isard (Sussex Univ.),
G. Sundholm (Nijmegen Univ.),
D. Israel (SRI, Stanford)
Ch. Travis (Tilburg Univ.),
P.N. Johnson-Laird (MRC Appl. Psych.
B. van Fraassen (Princeton Univ.},
Unit, Cambridge},
Z. Vendler (UCSD},
E. Keenan (UCLA},
Y. Wilks (New Mexico State Univ.},
S. Kuno (Harvard Univ.),
D. Wilson (UCL).
EDITORIAL ADDRESS: Journal of Semantics, P.O. Box I454, 650I BL Nijmegen, The Netherlands
Published by Foris Publications, P.O. Box 509. 3300 AM Dordrecht, The Netherlands. Cl N.I.S. Foundation ISSN 0167-5133
Printed in the Netherlands by ICG Printing
JOURNAL OF SEMANTICS CONTENTS VOLUME 4 ( 1 985)
A rticles FRITS BEUKEMA, Chronological Relations and Free Adjuncts in English . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
101
D.S. BREE, The Durative Temporal Subordinating Conjunctions Since and Until . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . DAVID M . CARTER, Common Sense Inference in a Focus-guided Anaphor Resolver . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . BERNARD COMRIE, Reflections on Subject and Object Control
237 47
..................................
247
CHRISTA HA UENSCHILD, Definite vs. I ndefinite I n terpretation of Russian Noun Phrases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
37 1
H.J. HERINGER, The Verb and its Semantic Power: Association as a Basis for Valence Theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
79
H.J. VAN JAARSVELD and R. SCHREUDER, Implicit Quantification of Temporal Adverbials . . . . . . . . . . . . . . . . . . . . . . . .
327
SEBASTIAN LOB NER, Definites
............................
279
ILKKA NIINI LUOTO, Imagination and fiction . . . . . . . . . . . . . . . .
209
BOELE DE RAAD, The Language of Social Acts
223
BART GEURTS, Generics
..............
GUNTHER ROHDENBURG, Dogs, Bitches and other Creatures
1 17
LI LIANE TASMOWSKI and S. PAUL VERLUYTEN, Control Mechanisms of Anaphora
34 1
JAN DE VUYST, The present Perfect in Dutch and English
1 37
.....
THOMAS Postulates
EDE
ZIMMERMAN,
A
Note
on
Transparency
67
Review A rticle ARNIM VON STECHOW: Max J . Cresswell, Structured Meanings. The Semantics of Propositional A ttitudes . . . . . . . . . . . . . . . . . . . . . . .
1 65
Book Notes AD FOOLEN: Gunter Saile, Sprache und Handlung
391
ULF FRIEDRICHSDORF and PETER S CHROEDER-HEISTER: G.E. Hughes and M.J. Cresswell , A Companion to Modal Logic . . .
389
B ook Reviews JAN LEMMENS: Jaakko H intikka (in collaboration with Jack Kulas), The Game ofLanguage. Studies in Game-Theoretical Semantics and its Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
265
PETER ROLF LUTZEIER: Fred Landman and Frans Veltman (eds.), Varieties of Formal Semantics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
193
D . E. OVER: Robert Stalnaker, Inquiry
........................
271
THOMAS PECHMANN: J. Engelkamp and H.D. Zimmer, Dynamic A spects of Language Processing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
200
WALTER J . PERRIG and H UBERT D . ZIMMER: A.J. Sanford and S.C. Garrod, Understanding Written Language: Comprehension Beyond the Sentence . . . . . . . . . . . . . . . ... . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
204
HAN R EICHGELT: T. van Dij k and W. Kintsch, Strategies of 275 Discourse Comprehension
D iscussion H. ECKERT: Valences LTD vs. Valences Associated, Comments o n Heringer's Association Experiment as a Basis for Valence Theory . H .J. HE RINGER, The Verb and its Semantic Power: Association as a
257
..................................................
1 00
Obituary
EDITOR'S NOTE
The editors wish to express their gratitude to the following colleagues who are not consulting editors of this Journal but have been kind enough during the past year to referee papers that were submitted for publication : D . S . Bree C. Gussenhoven G.L. M . van H ec k L . Hom R . Kempson A.A. McDonald R . van Oirsouw F. Plank G. Redeker H. van Riemsdij k B. Rigter C.-R. Rollinger A. Sanford A.J. Smolenaars W. Spooren L. Stassen P. Staudacher K. Stenning L. Wetzels D. Wunderlich The editors also want to thank Michael Arbib, F rank Heny, H an nes Rieser, and Robert Stalnaker for the support they gave to the Journal in their function as consulting editors during the period 1 982- 1 98 5 .
Literature & Linguistics
German, Romance, and English Studies
Max Niemeyer Verlag Tiibingen would like to announce that our books are now available for distribution
in the United States and Canada. You will receive prompt delivery of any of our recent publications as well as
a large number of our university course books and standard works.
Books which are indicated provisionally can be reserved and will be delivered on publication. Please address your order to:
Foris Publications USA, Inc. P.O.Box
5904
Providence, RI or call
02903, USA
(401) 232-27 22.
A complete stock list including publications since �98 5, our General Catalogue in German, and a special catalogue "Literature & Linguistics: Spring/Summer '87'' of forthcoming books are available. Orders for books which are not on stock in Providence can be placed either directly or with Faris Publications USA, Inc.
Max Niemeyer Verlag P.O.Box
�
2140 - D-7400 Tt.ibingen!West Germany
Max Niemeyer Verlag
COMMUNICATION AND COGNITION An International Interdisciplinary Ioumal focused on the interrela tions
between communication and
cognition
AIMS AND SCOPE C.0 CjX -----_
Communication &- Cognition is an interdisciplinary workgroup. Their objective is the study of the interrelations between communication & cognition as realized
in the fields of linguistics, logic, psychology, scientific methodology, artificial in teiiigence, information sciences, anthropology, and sociology. Toward gaining this
objective, Communication & Cognition publishes two journals, topical series, books and a newsletter. It organizes as weii workshops, seminars, symposia, and congresses.
&
FORTHCOMING ISSUES
Cognition, Emotion, and Behavior - M. Boekaerts, M & S. Maes (Eds.) with planned contributions by D. Eiser & C. Eiser, M Johnston, F. Vand amm e, M. Boekaerts, S. Maes, G. Van Heck, J. Hettner, B. Bermont, I. Sarason and B. Sarason, D. Rost and G. d'Ydewalle.
&
BOOKS Vandamme Femand, Informatika, of de Leer der Symboolverwerkende Systemen, Communication & Cognition, 1986, 255 pp., 450 BF. ISBN 90-70963-15-9. Aagaard-Mogensen L., Pinxten R. & Vandamme F. (Eds.), Wordlmakings' Ways, Communication & Cognition, 1986, 190 pp., 550 BF. ISBN 90-7-963-20-5.
&
SUBSC RIPTION INFORMATION Subscription rate: Institutions: 850 BF; individuals is 650 BF. Payment: by International Money Order, Personal Cheque (150 BF Billcharges if paid in non-Belgian currency), or with Eurocard I Accesscard I Mastercard I Din crsclub I American Express I Visa.
COMMUNICATION & COGNITION,
Blandijnberg 2, B-9000 Ghent/Belgium
LANGUAGE& COMMUNICATION An Interdisciplinary Journal
University of Oxford, Worcester College, Oxford OX1 2HB, UK
Editor: R HARRIS,
The primary aim of this uniquej ournal is to fill the need for a publications! forum devoted to the discussion of topics and issues in communication which are of interdisciplinary significance. It publishes contributions from researchers an all fields relevant to the study of verbal and non verbal communication. The investigation of language and its communicational functions are treated as a concern shared in common by those working in: anthropology, the arts, artificial intelligence, education, ethology, linguistics, physiology, philosophy, psychology and the soc1al sciences. Emphasis is placed on the implications of current research for establishing common theoretical frameworks within which findings from different areas of study may be accommodated and interrelate
US$90.00 US$171.00
FREE SAMPLE COPIES AVAILABLE ON REQUEST Adve11Jo ing rote card available on request. Bocl¢ i&Riea end ere also available in mlcrofonn. Priceo are to change without notice. Journal priceoo Include postage end in&urence. Starting prlceo are avoliable to UK and Ei"' customera on request.
current
aubsc:riptiont aubject
A selection of papers
Language and communic11tionaj efficiency: the case of TokPisin,P MUHLHAUSLER. What has gender got to do with sex? D CAMERON. En glish as the second language of the deaf, M DEUCHAA & H JAMES. The interfacing of language and music, M YCHEN. Intonation and meaning in conversation, G McGREGOR. Kinetic patterns of individuality in acoustic and g raphic productions, G G HAYDU & S FELDMAN. The garden path that historical linguistics went astray on, CH-J N BAILEY. Realism and language change, TPATEMAN. Against affability, M L SCHNITZER. The Shark Bay language and its implications for linguistic theory, J B M GUY. The demarcation of metaphor,P PETTIT. Evaluation of noncontent speech accommodation, R L STREET Jr. The messag e, the context and the medium, A FURNHAM. The future machine: a study of the span of speakers' anticipations in conversation, D D CLARKE. Language acquisition, date compression and generalization, J G WOLFF. What we might learn from acquired disorders of reading, R B FRIEDMAN. The social context of language acquisition, J BRUNER.
e
���!�=�:O
SCOPE OF THIS JOURNAL The JOURNAL OF SEMANTICS publishes articles, notes, discussions, and book reviews in the area of natural language semantics. It is explidtly interdisciplinary, in that it aims at an integration of philosophical, psychological, and linguistic semantics as well as semantic watk done in artificial intelligence and anthropology. Contributions must be of good quality (to be judged by at least two referees ) and should relate to questions of comprehension and interpretation of sentences or texts in natural language. The editors welcome not only papers that cross traditional discipline boundaries, but also more s le to and interesting for a wider specialized contributions, provided they are accesib readership. Empirical relevance and fonnal correctness are paramount among the criteria of acceptance for publication
INFORMATION FOR AUilfORS Typescripts for publication should be sent in 3 copies to the managing editor. They should be typed on M (or similar format), one-sided, double spaced, and with a wide margin and must be accompanied by an approx. 200 words summary. Footnotes and bibliographical references must appear at the end of the typescript. Diagrams must be submitted camera-ready. All papers submitted are subject to anonymous refereeing. Authors receive 20 offprints of their published articles and 10 offprints of their publish ed reviews, free of charge. IMger numbers can be supplied at cost price by advance ar rangement. Unless special arrangements have been made , copyright rests with the NIS Foundation. PRICES AND CONDffiONS OF SUBSCRIPTION The JOURNAL OF SEMANTICS appears in four issues per year of approx. 100 pages each. Subscriptions for private use are available at the reduced rate ofDtl. SO,- per year; the institution rate Is Dtl. 120,-. The price for single issues Is Dtl. 35,- (reduced rate: Dfl. 15,-). All prices exclusive packing and postage . Airmail- and SAL-rates are available on request. We regret that no delivery can take place before payment has been received Subcriptions not cancelled before October 1st automatically extend to the following year. Placement of orders implies the consent of the subscriber to these conditions. All orders should be sent to Faris Publications, P.O. Box 509, 3300 AM Dordrecht, The
Netherlands.
\1
.Jmc
4
De ember 19 S
Num�r �
(0'HNTS
..,I BA�IIA"'-i I ()n-..rR, D.Jannes H I V \"' J \1\RS\ I LD und R. "t( HRH D£R, Jmphc.n Quanuftt tton of Iempurtl Ad\cTbJah .. L Ill!\NL Mtclum.m
I A�MOWSKI
<•I
nntl S. PAll VERl t:\'lll'i. Control "\41
1\n.Jphora
( II RI�I A II 1\t I N'iC IIII () 1f R1
'\27
'l '\n•m Phr
l IF F RH URIC H�DOR� Hu•h " 1nJ \1 J Cr
(, I
\U r 001 E "- Cai.ir �
Dduutc
"
lndc;lrntte lnt
rprct.ltt n '\71
"C'i
r
\ •I
�d PET£ R S(HROE01 R-11£1<\IE R \liCII
•
rf
Compamon
ro
A.fodullo
\Jr tht und 1/or:d/unrr
tc
�flo) l')J
Journal of s�mantics
4
: 279-326
DEFINITES*
SEBASTIAN LOBNER
ABSTRACT The paper develops a uniform analysis of the defimte article in its various uses and of definite
1), 11 is argued that the logical role of
definites is that of indiVIdual expressions (terms). To treat definites as quantifiers is logically
inadequate, but (referential) quantification involves definite reference to the domain of quantifi cation (C!t. 2). In Ch. 3 a distinction between functional, relational, and sortal concepts is introduced along with a corresponding subcategorization of noun occurrences. This enables a systematic categonzation of the uses of definites, the main distinction being that between semantic and pragmat1c definites (Ch. 4). Semantic definites represent functional concepts which ex1st independently of the particular situation referred to, whereas pragmatic definites depend on that SituatiOn for unambiguous reference. In sketching a more elaborated theory of anaphors it is finally argued that the definite article always md1cates that the noun is to be taken as a functional concept. Definites in general receive functional concept Interpretations, but the definite article is unique in 1ts role: all other determiners require a sortal or relational concept interpretation of the noun.
I. INTRODUCTION
1.1. Definite Noun Phrases and Definite Descrip tions In what follows I shall refer to any N Ps which consist of the definite article the and a following simple or complex noun globally as "definite descriptions" (DDs). DDs are only a special case of definite NPs. Definiteness is not tied to the occurrence of the definite article. Rather it is to be considered a property of certain NPs and ProNPs, including complex N Ps with a demonstrative or possessive determiner, nouns with preceding genitives, as well as proper names, personal pronouns, and demonstratives. In German , certain preposi tional phrases in which the preposition and the definite article are contracted to a reduced form (as in zur < zu der, beim < bei dem, fii rs
•
The research reported in this paper was partly financed by the Deutsche Forschungsgemein
schaft in the project Wu 86/6 "Quantoren im Deutschen". I would like to thank the anonymous JS reviewers for helpful comments on a previous version of this paper.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
NPs in general. Startmg from the observation that singular and plural mass and count definites all refer to single, though poss1bly complex objects (Ch.
280 1. 2. Start and Goal The main theme of this paper is the meaning and function of the definite article. Semantically, however, definite descriptions are not a natural class. The analysis is therefore embedded in a study of definite NPs in general. I shall try to follow some basic heuristic principles which aim at conceiving the semantics of the definite article in a global way rather than restrict the consideration to some special cases. The first assumption is in the spirit of rigid compositionalism : (I)
I shall not pick out some special cases like anaphoric and deictic uses ofDDs. I n principle, also generic uses of DDs should be covered, but for the moment I have nothing to say about them. Many, more or less unrelated languages have a definite article. An analysis of the semantics of the definite article should not be l anguage-specific: (II)
The meaning of the definite article is essentially the same in different languages.
This principle does not rule out the possibility that the definite article in one language is used in cases where it is not used in another language. The meaning of an expression constrains its use but does not determine it. For example, in English in certain cases a possessive pronoun is used where in German or French the definite article is (at least) preferred: (1)
English: He kept his hand i n his pocket. German : Er hatte die Hand in der Tasche. French: II avait Ia main a Ia poche.
I shall try to do justice to principle (II) by describ ing the meaning of the definite article in terms which do not seem to be language-specific. Opposed to languages which have a definite article there are many languages that do not. (Some languages without a definite article have equivalent v:ays of expressing definiteness explicitly; cf. Kramsky ( 1 976) for a survey.) Many languages, however, e.g. Japanese, normally do not distinguish between definite and indefinite NPs. These languages presumably manage to express the same kinds of things without an article. Thus the meaning of the definite article cannot have any "substance" like the meaning of nouns, verbs, or adjectives. Let me put i t this way:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The definite article has one and the same meaning in all its uses. This applies to count noun as well as mass noun cases, and for singular and plural likewise.
28 1 (III)
The definite article is just the indication of a certain way of using the noun it is attached to.
This specific way of using the noun (or processing its meaning) is marked by the explicit forms of definiteness i n those languages which possess t hat possibility and it is not made explicit in those which lack a comparable device. This does not mean that the defi n ite article has no meaning at all . Nobody would say that n umber has no meaning. But number, too, is n ot made explicit i n many languages. 1.3. Singular and Plural, Mass and Count Dejinites
(2)
a. b.
She always puts an apple into the salad. She always puts apple into the salad.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I n traditional formal semantics, a sharp distinction is drawn between singular and plural D Ds and between DDs with a count or a mass noun head. Semanticists were forced to consider these cases as different , as long as they kept to the Russellian analysis of d7fin ites. According to Russell ( 1 9 19: Chs. 1 5, 1 6) singular count noun DDs are to be taken as existential quantifiers u niquely describing individuals in the sense of logic. Plural DDs, however, would not correspond to individuals but to classes of individuals. Thus, singular and plural DDs and consequently the singular and the plural definite article do n ot only have different meanings but even different logical types. Singular mass nouns seem to represent yet another type: they do not fit at all into any treatment along these lines. In view of the linguistic data, however, there is no evidence for such fundamental distinctions. If D Ds of the three possible kinds -plural count, singular count and 1 singular mass -differ in meaning, it would be most natural to attribute t he difference to grammatical nu mber and to the subcategorization of the noun, i nstead of treating the definite article as ambiguous. 2 It has been pointed out by Hawkins ( 1978:11 0ff.) and other authors that the syntactic and referential role of the different kinds of definites is the same: all three types of D Ds have the same complex range of uses. As Bunt ( 1 985: !Off.) has argued, the mass/count distinction does not apply to nouns as such but to noun occurrences. Some syntactic configurations require a mass noun interpretation and others a count noun interpretation. In particular, i ndefinite determiners and determiner phrases split into two subclasses with little intersection. Definite determiners however, are not discri minative. It seems to be reasonable to consider them neutral with respect to the mass/count distinction. As for anaphoric DDs, to mention only one frequent case, the DO inherits the interpretation of the noun from its antecedent. Consider the following two sentences (2a) and (2b) and their continuation (3):
282 (3)
The apple makes the salad a little sweeter.
Apart from being introduced as a count or a mass term in (2a,b), there is no clue to decide the mass/count status of the noun apple in (3). Taken in isolation, the syntactic configuration the+ singular noun in (3) is unspecific. The most natural interpretation of DDs is to take them as referring to one object in every case. Consider e.g. an anaphoric plural DO, such as the children in the following example (4)
She has two children . . . . The children are twins.
2.
THE
LOGICAL
TYPE
OF
DEFINITES
AND
THEIR
RELATIONSHIP
TO
QUANTIFICATION
Russell has introduced the idea that the individuals of first order predicate logic be identified with those objects to which proper names refer and which (referential) singular count DDs "describe". Plural NPs, in contrast, do not correspond to individuals but to classes, i.e. entities of a higher logical type (cf. e.g. Russell 19 1 9 : Chs. 1 5 / 1 6). For mass nouns there was no room c.t all in the logical framework. This is, however, only a singular theoretically possible application of first order predicate logic to natural language semantics, and in fact an inadequate one. In what follows, I will argue on logical grounds for a uniform treatment of singular, plural and mass DDs as individuals, which is one precondition for a uniform analysis of the definite article. An analysis with these formal properties and an alternative predicate-logical framework has been presented by Link ( 1 983).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The DO refers to one complex discourse referent introduced i n the first sentence as "two children". The plural in the continuation does not mean that there are several different discourse referents but that the discourse referent in question is composed of several units of the same kind. The mass/count status and the gram�atical number of the head noun can be considered part of its meaning. The definite article indicates that the DO refers to that, possibly complex, object to which the nou n , as a predicate, applies in the situation referred to. The children refers to the entire complex object to which children applies; the child to the entire object to which child applies (which is necessarily only one child); and the snow to the entire object to which snow applies. Lin k ( 1 983) has developed a logical system that allows a uniform treatment of all DDs in this manner. Individuals in Lin k's system can be either individu als in the classical sense, or "sums" of simple individuals, or homogeneous quantities of the same substance. In the next section I will provide semantic evidence for the adequacy of such a system for the interpretation of n atural language NPs.
283
2. 1. Definites are Individual Terms In the following discussion, I assume generally that natural language senten ces and the logical formulae thought of as equivalents have a truth-value only if certain conditions (i.e. presuppositions) are fulfilled. A sentence or formula is either true or false or neither. A sentence is true if and only if its negation is false. Negation, hence, is the strong, presupposition-preserving variant, and falsity must be distinguished from mere non-truth. Consequently, a sentence If' and its negation -q� either have opposite truth-values, or both lack a truth-value. In predicate logic, individual terms, or "terms" for short, combine with predicate symbols to form a formula. Let P be a one-place predicate constant and t a term, then P(t) is such a formula. For every predicate symbol P we can introduce its negation P. PrtJ is true if and only if P(t) is false. Let us say "P is true for t" if P(t) is true. The latter expression can also be used with arbitrary NPs in the place oft. The following conditions are formulated independent of the specific syntactic status of individual terms. In predicate logic, individual terms have the following logical properties, which are all concerned with negation : (TI)
Individual· terms cannot be negated.
(T2)
(Consistency) If P is true for an individual term t, then P cannot be true for t.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Formal semantics, from Montague ( 1 973) to Barwise & Cooper (198 1 ), has adopted Russell's identification of singular count terms and individuals. Definite NPs, now, however were "raised" to generalized quantifiers, deno ting sets of properties of individuals rather than individuals. This step made it possible to treat definite, indefinite, and quantificational N Ps uniformly as second order predicates, yielding a correspondence between the syntactic structure and the semantic interpretation of NP+VP-sentences which could not have been closer. The relationship between syntax and semantics should, however, be sub ject to empirical investigations and not be fixed as strictly isomorphic as is done in the dogmatic approach of Montague Grammar. Syntax and seman tics are (at least possibly) governed by independent regularities. In particular, it is not open to an arbitrary decision what logical status is to be assigned to certain classes of expressions. I ndividual terms e.g., and quantifiers play specific and different roles in the system of first order and higher predicate logic. If NPs are assigned the status of one or the other kind of expression in . the logical framework used, then their actual semantic properties should match those of their "translations" in logic. It will be argued, that natural language NPs do not form a logically uniform class. Definites are individual terms and as such have different logical properties than quantifiers, which in turn are the proper correspondents of quantificational NPs.
284 (T3)
(Relative Completeness) If P is false for an individual term t, then Pis true for t.
We can now decide if an NP must be analysed as an individual term or not by checking the respective properties: (T l ')
The NP can be negated.
(T2')
If (NP+ VP) is true and VP is the negation of VP, then (NP+ VP) cannot be true.
(T3')
If (NP+ VP) is false, then (NP+ VP) is true.
(I)
Many people have dish-washers and many people don't.
(2)
O ne girl is tall and one girl is small .
(3)
I have a car and you have a car.
The latter is not an overt case but is usually interpreted in a way that I have is incompatible with you have. Also condition (T3) does not hold for indefinite NPs in general. If, say, three students are Italians is false, it does not follow that three students are not Italians. (Imagine a case of four students, two Italian , and two not.) Quantificational NPs, to exclude a . second group of NPs, can violate each of these conditions. Quantifiers are semantically second order predicates (cf. Barwise & Cooper 1 9 8 1 , and Lohner 1 986) and as such they can be negated. Note that the negation of a quantifier dominating a sentence is tantamount to the negation of the whole sentence ("outer negation" ) and, in general, has a different semantic effect than the negation of the predicate ("inner negation"). At least some quantificational N Ps can be negated: (4)
Not
{ Mkny } children are watching the TV.
(T2) is violated by the so-called "weak" quantifiers including some N, many N and the like (Cf. Barwise & Cooper 1 98 1 : 1 82, 2 1 9). Universal quantifiers fulfil condition (T2), but not (T3), as the following examples show. Let us consider a group of children and assume that children are either boys or girls. Then the predicate "boy" is the negation of the predicate "girl" (for children). Now, the sentences
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let me start the discussion with the observation that indefi nite NPs are not terms since they characteristically violate (T2). Cases where both a predicate and its negation yield true sentences with an indefinite NP are easy to imagine.
285 (5)
All children are boys.
(6)
All children are girls.
can not both be true. 3 This illustrates t he consistency of universal quantifiers. But they can both be false (if the group of children is mixed). And this contradicts condition (T3), since the falsity of (5) does not automatically lead to the truth of (6). In contrast to quantificat ional and indefinite NPs, definites fulfil all three conditions. To begin with, they cannot be negated (T l ):
*Not
l
the children are the chil � i � the mUSIC IS
I
playing.
Negation is possible, though, in combination with focusing on the subse quent NP:
(7) a.
Not
l
the C H I LDREN are the CHILD is the MUSIC is
I
playing.
Focusing affects, however, the semantic structure of the sentence . Sentences (7a) are different from sentences (7). (7a) presupposes that somebody (or something) is playing and denies that it is the referent of the NP who is playing. Horn would call this an i nstance of "metalinguistic negation" (cf. Horn 1985, in particular pp. 166 ff.). I n contrast to the truth-fu nctional presupposition-preserving "descriptive negation " , with which we are concerned, metalinguistic negation can be appl ied to virtually every part of a sentence. Note that cases such as (7a), but not (4), call for a contrastive but-conjunction . In addition to condition ( T 1 ), defi nites obviously satisfy the consistency condition (T2): it is impossible that both a predicate and its negation are true for a definite NP. Sentences such as (8) are contradictions.
(8)
!
The children are The chil � i � The mUSIC IS
I
l
the children are loud and the chil� i � the mUSIC IS
I
not loud.
Condition (T3) is tantamount to the claim that the negation of the whole statemen t is equivalent to the negation of the predicate. For singular definites the equivalence is obvious. (9)
Cesar is a good fellow.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(7)
286 is false if and only if (9')
Cesar is not a good fellow.
( 1 0)
The children are dirty.
(II)
The children are clean .
( 10')
The children are not dirty.
(II')
The children are not clean .
But let us now assume that some of the children are dirty and some are not. Then, obviously, neither sentence ( 1 0) nor (II) is true. But they are not false either, not in the sense of "false" used here. If they were false, in the strong sense, then their respective negations ( 10') and (II') would be true. B ut they aren 't. ( 1 0), (II), ( 10') and (II') are thus all truthvalueless. It may be worthwhile to point out that the truth-value gap observed here is not specific for plural DDs. Such gaps occur in a com pletely analogous fashion whenever the predicate at issue happens to apply only to part of the i ndiviual referred to. Imagine little John among those children. John did not fall into the pool but stepped into it up to his knees. He is partly dirty but the rest of him is clean. Consequently the sentences ( 1 2) and ( 12') and their negations all lack truth-values in that particular situatio n : ( 1 2)
John i s dirty.
( 12')
John is clean.
If a definite NP is used as an argument of a predicate, the predicate must be
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
is true. But how about plural definites? In the Russellian tradition of formal semantics the plural definite article is treated as a synonym of all. If this were correct, plural DDs should violate the completeness condition in the same way as universally quantifying NPs do. They should exhibit a difference between the negation of the p redicate and falsity of the sentence. Let us consider an example. Suppose there are ten children playing in the backyard, near a pool of muddy water. Some of them may have fallen into that pool and are dirty. The others are clean . We assume that each child is either clean or dirty. The p redicates "dirty" and "clean" can thus be considered negations of each other i n this particular situation. If the group of children is homogeneous, i.e. if either all of them or none are dirty, there are no p roblems: sentences ( 1 0) and (II') are true if (II) and ( 10') are false and VICe versa:
287
2. 2. The Relationship Between Dejinites and (Referential) Quantification It may be useful at this point to go further into the details of the relationship between definites and quantification. I n Barwise and Cooper ( 1 9 8 1 ) and in LOhner ( 1986), a general notion of quantification is outlined , according to which there is always a fixed domain of quantification involved. Qantificati on, then , specifies if and to what extent the predicat 'e which is COll}bined with the quantifier applies to elements or parts of the domain of quantificatio n . In LOhner (forthcoming) it is argued that two kinds of quantification must be distinguished: generic and referential quantification. The basic picture of Generalized Quantifier Theory, with a given domain of quantification as the basis of quantifications, applies only to referential quantifica tion . Let me briefly illustrate the difference. Referential quantification refers to a given domain of quantification. Referential quantifiers can be paraphrased by partitive construction in which the domain of quantification appears explicitly as a definite NP. Consider the following cases:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
applied to the referent of the NP as a whole, as one thing. But if the referent is not homogeneous with respect to the predicate, if it is split , so to speak, by the predicate, then the predicate will not yield a truth-value. (One way out would be to take a majority view. To consider the children dirty, e.g. , if the majority of them are. But in this case, again, the application of the predicate would be such that P is true if and only if P is false.) For every NP there is a range of predicates that yield a truth-value in a given situation , and under the constellation considered the predicates "dirty" and "clean" do not belong to the range of the children. (They do belong, however, to the predicate range of all children.) Hence, since for arbitrary predicates P, either P is true and its negation false or neither P nor its negation yield a truth-value, it can still be mainta ined that if P is false then its negation is true. The relative completeness condition holds for all kinds of definite NPs. There may, however, be truth-value gaps. Completeness is thus only completeness relative to the respective range of predicates. We have thus shown that all definites and only definites exhibit the characteristics of individual terms in the sense of predicate logic. In particu lar, they do not share the properties of quantificational NPs which character istically violate conditions (TI) and (T3). 4 The analysis of definite NPs as individual terms makes a theory of predica tion both necessary and possible. If the denotation of an NP is a complex object, there are in principle several ways to apply a predicate to it. Some predicates will apply to the whole via application to every part (distributive predicates like snore), others will apply to the whole but not to parts (collect ive predicates like gather), and some will be unspecific (play, weigh). In general the global application to the whole will be less strict than the explicit application to every single element in case ofNPs containing every and each . 5
288 (13)
a. b.
(All I Most I Many I Few I Some I No} apples are sour. {All I Most I Much I Little I Some I No} water is poisoned.
If ( 13) refers to a certain quantity of apples (water), out of which some may be sour (poisoned) and some not, ( 14) is a paraphrase of ( 13): ( 14)
a. b.
(All I Most I Many I Few I Some I None} of the apples are sour. (All I Most I Much I Little I Some I None} of the water is poisoned.
( 1 5)
a. b.
The apples are The water is
all to the most part to a great part to a small part to some part (not)
sour. poisoned .
The corresponding paraphrases for the generic interpretation would be quite different . ( 16) has an indefinite plural subject and a quantificational adverb of frequency:
( 16)
a. b.
Apples are Water is
always mostly often seldom sometimes never
sour. poisoned.
The examples show t hat referential quantification involves defini te reference to a domain of quantification. In fact, referentially quantifying NPs can be used in the same configuration as si mple plural count or singular mass DDs referring to the respective domain of quantification. The effect of a prenomi nal or adverbial quantifier in addition to the reference to the domain of quantification, then, is a differentiation of the predication . Without a quanti ficat ional expression there is only the alternative between global truth and global falsity o'f the predicate with respect to the domain referred to. Quanti-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
But the sentences in ( 13) might also be taken as generic statements about apples and water in general. And under t his interpretation, ( 14) is not a paraphrase. This distinction becomes clearer if one considers adverbial instead of nominal quantification. Under the referential interpretation, ( 13) can' be paraphrased as ( 15), with a definitive subject and a quantificational adverb of extent :
289 fication allows the expression of intermediate cases and thereby extends the predicate range for the sentence. This point is also illustrated by examples (IO)-(II) and (5)-(6) above. If the group of children is mixed, the respective predicates do not belong to the range of the definite NP the children , but they belong to the range of any quantificational NP with the head child/children , regardless of the composition of the group. It follows, again, that the logical and the syntactic status of definite NPs and quantificational NPs is different. Quantificational NPs cannot replace the DO in sentences with adverbial quantifiers or in partitive constructions6 simply because the same predication cannot be quantified twice.7
2.3. Quantificational Theories of DDs
(17)
3x(N(x) & VP(x) & Vy(N(y) - y=x))
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Quantificational theories of noun-phrase semantics, from Montague's PTQ (1973) to Generalized Quantifier Theory (GQn, as introduced in Barwise and Cooper (1981), treat all NPs as quantifiers. Even simple personal names and pronouns are raised to the level of second order predicates (i.e. quanti fiers in the sense of GQT). This way a main obstacle to a logical semantics for natural language seemed to have been overcome: the apparent logical inho mogeneity of the category of NPs. The basic idea of Montague (1970a) was a grammar for natural language with a one-to-one correspondence between syntactic and semantic structure, a very rigid version of Frege's principle of compositionality. A closer investigation of linguistic data of the kind consi dered suggests, however, that NPs are not a homogeneous syntactic and semantic category and that the relationship between syntax and semantics might be more complex than in Montague's original model. Meanwhile, revisions of the uniform treatment of NPs are being developed, cf. Link (forthcoming) and L0nning (forthcoming): Independent arguments against a quantificational treatment of definites (and indefinites) have been put forward from the standpoint of discourse representation theory (Kamp 1981, Heim 1982, Barwise 1985). Quantificati on introduces scope boundaries that cannot transgress the boundaries of a sentence. This fact is incompatible with the idea of anaphoric reference as coindexing of discourse referents. Although these arguments are not absolu tely cogent - in principle, anaphora might be analyzed in a different way they are at least highly suggestive. (This does not imply that I consider the Discourse Representation Theories referred to as an adequate model for the general semantics of NPs. I'll come back to this point towards the end of the paper.) Let me conclude the discussion of quantificational theories of DDs with a last point. Russell's analysis considers existence and uniqueness of the refer ent of the noun of singular count DDs as the essence of the meaning of the definite article. The sentence the N VP "translates" into
290 In this interpretation, the role of the noun is that of a sorta/ concept: there is one and only one object of the sort N. Hence the uniqueness clause; which is necessary because, for any sortal concept N, it is always possible that there is more than one object of that sort. In what follows, I shall argue that this is not the typical use of the definite article. Often the definite article is combined with a noun that represents a concept which necessarily applies to only one object (at most), cf. DDs such as the sun, the present president of the U. S., the smallest prime number. If N is such an inherently non-ambiguous expression, the formula (17) becomes in some cases logically equivalent to (18), since the uniqueness clause is then redundant: (18)
3x(N(x) & VP(x))
(19)
a. b.
The smallest prime number is 2. A smallest prime number is 2.
2. 4. Uniqueness vs. Non-ambiguity The problem with Russell's account is that it is not uniqueness which is essential to definiteness. Uniqueness is a (possible) property of sortal con cepts, namely the property that there is only one object of the respective kind. Uniqueness theories of definites consider the head noun complex of a DO to provide sortal information about the referent. This sortal information must apply to exactly one object in the respective context. Thus, the role of the noun is to provide contrastive information which allows to single out the intended referent. This conception lies at the ground of Russell's analysis as well as the approaches of Hawkins and Heim. Hei!T), for instance, takes the definite article to impose the constraint that there must be a discourse referent previously mentioned or else "familiar" (i.e. immediately salient for deictic DDs), which satisfies the sortal informati on provided by the head noun. Definiteness is thus reduced to the property of the referent of being the unique object which exhibits certain characteristics that distinguish it from other possible referents. The basic idea of Hawkins approach is similar. He considers it essential for definiteness that the referent belongs to a certain "shared set". Both Heim and Hawkins claim that their theories are elaborations of Christophersen's familiarity theory. Christophersen, however, explicitly denies the uniqueness approach to definiteness: " the does not mark a contrast; it does not select a single object from among several things contemplated." (1939:70). His con-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This, however, is the standard Russellian analysis of the sentences of the form (a N)VP, with the indefinite article in place of the definite. But this conse quence is clearly unacceptable. With such nouns, the indefinite article is not only not equivalent to the definite article but unacceptable- another serious linguistic inadequacy of the existence-plus-uniqueness account, cf. (19):
291
3. PRELIMINARIES TO A SYSTEMATIC CLASSIFICATION OF DO USES: A SEMANTIC SUBCATEGORIZATION OF NOUNS AND CONCEPTS
Those studies that try to cover a broader range of DO uses - Hawkins (1978) and Heim ( 1 982) - regard the anaphoric (and deictic) uses of DDs as the paradigm cases. Other uses they try to cover by some sort of generalization of the anaphoric case. But there remain a variety of uses which evade such a treatment and cannot be handled satisfactorily in either approach. Hawkins' Location Theory fails to apply to such simple sentences as Close the window (cf. Chr. Lyons 1980), and he is unable to formulate in a precise way conditions under which associative anaphoric and similar uses are possible. Heim would need essential amendments to her theory for all but the anapho ric uses. If other uses were incorporated not much would be left of the original idea of File Change Semantics. I believe that the path taken by Hawkins and Heim - i.e. starting from the anaphoric and deictic uses of DDs as the basic pattern - is a dead-end. Instead, I shall take, as it were, the opposite starting point: those cases where the ? efinite article is necessary for semantic reasons.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ception of familiarity is different. For him, the definite article associates the referent with previous experience. "By the unambiguous reference to outside knowledge (not contained in the idea of the word itself), the word is felt to stand for one definite individual ..." (1939:73). It is thus not uniqueness, but non-ambiguity8 which is essential for definiteness. Non-ambiguity is the property of an expression that allows for only one interpretation (possibly under additional constraints). Uniqueness of reference is always an acciden tal property of a sortal concept,- the property that it happens to apply to only one object. Non-ambiguity, in contrast, may be an inherent property of (also non-sortal) concepts. Imagine a group of men and women. The concept "oldest woman among them" is inherently non-ambiguous; due to the super lative there cannot be more than one objects to which this concept applies. Under the same circumstances, certain sortal concepts, e.g. "woman with white hair" or "woman older than eighty", may allow a unique reference to the same person, but it is then merely accidental that there is only one person to which these descriptions apply. I agree with Christophersen that the crucial feature of definiteness is non-ambiguity of reference. Situation Semantics has recently presented an idea which points into the same direction. Barwise and Perry (1983: 152ff.) regard definite NPs as func tions from situations to objects. In the remainder of the paper I shall develop a similar line of reasoning, arguing that definites are unambiguous functional expressions, assigning the head noun a quite different conceptual role than that of a sortal concept. The properties of existence and non-ambiguity are already explained by the logical status of definites as terms. If terms refer at all, they refer to something and they refer unambiguously. But I shall formulate a more substantial theory of definites below.
292 It is these uses which should provide the clue to the semantics of the definite article. For a proper definition of these cases and a systematic classification of DO uses we first need an appropriate semantic subcategorization of nouns. 3. 1. Sortal and Relational Nouns
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It is usually assumed that nouns can occur with any determiners, in particular with the definite article as well as with the indefinite (or, equivalently, without a ny article in case of mass and plural count nouns). There are, however, semantic subclasses of nouns that differ in the range of determiners with which they combine in certain contexts. Some nouns, in a sense, require the definite article. Also, some nouns, such as father, occur with possessive complementations while others, e.g. man, do not. In main stream modern linguistics the category N of nouns is usually not subdivided. In the work of more traditional grammarians, however, we can find such distinctions; Cf. e.g. Behaghel (1923, I: 22ff.) who distinguishes between "absolute Begriffe" and "relative Begriffe" (absolute and relative concepts). It is an open question whether the semantic subcategorization I have in mind requires a syntactic differentiation. I shall first introduce the subcategorizati on as one of lexical noun meanings, just in order to illustrate the difference by means of typical cases. Later it will become clear that the distinctions apply to uses or occurrences of nouns rather than to nouns as such. Nouns have two basic interpretations. Taken in isolation they can be considered either sortal nouns or relational nouns. Sortal nouns classify objects, whereas relational nouns describe objects as standing in a certain relation to others. These are two fundamentally different ways to character ize objects, and one cannot be reduced to the other. Let me illust rate the difference with something like a minimal pair: woman and wife. Woman is a sortal concept. It is the conjunction of other sortal concepts such as human, female, adult. Logically, sortal concepts are just one-place predicates. They can be negated, and some sortal nouns have lexicalized negations within a certain range of objects (cf. the complementary character of man and woman within the range of human adults). Wife, on the other hand, is a relational concept. Although it contains also sortal specifica tions ("human" and '.'female") the essential meaning is being the female spouse of someone. A "wife" is always the wife of someone (although in certain contexts the husband may become the victim of abstraction). It is the relationship of being married to a man which determines the referent of wife. This relationship cannot be replaced by a conjunction of one-place predica tes. Even the complex sortal noun married woman is not synonymous with wife, cf. John's married woman as opposed to John's wife. If one abstracts from the relationship and regards the sortal content of a noun only, sortal and relational nouns can become, in a sense, coextensional. Thus every woman (in the more general sense of female human) is a daughter, and every daughter
293
3. 2. Subtypes of Relational Nouns The type of relational nouns can be further subdivided into several subtypes in a way that will be relevant for the following discussion. The most impor tant subclass is that of functional nouns. For functional nouns, the relation that defines their reference is a function. Functions relate objects unambigu ously (or one-to-one) to others. Hence, functional nouns are inherently unambiguous, and this semantic property matters in the present discussion. Functional nouns (if they refer at all) always identify a referent. Sortal nouns, in c ontrast, only classify their referents. Under certain circumstance s it may happen that there is exactly one object which fits the classification. But this would be accidental. Functional concepts, e .g. mother of John , do not allow for more than one referent. I assume that the functions underlying functional nouns are in general partial functions. But for the sake of simplicity I will not make this explicit in each case. Functions assign values to arguments, unambiguously. What, then, are the arguments of the functions underlying the meaning of functional nouns? Consider a functional noun such as weight. Weight is clearly a relational noun. lt is a certain measure, related in a certain way to physical objects. And since every object has only one weight at a time, weight is a functional noun. Now, weight is always the weight of some object. Hence some object is involved as an argument: weight is a function from objects to values on a certain scale. But the weight of an object is again variable. It depends on time, location, and circumstances. My weight today is different from my weight tomorrow, it is different on earth and on the moon, and it would be different if I ate more. In possible world semantics one would say that the referent of a functional noun also depends on a possible world index. I prefer to talk of situations instead. In view of the fact, that the referent (or value) of a
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
is a woman, so daughter and woman (as sortal concepts) have the same extensions - but they have of course quite different meanings. Relational nouns are logically not one-place predicates but correspond to predicates with two or more arguments (i.e. to relations). Many nouns are ambivalent in that they can be used as sortal nouns and as relational nouns. An example may illustrate this point. When I enter a furniture shop and ask for "a table" I use table as a sortal noun. As such it contains certain conventional characteristics concerning size, shape, height, and proportions (among others), that distinguish tables from beds, trunks or an orange box. But if John invites Mary to his sort of improvised room, it may well happen that he points to an orange box and tells Mary: "This is my table. Please, take a seat". In this case he is using the noun as a relational noun. The object in question exhibits the features of a typical table only to a very modest degree, but what he means is that it plays the role a table plays, it functions as a table.
294 functional noun depends, in general, on the situation referred to, I regard all functional nouns as involving a situation as one of their arguments. Different functional nouns may relate to different components of the situation, and for special purposes it is certainly important to distinguish subclasses of functional concepts, e.g. those with time-dependence and those without. Others, in particular mathematical functional nouns, may appear to be completely independent on the situation, but since nothing is really absolute it is better to provide for a situational argument in case it may be needed. The situational argument of functional nouns plays a crucial role in certain types of intensionality, involving alternatives of the actual value of a functional noun.9 Consider sentences such as Her weight has changed.
(2)
Guess her weight.
(3)
Taken on the moon, her weight would be ideal.
As the examples illustrate, the situational argument is usually not explicit. It may, of course, appear explicitly if necessary. There are a n umber of functi onal nouns which involve only a situational argument. Weather is one, time another (in uses like what time is it), sun, moon, earth are used in this way, and in theoretical contexts also speaker, hearer, etc. Most functional nouns, however, involve a second argument. I already mentioned wife, which can be considered representative for all those functional nouns that refer to certain social or other roles, cf. president, referee, driver, bride, victim, murderer. the probably most perspicuous cases are nouns referring to unique parts of objects: head, body, roof, top, surface, etc. Other nouns of this type are those that represent conceptual dimensions of various kinds, such as height, age, weight, status, profession, nationality, character, name, address, telephone number (cf. Lobner 1979: l l l ff., 173 and LObner 1981). Or, to cite some linguistic terms: subject, intonation, interpretation, meaning, category (of something) etc. Yet a different kind of functional nouns assign to objects certain events which occur only once (cf. the discussion of "once only-events" in Galton 1984:56ff.): birth, death, beginning, end. Functional nouns are ubiquitous both in everyday speech and in scientific terminology. Some but not very man y functional nouns are based on two further arguments in addition to the situation; I am thinking of examples such as relationship, difference, distance. It should be mentioned that relational and functional nouns are not necessarily count nouns. Hair, blood, kinship, property, baggage (of someone) are functional mass nouns with an additional argument. A ir in the air was mild is used as a simple situational functional noun. Not all relational nouns, however, are functional nouns. Many relational nouns are "only" relational, i.e. they involve relations that are not one-to-one
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(I)
295 but (possibly) one-to-many. In what follows I use the term relational to refer to these relational nouns in the narrower sense. The most concrete examples are provided by kinship terms such as sister, daughter, aunt, cousin, grand mother. Others refer to social relations: friend, relative, neighbour, room-mate, compatriote, competitor, and the like. Hand, eye, part, member refer to parts which can occur more than once. All these examples involve an additional argument.
3.3. Functional, R elational, and Sortal Concep ts
Montague has introduced the notion "individual concept" (first in Monta gue 1970b: l 32; see also 1973:260). Individual concepts are functions from possible worlds to individuals. If one replaces possible worlds by situations, functional concepts of the simple type without additional arguments are individual concepts in that sense. But the converse is not true. "Individual concept" is a set-theoretic notion, and although it is meant to model intensi onal entities, it is an extensional conception. An individual concept is just a set of ordered pairs with the only restriction that there cannot be pairs <s,x> and <s,y> with different second members. This does not mean that they really represent a concept. The so-called functions of set theory are only potential extensions of functions in the intensional sense. The original idea of a function as developed by Frege (1891) is an (effective) assignment of values to arguments, following some general rule. The functions of set theory are the extensional shadows of such functions, the same way as sets themselves are the extensional shadows of sortal concepts. If we regard only such concepts as "reasonable" which can be explicitly defined with a finite amount of informa tion, then there are many more sets of ordered pairs than there are relational and functional concepts. Concepts in the sense relevant here are always concepts in the intensional sense. They can be understood as effective mental procedures with a certain input/output characteristic. The concept of "fat her", e.g., is a procedure that effectively assigns persons to persons given the relevant situational information. It may, in practice, be very difficult or even impossible to determine the father of some person. But such a situation
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Up to now I have spoken of different types of nouns. However, although there are centainly typical nouns that exemplify the distinctions in question, nouns, taken as lexical units, cannot be subcategorized in a strict way. The distinc tions rather apply to (types of) uses ofnouns. Henceforth, I shall therefore talk of functional, relational, and sort a) concep ts instead. Nouns can contribute to the meaning of a sentence as concepts of one of these kinds. Some nouns represent sortal concepts and others represent relational concepts. But relati onal meaning always involves also sortal specification, and accordingly many relational nouns can also be used as sortal concepts. On the other hand, sortal concepts can also turn into functional concepts, e.g. if in a special environ ment certain sortal characteristics become tantamount to a specific role.
296 would be due to an insufficient supply of relevant information rather than to a deficiency or indeterminacy of the concept "father". In the subsequent definitions I use the abbreviation EGP for "effective general procedure" in the following sense. A procedure is general if and only if it applies in the same way to different possible inputs; and it is effective if and only if it produces an output when provided with all the relevant input.
One-place relational concepts and sortal concepts may be difficult to distin guish. But this is not a central point in this connection. In what follows, I will often talk of functional and relational concepts as linking their value to the situation (and the argument) given.
3. 4. Noun Types and Concep t Types .
DDs with a functional head noun obviously yield functional concepts. Since DDs are terms, the basic interpretation will be a FCI. The arguments of a functional head noun can be made explicit in ways which differ for FC1, FC2, and FC3 nouns. The arguments of FC3 nouns (distance etc.) are added with prepositional phrases of the form between A and B. The non-situational argument of FC2 nouns is specified by means of various possessive construc tions: my father, John's address, the name of the town. The relation between the second argument and the value is generally treated as possession: a country has a government; a person has a father, a head, hands, a profession, etc.; a sentence has a meaning, a structure, a subject. Conversely, the value is said to belong to its "possessor". Situational a rguments take the form of prepositional, adverbial, or adjecti val attributes: the weather today, the present situation, the temperature in Bangkok. The argument place can also be occupied by a variable, either correspond i.ng to an indefinite the colour of one of your hats or introduced by a quantifier - everybody's mother. If the arguments are not explicitly specified, -
-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Definitions A one-place functional concep t (FC1) is an EG P which assigns objects to given situations. A two-place functional concep t (FC2) is an EGP which assigns objects to objects in a given situation. A one-place relational concep t (RC l ) is an EGP which in a given situation assigns to objects the value "true" iff a certain relationship obtains between the object and (elements of) the situation. A two-place relational concep t (RC2) is an EGP which in a given situation assigns to objects the value "true" iff a certain relationship obtains between that object and another obj ect in that situation. A sorta/ concep t (SC) is an EGP which in a given situation assigns to objects the value "true" iff that object in that situation exhibits a certain property.
297 they are assumed to be given in the context. Relational nouns, when used in the plural with the definite article, also give rise to functional concepts - cf. her arms, his parents, the Members ofParliament. They assign to the addition al object the totality of the objects which stand in the relevant relationship to it. Often relational nouns can be disambiguated to yield a functional concept by certain attributes. Cf. the left hand, the Foreign Ministery, my eldest son. Since functional nouns are inherently unambiguous, they do not allow the indefinite article as long as it can be presupposed that they have a referent. Only in those contexts that focus on the very existence of a referent, the indefinite article or its negation no is natural with such nouns, cf. the following examples: Does a makak have a rain
(5)
This car has no clutch.
All determiners, with the exception of the definite article and the possessive pronouns, require the noun to be taken as a sortal or relational concept. They all involve the possibility of several objects of the same kind. Numerals and the indefinite article count objects of one kind; demonstratives select between objec;ts of the same kind, but in different regions; quantificational determin ers refer to a totality of objects of the same kind. Two-place functional nouns do, however, occur with other determiners. These cases can be explained by reference to more than one argument. If the additional argument of a two-place functional noun is implicitly or explicitly existentially bound (or left open), the result is a sortal concept. This sort of existential closure is present when a functional concept is combined with an inherently comparative adjective (all scalar adjectives are of this sort). (6)
She is a good mother.
It also underlies statements such as (7)
He doesn't like mothers.
and is explicit in (8)
The mothers of some of the children didn't like the proposal.
What appears to be quantification with functional nouns, in sentences such as (9)
Every mother loves her child.
is in fact quantification over implicit arguments of the functional noun. (9) means the same as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(4)
298 (10)
Every child's mother loves him.
Likewise, plural is possible with functional nouns, if there is a plurality of arguments involved either explicitly (cf. 8) or implicitly (as in 7)). 3. 5. Semantic and other Functional Concepts.
4. USES OF DEFINITES. SEMANTIC AND PRAGMATIC DEFINITENESS
After the introduction of functional concepts, it is now possible to discuss the various uses of definite NPs in a more systematic manner. I will focus the discussion on definite descriptions but include the other types of definites as well, since they fit in naturally. Roughly speaking, there are two kinds of uses for definites. In those cases which I want to call "semantic definites" the referent of the definite is established independently of the immediate �ituati on or context of utterance. It is the semantic definites which evade theories such as Heim's (1982) and which fit most easily into the theory to be developed here. "Pragmatically definite" NPs, on the other hand, are essenti ally dependent on special situations and contexts for the non-ambiguity (and existence) of a referent. They include deictic, anaphoric, and endophoric uses of definites. (Endophoric DDs, sometimes called "cataphoric", carry with them an attribute that Jinks their referent explicitly to the situation referred
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Functional concepts do not necessarily owe their non-ambiguity merely to semantic constraints. The concept of "mother" might be considered functional on semantic grounds, if it is taken to mean the woman who gave birth to a person (but in vitro fertilization has already complicated this matter.). It is impossible to conceive of a person as having more than one mother unless the term mother is taken in different senses, corresponding to different functional concepts. The notion wife, however, is a case of a functional concept which is one-to-one only under an additional (non-se mantic) monogamy constraint. Whether in a certain language community a noun can constitute a functional concept, in general depends on all sorts of constraints, some being semantic and others belonging to other realms of life. We shall see in the next chapter that there is a whole scale between logical necessity and near-accidentality. Let me conclude this section with the remark that functional concepts are not just one variety of noun interpretation. They are in fact ubiquitous in natural language semantics. Janssen (1984) has pointed out a considerable range of phenomena in the area of NP semantics where individual concepts (in fact, I want to add, functional concepts) are involved. But functional concepts also underlie several other phenomena, including wh-questions, focusing and thematic roles. I will come back to the latter, when I discuss the anaphoric use of DDs.
299 to.) Semantic defi nites refer u nambiguously due to general constraints; prag matic definites depend on the part icular situation for unam biguous referen ce. I define: An NP is a semantic definite iff it represents a functional concept, indepen dently of the particular situation referred to.
4. 1. Semantic Definites (1): FCIs
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Proper Names and Similar Expressions The clearest examples of semantic definites are proper names. Within a certain range of situations, proper names refer unambiguously to certain objects. They constitute constant functional concepts, as their value does not vary with their possible argu ments. Equal names for different objects can be considered different partial functions with different domains of arguments. Some names have a broad domain of situations in which they refe r to t he same object, e.g. geographical names or names of abstract entities such as dates or numbers. Personal names have a more limited domain of use, in particular bare first names. But they do not rely on the part icular situation alone for reference. Personal names in English have not the form of a D O . But in colloquial German, we u s e personal names with t h e definite article throughout. Note that in English artificial names such as "the Empire State Bu ilding" or "the London Symphony Orchestra" have the definite article. Very close to proper names are those DDs that consist of the definite art icle, a sortal noun , and a proper name of some sort, cf. the year 1984, the word ' the', the People's Republic ofChina, the number zero, the opera Rigoletto . They form FC I s in the same way as proper names do thanks to the proper name they contain . Phrases involving functional and relational specificatory nouns, such as president Marcos, ourfriend Ferdinand, his wife Imelda have a similar status. Although the additions are often disambiguating, they do not establish the FC I involved. Hawkins ( 1 978: 1 40ff. ) mentions another group of cases which can also be subsumed under this category. They consist of an abstract sortal head noun and a subordinate clause which explicitly specifies a certain abstract entity of the sort indicated. Cf. the rumour that Reagan is going to resign, the question whether the indefinite article is a numeral, the idea to have pizza now, the fact that she was married, the dream to become rich . There is no semantic difference between proper names and denotations of certain abstract entities, such as the definite article or the past tense. They, too, are constant FC ! s . Sometimes, due t o special circu mstances, D D s with sortal head nouns acquire the status of a proper name at a certain place. Hawkins ( 1 978: 1 1 8ff.) calls this use "larger situational use based on specific knowledge", which is probably not a very fortunate term. This use is situational only in the sense that the domain of FC Is of this kind is locally restricted. Examples would be the Little Mermaid in Copenhagen, the Wall in Berlin, or the Tower in London. They can be considered a special sort of proper names, too.
300
lndexicals A more interesting case of FC i s is provided by first and second person singular personal pronouns. They are (no n-constant) FC Is restricted to situations of utterance. I assigns the uttere r,you the addressee to the situation of utterance. I take I and you to be semantic definites, as there are effective general p rocedures to determine the utterer and the addressee of an utteran ce. The status of we and you (plural) is less clear. What is included in the refe rent of we and you in addition to the utterer or addressee respectively may depend on the particular situation. Probably the best way to look at we and you p1 is to consider them RC2 concepts with the implicit arguments I and you511 , respectively.
{
in England of England
(I)
the weather
(2)
the pnme mm1ster
'r!
. .. {
}
'r! in B ritain . . o f 8 n t am
}
The argument of concepts such as prime minister is not a location or territory but a state, i.e. a more abstract kind of thing: Better candidates for FC I s with a l ocal domain are public institutions which fulfil a specific function for a particular district: the post office, the station, the pub, the laundry. There are general constrai nts which allow speakers to presuppose the existence and u n iqueness of a referent in an arbitrary location. The constraints may be of limited validity though, and the regions may be li mited too. Within a normal
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Simple FC 1 DDs Among DDs in particular, the basic representatives of semantic definites are FC I s . There are first of all DDs with a simple FC I head noun that depend on immediate constituents, com mon to all situations, e .g. to the temporal and spatial location. Such concepts are rather rare, cf. the weather, the time, the air, the moon. There are quite a few examples which seem to be of this kind, but are better considered FC2 concepts with an implicit argument. Hawkins ( 1 978: 1 1 8ff.) calls them "larger situational use based on general knowledge". Examples are DDs such as the Prime Minister. The relevant situations would then be located within the territory of a state which has a prime minister. I n Lobner ( 1 979) I took this view myself b u t n o w I prefer the FC2 interpretati on . There is a linguistic argument for this view: nouns such as weather have their (only) situational argu ment specified in a different way than the argu ment of prime minister.
30 1 size apartment there is normatiy one toilet, one kitchen, one l i ving room etc., and it is due to that general constraint that the toilet, the kitchen, the living room can be used as functional concepts. Crucial for the FC character of a DO, and hence for the possibility to use it without former explicit introducti on of a referent, is not the extension of the situation of reference but the existence of some general constraint due to which the denotated object has a certain function in that environment . Hawkins ( 1 978: 1 1 9) describes the condition that the DO has to provide an FC 1 on the basis of general constraints only very vaguely when he writes: "What seems to be required is only that situations of various kinds, e.g. weddings, villages, countries, etc. generally contain certain objects." This formulation is i nsufficient in that it does not requi re a unique role for the respective object.
(3)
the { next I last I third I most success ful} president of the association
Only is similar i n the sense that it involves the possibility of other referents of the same kind. I f superlatives or last are combined with plural nouns as in (4 )
the fattest wrestlers
the outcome is usually vague. Th is does not mean, however, that DDs of this type are ambiguous, but that their respective reference depends on a contex tually determ ined relation. Another case which belongs here is N Ps of the form the other+N . The adjective other adds a relational meaning to the objects that fall under N: to be different from some other "N(s)" given i n the context. The result is a fu nctional concept : "those N(s) other than xy". 4. 2.
Semantic Definites (2): FC 2s
FC2s provide one-to-one (i.e. unambiguous, functional) links from argu ments to their values. The argument, in turn, will be linked to the situation referred to in one of several possible ways. Let S be the situation referred to, A the argument, and V the value. The basic pict ure is that of a chain of links:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Complex FC J DDs Certain adjectival attributes combine with sortal or relational nouns to form com plex FC 1 expression. A mong these are superlatives and ordinals as well as next, last, only, same. With singular nouns, superlative and ordinals are obviously functions that single out a certain member from a collection referred to implicitly. They require the possibility of alternatives and hence relational or sortal nouns. When applied to functional nouns, they force implicit reference to diffe rent arguments with different function values. Cf.
302 (S)
0�------101---·0 FC2
(6)
the mayor of a small town in Wales
(7)
the mother of one of my students
In (6) there is a chain of refe rence involving an indefinite argu ment : (6' )
---4 \-----·0 @-�------0 (indefin ite)
FC2
Hence, the overall reference of the NP (6) is not determined. Yet the definite article is used here as well as in (7), which is referentially more complex :
(7') my students
one of my students
the mother
Quantificational argu ments, too, lead to a disru ption of the referential chain by underdetermined arbitrary links. Cf. (8)
Every man loves his mother.
The FC2 NP his mother contains (logically) a bound variable argu ment.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Now I would claim that definiteness of an N P involving a FC2 noun does not concern the nature of the overall link between S and V, but only the nature of the link between A and V, i.e. the link immediately provided by the functional noun. The usual conception of definiteness is such that the refe rent of a definite expression must be determi ned in an absolute sense. Heim's ( 1 982) and Ha wki1_1s' ( 1 978) theories are implicitly committed to this notion of definite ness, which can be found explicitly in Lyons ( 1 977: 1 78ff.). Under this per spective, definiteness is considered a property of reference . This, however, applies only to the simple cases, as provided by the F C I defi nites just discussed, and the pragmatic defin ites yet to follow. In the simple cases the referent of the definite is directly linked to a situation, hence the natu re of reference coincides with the nature of the last link in the chain. As already mentioned in the previous section, however, the argument of an FC2 noun can be non-defi n ite. Consider NPs such as
303 Definiteness - as fa r as it is indicated by the use of the definite article (or possessive constructions) - is not a matter of overall reference, but only a matter of a link which is provided by the noun fol lowing the article immedi · ately. If the link provided by that noun is unambiguous, then the definite article will be used. The arti�le is, as it were, only responsible for the link it is i mmediately concerned with. Accordingly we need two diffe rent conceptions, one for the defi niteness of a noun as expressed by the use of the definite article and another for the determinedness of the referent of the overall N P. 10
(9)
my wife
( 1 0)
the President of the U.S.
(I I)
the meaning of the definite article
( 1 2)
the second wife of the former husband of my father's lover
Among those definites which are definite "through and through" , there are of course all sorts of possible links, including deictic and anaphoric links . . Hawkins ( 1 978) covers only reference chains with no more than two links.
FC2s with implicit arguments In many cases, the argument of an FC2 is left implicit. It is then, in general, either a quantified variable or p ragmatically definite. H awkins ( 1 978) men tions t wo uses of definites which belong to this category. One is what h e calls the "introductory situational use", and the other is the so-called "associative anaphoric use". The former use is exemplified by sentences such as ( 1 3)
This is the clutch.
which is a case of a FC2 noun (clutch) with an argu ment (a car) that is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
FC2s with explicit arguments In what fol lows, I will restrict myself to FC2 N Ps with definite arguments. As I already mentioned, the proper way of connecting an argument to an FC2 noun is a possessive construction. Possession is also the basic conceptual scheme in which the association of a function value with its argument is expressed i n statements such as: everybody has {a head/ a father/ a name/ a weight} , etc. For personal pronouns there is the special possibility of corresponding possessive pronouns; some NPs can be preposed in the genitive; else the definite article precedes a complex consisting of the FC2 noun and a preposi tional phrase of the form of+ NP. If the argument NPs are all semantically definite then we get definites that are capable of first mention use ( Hawkins 1 978: 1 38 ff. ):
304 provided by the immediate physical environment. In that sense it has an implicit deictic argu ment. The introductory character is not essential to this kind of use, which would also be instantiated by an utterance of ( 1 4)
Step on the clutch!
Introductory statements such as ( 1 3) are, however, interesting in that they could not be used as such if the noun were not a functional noun in its own right, i.e. if it would not represent a general functional concept. A further example of implicitly deictic FC2 definites is Hawkins' "larger situation use based on general knowledge" , present in sentences such as ( 1 5). The Pri me Minister has resigned.
In these cases the location of the utterance belongs to the territory of a state to which in turn the FC2 "Prime Min ister" is assigned. (Of course, this is only one possible interpretation. The i mplicit argument can also be determined otherwise.) Very frequently, texts contain anaphoric expressions which refer indi rectly to objects (or to be precise, discourse referents) introduced earlier. Standard examples are sentences such as Hawkins' ( 1 6)
Fred discussed a book in his class yesterday. He knows the author.
These are cases of FC2s with an implicit anaphoric argument. Neither Hawkins ( 1 978: 1 23ff.) nor Heim ( 1 982:372ff.) manage to formulate the proper conditions under which this use (and FC2s with implicit arguments in general) is possible. It is not a matter of sufficiently close relations that occur in a sufficient frequency, as H awkins assumes. This would not guarantee an una mbiguous link. On the other hand, "sufficient frequency" is not essential either. Some functional concepts can apply to very few arguments and also the relation may be qu ite indirect. The crucial condition is that the head noun in these uses provides a general two-place fu nctional concept for which there is an appropriate argument in the immediate linguistic context.
4. 3. Semantic Definites (3): Configurational Uses There is one use of definites which is very interesting in this connection , because it provides an insurmountable obstacle to all theories of definiteness based on sortal u niqueness. I am thinking of sentences such as ( 1 7)
He was the son of a poor farmer.
( 1 8)
I do not want my daughter to marry the lover of her sister!
( 19)
He put his hand on her knee_ l 1
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 5)
305 ( 1 7) does not imply that "he" was the only son of his father. ( 1 8) impl ies that the speaker has at least two daughters and does not exclude the possibility that the "daughter" referred to has more than one sister and that the "sister" referred to has more than one lover. ( 1 9), finally, does not imply that "he" puts.his only hand on her only knee. Of course, such examples can also involve real functional concepts. Cf. (20)
He was the husband of late Alma Mahler.
(2 1 )
I do not want m y mother t o marry the father of m y wife.
(22)
He skratched her back.
(23)
I do not want my daughter to marry the lover of the foreigner.
The examples cited here differ from the ones discussed above in that the definite N Ps in these sentences are not really referential, except for the respective subject N Ps. Th is is more obvious in the first examples ( 1 7)-( 1 9), which involve relational concepts (son, daughter, lover, sister, hand, knee). If they were referential, these concepts would not be unambiguous in the situation at hand. I nstead of referring directly to a particular situation, such statements are, in a sense, generic. They refer to an abstract situation (in the sense of Barwise and Perry ( 1 983) in which only those objects figure that are explicitly mentioned. Within that abstract situation and, as we shall see, relative to the way the situation is built up, the concepts used are unambigu ous. As abstract situations consist essentially of objects and certain relations between them, I shall refer to them as "configurations". The configuration referred to in ( 1 7) consists of just two objects, the father and the son : (24)
son
o--o
"a poor farmer"
"he"
No other objects, in particular no further sons of that farmer figure in that constellation. Hence the relational concept "son" in fact provides afunction al link from the father to the son. Similarly, in ( 1 9), fou r objects constitute the configuration expressed:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The nouns involved are necessarily either relational or functional. Statements of this kind are impossible with sortal nouns. Cf. the inappropriateness of (23) in situations where there is no specific referent for the foreigner:
�
306 (25)
"he" hand
.......
�
-.....
"his hand"
�
"she" k nee
put on
"her knee"
Again, the relational concepts "hand" and "knee" are used unambiguously relative to this configuration. The configuration referred to in t he embedded construction in ( 1 8) is slightly more complex:
(26)
''I''
t
This configuration contains t wo daughters, b and c, of the person a. But only b figures as a's daughter w hile c figures as b's sister. The respective links between the constituents of the configuration are an essential part of it. We now understand why sortal concepts cannot play the role played by relational or functional concepts in this kind of use: they would not provide a link between different objects. Sentences involving such abstract situations con vey the notion that the real situation referred to is of that type, in the sense of Barwise and Perry ( 1 983). They are true whenever the abstract situation i nvolved can be anchored in the real situation. In the case of( 1 8) and (26) the actual situation may well contain even more daughters of the spea ker and more lovers of his daughters, and in this case ( 1 8) or (26) would apply to any sub-constellation of this type. Sentences ( 1 7) and ( 1 9) are diffe rent from ( 1 8) in this respect: there is in any case only one subsituation of the actual situation of the type (24). In that sense, ( 1 8) is more general or generic than ( 1 7) and ( 1 9). I t is worthwhile to note that the relationship between the abstract situation which is expressed and the actual situation, i .e. the kind of reference involved , i s indefinite: there is a subsituation of that abstract kind contained in the actual situat ion. Indefinites, both in Situation Semantics and in Discourse Representation Theory, give rise to free variables (i.e. "indeterminates" in the terminology of Barwise and Perry 1 983, or "discourse referents" in the sense of Heim 1 982). The configurational definites in the examples discussed also introduce free variables into an abstract situation. Both the uses of indefini tes and of configurational definites generate in a first step an abstract situation (roughly equivalen t to the discourse representation in Heim ( I 982) and Kamp ( 1 98 1 ) which then m ust be anchored in one of several possible ways in the actual situation referred to.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
"my dau hter" "her sister" a )-----___,o-{ b t------+f daughter sister lover marry ---"the lover of her sister"
307 Again, we see that the definite article does not express definite reference but only the non-ambiguity of the link provided by the following noun. The same kind of indirect reference via an abstract situation is present in sentences such as (27)
He was taken to (the) hospital.
I n many such cases the definite article is omitted in English but would be used in German: E: She goes to church. G: Sie geht in die Kirche.
(29)
E: He came after lunch. G : Er � am nach dem Mittagessen.
In such cases one does not refer to any particular object but rather to an abstract situation, in which a hospital, a schoo l , a lunch is an essential constituent. Further examples would be (30)
He is my friend.
(3 1 )
Switch the light on.
(32)
He broke his leg.
all of which withstand a plain referential interpretation. 4. 4.
Pragmatic Dejinites
In the cases treated so far, the head noun was always a functional concept , either by virtue of i t s lexical meaning plus, occasionally, additional general constraints, or due to the restriction to limited abstract configurations within which relational concepts can be used unambiguously. Pragmatic definites, in contrast, have sortal or non-functional relational head nouns. They depend on the particular situation for unambiguous reference. I shall distinguish three uses: endophoric, anaphoric, and deictic definites.
Endophoric Dejinites By endophoric definites I mean noun phrases with relational or sortal, i.e. potentially ambiguous, head nouns which carry a disambiguating attribute with the m . They are someti mes called cataphoric. The attribute links the refe rent of the head noun unambiguously to other objects and may be a relative clause or a prepositional or adverbial phrase. One relevant case are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(28)
308 N Ps with "establishing relative clauses" (Hawkins l 978: l 3 l ff.), such as the italicized N P in the following example, taken from H awkins: (33 )
What's wrong with B ill? - Oh, the wor, ·an he went out with last night was nasty to him.
(34)
What's wrong with B il l? - Oh, he went out with a woman last night and the woman was nasty to him.
Furthermore, in those cases where semantic and pragmatic definiteness is overtly distinguished, endophoric defi nites group together with anaphoric definites. Not every relative clause is an establishing relative clause. The crucial condition is that it must provide a functional, i.e. one-to-one,linkofthe referent to other given material. This material can be deictically or anaphorically linked or a bound variable. For the latter case cf. examples such as (35)
Each man has the wife he deserves.
A deictic e ndophoric DO occurs in the sentence (36)
Who was the person that called you up this morning?
In a novel of Chandler I found an endophoric DO with a link to an anaphoric definite: (37)
Her name, according to theplaque on her desk was Adrienne Fromsett.
When this sentence occurs, the desk and a referent for her have been explicitly introduced before, but no plaque was mentioned. In this case, the attribute on
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The phrase woman X went out with last night is a RC2 linking women to X, and i f it is further presupposed that X (in this case, Bill) did not enjoy the company of more than one lady, it serves in fact as an FC2. This FC2 links the referent of the DO to "Bill". I f one disregards the inner structure of such DDs - the sortal head plus the establishing relative clause which provides a functional link - one could si mply regard them as FC l s, and hence as semantic definites. The linguistic evidence, however, favours a classification as pragmatic definites . First , we have seen that the definite article only concerns the definiteness of the i m mediately following noun . And, second, the kind of definiteness (or refe rence) involved here is apparently no different from the following case, where the relative clause has been turned i nto a preceding main clause and the woman is j ust anaphoric.
309 her desk not only locates the referent of the plaque i n a limited area. It also provides a functional link in that context. The desk belongs to a certain person, and the plaque indicating the user of the desk belongs to that desk , in turn.
(38)
The woman that squints has found a husband after all.
(39)
Yesterday I met the man who studies buddh ism.
Such DDs, when taken as a whole, represent j ust sortal concepts and require previous mention of an object which fits the whole description. Note the possibility of demonstratives here in contrast to nouns with an establishing relative clause. The relative clauses here clearly do not provide a functional link for the referent of the head noun, but just additional sortal information. I shall argue below that anaph<;>ric N Ps constitute a semantically heterogene ous class. Demonstrative NPs function differently from definite N Ps. (Hence the difference in distri bution. Demonstrative defi nites cannot occur in all the uses of definites mentioned before, including the endophoric use).
Deictic Definites Deictic definites, which comprise DDs and demonstratives with sortal or relational head nouns, differ from anaphoric defin ites in that they refer immediately to components of the situation of utterance. Hawkins ( 1 978: l l Off.) distinguishes two "deictic" uses: what he calls "visible situation use" and so-called "im mediate situation use". These two uses are quite different. As examples of the second type, he cites sentences such as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Anaphoric Definites It is commonly accepted that there are direct and indirect anaphors. D irect anaphors have an antecedent with the same head noun and an indefinite article, if they are DDs or demonstrative definites. If. the anaphoric expressi on is a pronoun, it agrees with the antecedent in n u mber and gender. Hawkins ( 1 978: 1 07) cites instances of more indirect kinds of anaphora, including nominalizations and paraphrases. I shall restrict myself to the straightforward cases (with the exception of the yet different "associative anaphora" involving FC2s, discussed above). The theoretical difficulty with anaphoric definites is the fact that they have sortal or relational head nouns which are not inherently u nambigous, but refer back to an object previously introduced. In the next section I will offer a theoretical explanation of my own. Suffice it here to say that the subclass of anaphoric defi nites has a unique composition. It includes D Ds with sortal or relational (but not functional) head nouns, demonstrative noun phrases with sortal or relational head nouns, demonstrative pronouns, and third person personal pronouns. A mong anaphoric defi n ites there are also cases with a non-establishing relative clause, cf.
310 (40)
Beware of the dog!
(4 1 )
Harry, m ind t he table !
(42)
John, would you pass me the water, please?
John, still half asleep , reaches for the bottle without even trying to look at it a nd passes i t to Mary. He just knows where the bottle is. This use of the water is an FC l use. Although the range of situations is very restricted, there is a general constraint which establishes the FC I "Mary's and John's bedroom bottle". It is not a particular bottle containing a particular volume of water. Rather the bottle will be replaced by another one when it is empty. But how if John and Mary do not have this custo m , but just happen t o have a bottle of water beside the bed? The utterance of (42) could be equally successful. Neither Mary nor John might be able to perceive the bottle, they just know that it is there, in this particu lar situatio n . Is this use deictic? It is certainly not demonstrative in any sense, for sure. I would tend to regard it as deictic. One could t hen define deictic uses as those which employ accidental information about the particular situation of utterance. A clear case of deictic defi nites is the second use mentioned by H awkins, his "visible situation use". This use is possible if one and only one object of the kind belongs to the situation of utterance, if that object is visible (at least for the addressee) and known to him as such. As an example, Hawkins cites (43)
Pass me the bucke t !
The pragmatic description is s o far correct, I assume. B u t I should not like to restrict the use to visible cases alone. Other perceptions are possible as well: (44)
Where does the terrible smell come from?
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In both cases the referent is s upposed to be invisible, and the existence of a referent is brought to the mind of the adressee by the use of the DO (which i m plies its existence). These uses, however, in fact involve functional con cepts: the dog that belongs to the house, the table that always stands where it stands. They are deictic in that they refer to the situation of utterance for the argu ment of the FC. But being FCs, the D Ds in (40) and (4 1 ) are semantic definites and hence do not belong here. Note also that the definite article in (40) and (4 1 ) cannot be replaced by a demonstrative. It is not easy to draw the line between FC I s and deictic defi nites. Consider the following example. John and Mary (believe i t or not ) always have a bottle of mineral water beside their common bed. One night, John is already sleeping. Mary wakes up and feels thirsty. She fumbles for the bottle in t he darkness, but cannot find it and wakes up poor John :
311 (45)
Can't you stop the noise?
Other definites which can be used deictically are demonstrative pro-NPs and N Ps with a demonstrative determi ner. I n English also third person personal pronouns occur in deictic use, in contrast to German. In a situation where John and Mary are sitting in their garden and notice an u n known woman entering their property, Mary might say (46)
What's she doing here?
but in German (at least in my dialect) I would use a demonstrative pro-NP: Was macht die denn hier?
4.5. Remarks about the Significance of the Distinction between Semantic and Pragmatic Definites The basic distinction drawn above between semantic and pragmatic definites is significant in several ways. 1 . We have seen that, because of their inherent non-ambiguity, functional nouns allow only for the definite article, except of existential contexts. The article is thus redundant in connection with functional nouns, at least if the functionality of the noun is based on its lexical meaning. The redundance of the definite article in semantic definites offers an explanation for the absence of any article in certain cases of definites. One general example are personal and other proper names. But there are also cases involving count nouns used as definites without an article. English as well as German are languages which only gradually aquired articles, both definite and indefinite. The present state, characterized by near-obligatoriness of articles with singular count nouns is the resuli of a steadily growing range of application of the defini te (and the indefinite) article. A last set of definites that occur without an article is a set of some functional NPs of general use, such as God, and especially many more or less idiomatic configurational uses like go to school, loose patience. face toface (cf. Jespersen 1 933: 1 67ff.). 2. In German, cliticization of the article to a preposition as a host (the contracted forms mentioned above) is possible if and only if the NP is semantically definite and not too complex, as has been pointed out by Haberland ( 1 985) and Hartmann ( 1 980). Cf. examples such as (48)
Ich habe einen Brief vom Harald bekommen . [I got a letter from H arald.]
(49)
Er schreibt vom Nordpol. [He's writing from the North Pole.]
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(47)
312 (50)
am A bend, im Mai, zur Halbzeit [in the evening, in May, at half-time]
(5 1 )
Er hat beim Poker 2 Pfund veri oren [He lost 2 Pounds playing poker.]
Contracted forms are particularly frequent in configurational uses: Er mul3 {•in das/ins} Krankenhaus/ {•zu der/zur} Schute. [He has to go { to(*the) hospital/to (*the) school } .]
(53)
Sie will mit ihm {•zu dem/zum} Standesamt gehen. [She wants to go with him to the registrar's office.]
Characteristically, the contracted forms are not possible in the same phrases as soon as they are used to refer "pragmatically" in our sense to specific o bjects:
(54)
Er mull wieder {*ins/in das} Krankenhaus zuriick, aus dem er schon entlassen war. [He must go back {•to/to the} hospital from which he had already been discharged.]
3. In some German dialects in the Rhineland there are two different definite art icles (cf. Hart mann 1982). One can be characterized as weak. I t has a schwa in all its forms (e.g. (d ;l R]/(d;l ]/[;l t] for the three genders of nomi n ative singular). The other i s strong, with short tense vowels (the respective forms being [de]/[di], [dat]). Both forms are unstressed and cannot be considered demonstratives. The weak article is used exclusively in semantic defi nites, the strong one in pragmatic defi n ites. 4. Karen Ebert in her excellent study of the Frisian dialect Fering (used on the i slands of Fohr and A mrum off the German North Sea coast) also reports the existence of two different definite articles (cf. Ebert 1 970 and Keenan and Ebert 1 973). Again, one form , the " A-article" is phonetically weaker than the other, the " D-article" and the ranges of the respective articles coincide with the distinction between semantic (A-article) and pragmatic (D-article) defini teness. 4. 6. Referentially Quantifying NPs In Section 2.2. I claimed that refere ntial quantification involves definite reference to the domain of quantification. Quantification requ ires the possi bility of dividing the domain of quantification into a positive and a negative part with respect to the pred icate quantified. Explicit reference to the domain of qu antification can accordingly only be made with plural or mass definites.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(52)
313 A look at the different uses of definites shows that referentially quantifying N Ps indeed have the same range of uses. Cf. the semantically inherently definite quantifiers in (55)-(57): (55)
All the water is polluted.
(FC I )
(56)
Many members of the parliament voted against the law.
(FC2 + argument)
(57)
You had better use both hands.
(configurational)
The subject in (58) involves endophoric reference:
Some people who watched the last episode of "Schwarzwaldklinik" were shocked.
Sentence (59), when uttered in an appropriate situation, can be deictic: (59)
Look here, three eggs are broken.
For an anaphoric quantifier cf. the NP each lass in the following text, taken from a fairy tale: (60)
Once upon a time there l ived two lasses, who were sisters, and as they came from the fair they saw a right handsome young man standing at a house door before them . . . . . A nd he had a golden ball in each hand. He gave a ball to each lass, saying she was to keep it; . . .
The NP each hand is semantically definite, involving t h e FC2 hands with an implicit anaphoric argument "he".
5. THE MEANING OF THE DEFINITE ARTICLE
In the previous section we have seen that functional nouns require the definite article unless they occur as arguments of existential predicates. We have also seen that definiteness as it is expressed by the definite article is not a matter of the overall reference of the N P but only of the immed iate head. This became clear in view of FC2 definites with non-definite argu ments. (A further point in this connection is the fact that endophoric definites - which have sortal heads but functional concept meaning - require the "pragmatic" definite article in dialects with an overt distinction.) On the other hand, the defi nite article forces a functional reading onto relational nouns. This can be seen from the configurational uses as well as from simpler examples such as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(58)
314 (I)
I wanted t o buy that book. B u t I had forgotten the author.
5. 1. The Definite A rticle in Endophoric DDs12 A mong the pragmatically definite N Ps it is the endophoric DDS that provide the key to this class as a whole. Endophoric DDs consist of a sortal or relational head noun and additional attri butes which under the sortal or relational constraints that are provided by the head noun yield a functional concept. It is furthermore essential that the referent is linked to other objects; otherwise we get cases of semantic definites as discussed in 4. 1 . , with the characteristic "wea k" articles in those languages which distinguish between semantic and pragmatic defin i teness. Let us now have a closer look at the i n ner structu re of endophoric DDs. Hawkins in h is discussion of "establish ing relative clauses" (which are the crucial disambiguating attributes of endophoric D Ds), contrasts the following sentences, taken as answers to the question what's wrong with Biln
(2)
The woman he went out with last night was nasty to h i m .
(3)
The woman was nasty to h i m .
(4)
The woman who was from the South was nasty to h i m .
(5)
The woman who some sailor dated last night spread nasty rumours about Bill.
Among these sentences, only (2) has an endophoric subject. The addition he
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Books can have more than one author. H ence, the noun author is only a rela tional concept, as far as its lexical meaning is concerned. But the use of the author in ( I ) excludes the possibility of more than one author. If one looks at the cases of semantic definites only, one might thus want to say that the definite article is used if and only if the head noun is interpreted as a functional noun. But semantic definites always involve relational nouns. How about the sortal head nouns of pragmatic definites? It would be easy to sto p at this point and e mbrace the view that there are in fact two different defi nite articles with different meanings. But I think that we may go further and gain not only a uniform semantics for all uses of the definite article and of definites in general, but also a stronger theory of anaphora. My central claim is this: In all its uses, the definite article has the meaning ofindicating that the noun is to be taken as a functional concept. The fu nctional concept can be either independently established or situatio nal . I will spend the rest of this section to explain the notion of a "situational" functional concept.
315
(6)
Situation
GOBL
Bill/he
the woman, . . .
g f is the functional link linking Bill to the situation. g is the actor/action-link between Bill and GOBL and h is the comitative link from GOBL to Bill's company. The event GOBL is an important constituent of t he chain . It is introduced by the verb, and the links between it and the neighbour consti tuents are provided by the thematic roles of the verb. 5. 2. Digression: Thematic Roles are Functional Concepts In the example just considered the thematic roles of the verb go out with correspond to functional concepts. This is not an accidental property of this verb but a general structure: verb roles are functional concepts. This fact is reflected by the basic assumption of case grammar and related theories that within one sentence every thematic role can be specified at most once. (The fact that one verb complement may fill two or more thematic roles, does not contradict the functional nature of the thematic roles: two fu nctions may have the same value.)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
went out with last night can be used to establish a functional link from "Bill" to the referent of the woman, Bill himself being linked to the situation. Thus, the referent of the DO, too, is (indirectly) l in ked to the situation. I n (5), the definite NP Bill is replaced by an indefi nite NP. The chain is interrupted (or incomplete), "the woman" not l in ked to the situation. (3) is not endophoric because there is no disambiguating information at all. And (4) is si milar in that the whole information is sortal and does not link the referent of the woman unambiguously to anything. The relative clause can be taken as providing a relational link from the South to that woman, but this link is a one-to-many link. Now the only endophoric D O , the subject of (2), works as follows. First there is he, a definite pronoun with a pre-established (or presupposed) referential link to the situation. "He" is said to have gone out last n ight with somebody. The temporal specification "last night" is definite and it is restrictive enough to single out a single event of going out (one usually goes out with somebody only once a night). Hence there is an individual event GOBL of going out linked to Bill (the referent of he). Since Bill figures in a unique role of that individual event GOBL, there is a relational link between Bill and GOBL; in addition, it can be assumed that the link is the only one from Bill to an event of this kind, and hence, in fact, functional. Now to every event of the sort go-out-with-somebody there is one and only one object figuring in the object role (note, that the object may be complex'.). I n our case, the object is said to be a woman, hence a single person. Simpli fied, the chain of links between the situation and the referent of the subject looks as follows:
316
(7)
Bill is bold.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Many, but not all, verbs can be used to refer t o events, i.e. abstract entities located in time. For such verbs, the semantic roles correspond to functional concepts that assign to individual events of that kind the respective partici pants. For acts of the kind "give", e.g., there is the role of the giving part, of the object given, and of the receiver, each corresponding to a syntactic argu ment. There are other functional concepts, too, which may play a role. The time and location of an act of giving; the equivalent received for the object given; the purpose of the act; the reason etc. The values of these functions can be specified by means of various adverbial expressions. A very interesting question is whether there is a stock of general functional concepts which apply to large sets of verbs, or whether the functional concepts are more or less specific for every single verb. We cannot discuss this question here in any adequate manner. But there seem to be hints that point into the direction of the first alternative. For action verbs, the role of the agent is not only a functional concept but it is even unambiguous in the reverse d irection (i.e. bi-un.ique), given sufficient temporal (and contextual) restriction of the situation. For an individual act of giving at time t0 there is one and only one agent involved; but in the opposite directio n , for that agent at time t 0, there is one and only one action (s)he performs. We can ask who gives. . . as well as what does (s)he do? In a similar way direct objects of transitive verbs or themes can be assigned to the individual events they figure in. But the answer to the question what happens to X? is, in a way, less determ i ned than the answer to the question what does X do? If it is right that the subj ect role of many verbs is distinguished by the property of providing a functional relationship in both directions, then we would be closer to an explanation for the affin ity of subjects, agents, themes, and topics: subjects of agentive verbs can be used as given material to which new events are linked, which, in turn, provide the basis for the introduction of additional objects via their thematic roles. Not all thematic roles, however, involve events. Some verbs, e.g. to have and to be, express directly relations between their syntactic arguments. I have a hammer provides a possessive relation between the speaker and a hammer, which of course is (potentially) one-to-many. Simi larly, Iocational predicates provide spatial relations between locations and objects or between objects. The nature of the relation depends on the situation. If the spatial region referred to is small enough, the relation may be functional . The location of an object is the most elementary FC2 which is u nambiguous in both directions. Every (real) object is at one and only one location. And in every location there can be one and only one object. For this reason , very often locative specifica tions are used to form simple functional concepts such as the girl behind you, the house over there, the book on the table. The mere copula use of to be does not provide any relational links, unless it combines the subject with a relational concept . H ence, sen tences such as
317 (8)
Bill is a good pianist.
involve only one object and a predication about it. V Ps o f this sort can never be used to form endophoric DDs: (9)
the man who is bold I the bold man
( 1 0)
the man who is a good pianist
5. 3. The Definite A rticle in Anaphoric DDs Once we have analysed the role of the definite article in endophoric DDs, we can account for the anaphoric cases too: they are, as it were, endophorics without the explicit specification of the functional link. I assume that the stepwise construction of a universe of discourse is comparable to the braiding of a complex multi-di mensional network. The discourse participants start from given objects or locations that figure as the first nodes of the net. (In case of completely fictitious discourse, such as fairy tales, a starting point is artificially introduced by specific phrases, such as " Once upon a time. . . " . ) Further nodes are tied to existing ones exclusively v i a relational concept links; some nodes are objects, some are events. Many of the relational lin ks are conceptually functional and the rest gains the status of functional links due to the unique position of the node and the link within the network. In addition, the nodes carry all sorts of predicative i n formation. VPs either supply predicative information for existing nodes or introduce new objects or event nodes. Every node in the network is a potential discourse referent. The whole network is a complex abstract situation, in which every constituent object or event - plays a unique role. Anaphoric DDs are used to refer to nodes in the net. They usually provide only sortal information for the retrieval of their referents. But they can be completed to make the functional link explicit that links their referents to the net. Consider the following example: (I I)
. . . The ball ran right forward t o the door o f a house that stood there, and the ball went into the house and she saw it no more.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The additional information contained in a relative clause of endophoric D Ds provides functional links between given material and the referent of the head nou n, making use of the relational and eventually functional concepts provided by the thematic roles. Whether the thematic roles single out a unique object of the kind said, depends on the situation referred to. The defi n ite article is in general not redundant in such D Ds. It instructs the hearer to take the DD as a functional concept and !n order to do so (s)he has to take the situation as constrained as is necessary to make the crucial relations una mbiguous. A common way to do this is just to assume that there are not more objects of that kind in the relation specified .
318 The fi rst occurrence of the D O the ball corresponds t o a node k 1 already established; via the functional link "what happened (to the ball)?" an event node k 2 with the sortal information "run right forward" is tied to k 1 • The fu nctional concept "goal" links the node k 3 , "the door," to k 2 . The definite article indicates that the noun door is to be taken as a functional concept and, indeed, it is linked as a functional concept to t he node k 4, via the reverse fu nctional relation "belongs to". k 4 carries the predicative information "house". (Note that the indefi nite does not establish a discourse referent out of the blue, but specifies the value of a common functional concept for an argu ment independently introduced ! So far, the part of the net we have been considering, has the following structure: the ball theme
===: :;( o� · what �o k1 k2
belong to
goal
��n )c;'
�--
happened?
house
door
run . . .
• door
�o k4
We skip over the relative clause and continue. After the event k 2 , a new event is reported, k 5 , again linked to k 1 by the FC "what happened" (we neglect temporal and causal structure). To the event k 5 of the sort "go" a goal is l inked, "the house" . Now, the hearer or reader of ( I I ) will not introduce a new node for the N P the house, because the definite article instru cts her/him in this case, also, to use the noun as a functional concept. This functional concept is provided by the particular network built u p so far. Explicitly, it would be "the house the door of which the ball ran right forward to". Apparently, it is not necessary to repeat all this. (But it is possible, and in actual speech often enough such expl ications are used to disambiguate anaphoric expressions.) It is sufficient to employ some distinctive sortal i nformation in order to refer to the node in question. By indicating by means of the definite article that the sortal noun is to be used as a functional concept, the speaker refers the hearer t o a situation which is constrained to supply t he necessary additional relational info rmation. In this way, the use of anaphoric defin ites contributes to the coherence of texts. I mentioned earlier that all relational concepts also carry sortal information . With functional concepts, i n particular, we can observe a trade-o ff between relational information and disambiguating sortal informatio n : sortal restrictions often render a one-to many relational concept a functional concept. The division of labour is overt in the case of endophoric DDs, and enables the suppression of explicit relational information in a naphoric DDs. In many cases even the sortal i nformation can be reduced to a minimum. Third person personal pronouns can be taken in this sense: as definites with a minimum of sortal information that stand for more explicit functional concepts.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 2)
319 5. 4. The Definite A rticle in Deictic DDs
( 1 3)
Pass me the cornflakes, please.
Mary is aware of the cornflakes on her plate, in John's bowl, in the pack on the table and may be of a spare pack on a shelf nearby. But she will know which ones she has to pass, because several constraints exclude all but the cornflakes in the pack on the table. John instructed her to use the noun cornflakes as functional concept, due to which the cornflakes he wanted figure in a unique role within a effectively constructible abstract situation. Again , it is not essential that the head noun of the DO is sortal. The sortal content of the head noun is only used to construct a functional concept out of an appropriately restricted (and abstracted) situation. 5. 5. The Meaning of the Definite A rticle We may thus conclude with the hypothesis that the meaning of the definite article consists in the indication that the noun is to be taken as a functional concept. The uniform analysis of the definite article depends on the possibility of reinterpreting sortal nouns as clues to contextually given fu nctional concepts. To prevent a possible misunderstanding: I do not mean that in case of pragmatic defi nites sortal or relational nouns change their interpretation to become functional nouns when com bined with the definite article. Rather the information given with their semantic content, which by i tself is insufficient, is combined with additional non-lexical information to yield a functional concept. If the head of a defi nite NP is lexically a sortal concept noun, the application of the defi nite article does not have the effect of forming the individual concept of "the only satisfier of that concept". (Rather, this would be the effect of the combination of the noun with the only.) It is functional connections that allow the speaker to refer to the referents of pragmatic defi nites unambiguously, and the crucial identifying information is not provi ded by the noun in those cases.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The use of the definite article in deictic D Ds is in principle no different from the anaphoric use. Deictic DDs are hidden FC I s with respect to t he situation of utterance. The common feature of all deictic definites is the condition that the referent m ust be a part of the situation of utterance as it is accessible to the discourse participants. The access to the situation can be mediated either by perception or by memory . Both perception and memory provide structured pictures of situations within which the representation of real objects have a well-defined and unique place. Sortal or relational restrictions can help to single out certain objects within the situation. Again, this will in general require an appropriate restriction of the situation as a whole. Such restric tions are provided by various constraints. Think of Mary and John having breakfast together. John says to Mary
320
5. 6. Discourse Representation Theories of the Definite A rticles I n view of the theory developed above it should be clear that I do not agree with the D RT approach to definiteness. To begin with, it can by no means claim to be a theory of the definite article in general. It is just a theory of (simple and straightforward forms of) anaphora. The basic idea is to store and structure the verbal information of the previous discou rse and link it to N Ps. In this way, neither deictic nor semantic definites can be handled. For the deictic definites, situational infomation provided by the situation of utterance must be available to the constructor of the discourse universe. But i t is not at all obvious how this kind of information, which is provided by non-verbal channels and is incomparably more complex, should mix with the meager verbal information of DRSs. For semantic cl.efinites one would need an immense additional apparatus in order to update the discourse representation appropriately i n any case only to be able to cover just the more trivial cases such as proper na mes or "associa tive anaphors". Semantic definites refe rring to the situation of utterance would still provide the same problems as deictic definites. 1 3 A s a theory of anaphora, too, DRT is too weak. The D R S which is constructed by the theory is essen tially unstructured. DRT cannot account for the coherence of a text , although there are cross references between the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This analysis accounts for the global properties of DDs, such as existence of a referent and non-ambiguity of the nou n . It explains the logical status of DDs as terms (and not quantifiers proper). It predicts correctly that the definite a rticle is obligatory with functional nouns and non-redundant in all other cases. The theory implies that the semantic scope of the definite article is not the whole discourse, as suggested in DRT (Discou rse Representation Theory; I refer the reader to Heim 1 982, Kamp 1 98 1 , Barwise 1 985, and related works). According to Heim ( 1 982), for example, the contribution of the defi n ite article consists in a fel icity condition constraining the previous discourse. The scope of the definite art icle is not even the whole sentence within which it occurs, nor the whole NP following it; it is j ust the noun it precedes. The definite article indicates that the noun m ust be interpreted in one of two fundamental ways, viz. as a functional concept or as a sortal concept. Both kinds of concepts are equally represented in the lexicon of natural languages. Under a sortal interpretation the referent of the noun is taken to be of a certain sort, under the functional interpretation it is linked to other objects by general relations. The fact that practically all fu nctional concepts involve a situational argu ment furthermore reflects the deictic origin of the definite article in many (all?) languages which have a definite article. As for languages which do not have a defin ite article, it is plausible to assume that they just do not explicitly express the way the nouns are to be interpreted.
321 cards. I t just assumes, but does not explain, that new discourse referents are introduced into the discourse representation structure by means of indefinite N Ps. According to the theory sketched above they do not do this by them selves but they serve as sortal information about the occupants of roles that are provided independently through the thematic roles of verbs or other expressions such as prepositions. This way it can also be explained how the referent of a noun can be u nknown or unfamiliar when introduced and known or familiar in the next moment without any new information having been added in between . The mere fact that an indefinite with the same head noun has occurred in the previous discourse cannot be the crucial point, since the anaphoric case cannot be taken as representative for all uses of the defi nite article.
.
Not only D Ds but definite N Ps i n general must be taken as functional concepts. This has already become clear for proper names and possessive N Ps during the discussion of semantic definites. A word should be added about demonstrative N Ps. All "determi ners", except for the definite article and the possessive pronoun require the noun to be taken as a sortal or relational concept. For quantificational determiners this means that they require the domain of quantification to be divided into equal parts (cf. the remarks above about quantification with functional nouns). For the indefinite article it means that it serves to form a one-place predicate together with the fol lowing substantive phrase. 1 4 Also N Ps with a demonstrative determiner taken as a whole represent functional concepts. But the roles of the noun and the determ iner a re quite different. The noun must be taken as a sortal or relational concept, and the determiner instructs the hearer to pick out one of the possible referents. The choice, however, is n·ot arbitrary but unambiguous. Usually, demonstratives distinguish degrees of closeness or remoteness. The demonstrative is then tantamount to a functional specification such as the N at a certain distance. We thus observe the same division of labour as with endophoric DDs: the head noun is sortal and the rest provides for the functional connection. The localistic explanation of demonstratives is probably not general enough. What is crucial is not spatial distance but rather the degree of awareness of the object. If a demonstrative expression is accompanied by an ostensive act, this act is meant to direct the attention of the other person to a certain object in order to make it most or at least more prominent to his mind. Simi larly, when demonstratives are used anaphorically, they direct the hear er's attention to objects of a certain degree of awareness. Again, "the N which is most/less prominent" is a fu nctional concept. But it is different from the fu nctional concepts used for the interpretation of anaphoric or deictic D Ds, which refer to the situation as a structured complex. This explains why demonstratives cannot be combined with functional
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
5. 7. Definite NPs in general
322 nou ns, unless additional arguments are i mplicitly involved, e.g. in a sentence such as the following: ( 1 4)
This president is a tough president.
( 1 5)
A car stopped in front of the post office . { The/*This/*That l driver got out and . . . .
A fu rther difference between the definite article and demonstrative determi ners is also explained by the diffe rent status of the noun. The explicit sortal descriptive content of a demonstrative NP can be further reduced by the total omission of the head noun, leaving just the demonstrative in the function of an NP of its own. The same pronominalization p rocess occurs with quantifi cational determiners . But this is i mpossible with the definite article - as our theory predicts. To sum marize , the definite article not only indicates the functional concept role of the noun, but it is also the only determiner which has this function. A question which I cannot address here is whether the proposed analysis of the determ iners should have consequences for the syntax of N Ps. Do functio nal nouns, or better, nouns in functional concept role, constitute a syntactic subcategory of their own, different from a respective subcategory of nouns in sortal concept roles? There are data which seem to support this view: charac teristic complements of functional nouns and the determiner distri bution . Or should we keep to a uniform , undivided category of nouns? This is, in the first line, an e mpirical syntactic question, independent of semantic considera tions. If this issue is settled we can start to think about how to relate the syntactic structure and the semantic roles of its constituents. I believe that it is possible to achieve a compositional analysis of NPs with uniform determiner and noun meanings. I do n ot know, t hough, what a theory would look like which connects syntax and semantics and lives up to the rigid standards of e.g. Montague Grammar. What I tried to do here is provide an independent
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
where the president (of the time and organization implicitly referred to) is contrasted with the presidents of other t imes, organizations, or alternative circumstances. The inherent non-ambigu i ty of functional concepts is incom patible with the possibility, which is i mplied by the use of demonstratives of alternatives of the same sort or, a t least, the possibil ity of alternative loca tions. The only way to provide alternatives in either sense for the value of a functional concept consists in referring to alternative arguments. If this is i mpossible, i.e. if the arguments are all fixed, then demonstrative determiners - in the straightforward sense considered here - cannot be used with function al concept nouns. This accounts for the fact that demonstratives do not occur in uses such a s the "associative anaphoric use", with FC2 heads referring i mplicitly to arguments fixed in the context.
323 semantic analysis of definiteness which might contribute to the discussion of an overall theory integrating syntax, semantics, and pragmatics. Seminar fiir Allgemeine Sprachwisseruchaft Universitdt Dusseldorf Umversitd/Sstr. I 4000 Diisseldorf Fed. R�p. Germany
NOTES I.
In addition to the t h ree standard types, Bunt
( 1 985: 1 3) gives exam ples of plural mass nouns
such as oats or mash�d potato�s; but I think they can be considered marginal. For formal theones which ass1gn different meanings to smgular and plural definite article,
3.
( 1 973), Scha ( 1 9 8 1 ) and Barw1se and Cooper ( 1 98 1 ). ,
I generally exclude the pathological case of empty universes. I n the next section it will
become clear that all cases of what I call "referential" quantification presuppose a non-empty domain of quantification.
4.
in Barwise and Cooper
5.
is shown there that is considered the charactenstic property of definites
Cf. Lohner (forthcoming) for a more formal version of the argument. I t
definites constitute not only filters - which
( 1 98 1 : 1 831) - but in fact ultrafilters.
Scha ( 1 98 1 ) offers an extended discussion of the complexities of plural predication. See Link
(forthcoming) and umning (forthcoming) for recent analyses i n th1s area. 6.
Cf. d e Jong (forthcommg) for that. The determiner all
is an exception in that it occurs in
place of the definite article in partitiVe constructions. But then it is not a quantifier. Cf. the exceptional treatment of all in Barwise and Cooper
( 1 98 1 : 2061) and the detailed discussion in
Link (forthcoming). I do not talk about constructions such as thru out offour Americans. The NP after of is again not quantificational here, and the whole phrase can only be taken as a generic quantifier.
7.
Somethmg which at a first glance m1ght appear as double quantification can occur if
adverbial quantification of extent is applied to singular objects, as in (1)
Cologne is partly destroyed.
and if this k i nd of quantification 1s then em bedded in a nominal quantification: (1i)
Many towns are partly destroyed.
O f course, these are two different quantifications. The quantifier many modifies the apphcation of the predicate b� partly dntroyed to "the towns", while partly concerns the application of the predicate b� destroyed to single towns.
8.
There is a German word which exactly expresses what I mean with "non-ambiguity":
the term Eind�utigk�it. Unfortunately, there does not seem to be a straightforward English equivalent.
9.
For a n extensive discussion of this range of phenomena see Ulbner ( 1 979) and the summary
in Lohner
( 1 98 1 ). The type of intensionality referred to is that in TM t�mperature is rising. It is
intrinsically connected with a functional interpretation of the NPs involved.
1 0.
One of those connections where the latter (overall) notion of definiteness is relevant
is the
so-<:alled definiteness effect in existential th�r� sentences which allow only for indefinite N Ps in subject position. Woisetschlaeger
( 1 983) discu= many examples of definite N Ps (in the sense
used here) which in fact can occur in the subject positions of such sentences. One group of examples involves FC2 nouns wit h an indefinite argument, e.g. Ther� was th� sm�/1 ofpot all over th� appartm�nt. What matters, apparently,
is the overall indefiniteness of the s ubject NP.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
2.
cf. e.g. Bartsch
3 24 I I.
It might be objected that instead of ( 1 7) and
( 1 8), ( 1 7) and ( 1 8') would be preferred, and
that the genitive as well a s the possess1ve pronoun construction might not be definite.
( 1 7')
H e was a poor farmer's son.
( 1 8')
I do not want my daughter to marry her sister's lover.
In German a nd in French, however, the analogous constructions are i m possible. In German, e.g.,
( 1 7}-( 1 9) would translate as follows: Er war du Sohn eines armen Farmers.
( 1 8'')
lch mOchte mcht, daB meine Tochter dm Geliebten ihrer Schwester heiratet.
1 9'')
Er legte ihr di� Hand aufs Kme.
1 2.
With regard to the analys1s of relative clauses in this connection, I owe much to the
discussion i n Ebert
1 3.
( 1 970: 1 76ff.).
Cf. the considerations in Heim
( 1 982: 364ff) about how her theory should be modified in
order to account for semantiC defimtes. The modifications appear to be so fundamental that not much would be left of the original des1gn of DRT.
14.
I am fully aware of the msufficiency of this remark about the indefimte article. But I find
myself in agreement with all major theories on th1s point. Cf. Hawkms
( 1 978: 1 75ff.) who
considers the non-inclusiveness as fundamental - which excludes a functional interpretation of the noun. DRT translates indefinites as one-place predicates about discourse referents, and Situation Semantics, as far as can be J udged from the scarce remarks m Barwise and Perry ( 1 983) shares a similar view: the noun of indefimte NPs 1 m poses sortal constraints on the anchoring of indetermmates.
R E F E R ENCES
1 973: The semantics and syntax of n umber and numbers. In: J. Kimball (ed.), 2. Seminar Press, New York, pp. 52-93. Barwise, Jon, 1 985: A Model of th� Tr�atm�nt of Anaphora in Situation S�mantics. (Tech. Rep. Bartsch, Renate,
Syntax and S�mantics Vol.
Working Paper # I ). CS L I , Stanford. Barwise, Jon and Robin Cooper, and Philosophy
198 1 : Generalized quantifiers and natural language. Linguistics
4: 1 59-2 1 9.
1983: Situations and A ttitud�s. M . I .T. Press, Cambridge, Mass. 1 923- 1 932: D�utsch� Syntax. Carl Winter, Heidelberg. Bunt, H a rry C., 1 985: Mass T�rms and Mod�l-th�oretic s�mantics. Cambridge University Press,
Barwise, Jon and John Perry, Behaghel, Otto, Cambridge. Christophersen,
Paul,
1939: Th� Articl�s. A Study of Thdr Th�ory and Use in English.
M u nksgaard, Copenhagen and M ilford, Oxford.
0., 1964: A Grammar of th� German Languag�. 2nd edition. Frederick Ungar,
Curme, George New York. Ebert,
Karen
H.,
1970: R�fn�nz, Spruhsituation und di� b�stlmmt�n A rtik�l tn �in�m
nordfri�sisch�n Dial�kt (F�nng}. Doct. Diss., Univ. of Kiel. Nordfriisk lnstitut, Bredstedt,
1 97 1 . 1 89 1 : Funktion und &griff Reprinted in: Gottlob Frege: Funktion, &griff, 1 969, pp. 1 7-39. Galton, Antony, 1 984: Th� Logic of Asput. Clarendon Press, Oxford. Frege, Gottlob,
B�d�utung. Vandenhoeck and Rupprecht, Gtlttingen,
Gl!rdenfors.
Peter
(ed.),
forthcoming:
Gmua/iud
Quanti.fi�rs:
Linguistic
Approach�s. Studies in Linguistics and Philosophy Series. Reidel, Dordrecht.
and Lagtca/
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 7")
325 1 98 1 : Formal 1 35/ 1 36). Mathemaucal
Groenendijk, Jeroen A . G . , Theo M . V. Janssen, and Martin B . J . Stokhof (eds.), M�thOtb in th� Study of Languag� (= Mathematical Centre Tracts Centre, Amsterdam. Haberland, Hartmut,
1 985: Zum Problem der Verschmelzung von Prliposition und bestimmten
Artikel im Deutschen. Osnabriicku B�itrtig� zur Sprachth�ori� 30: 85- 106. Hartmann, Dietrich, 1980: O ber Verschmelzungen v o n PrlipositiOn u n d bestimmtem Artikel. Z�itschrift for Dial�ktologi� und Linguistik, Hartmann, Dietrich,
47: 1 60- 1 83. 1 982: Deixis and anaphora in German dialects: the semantics a n d
pragmatics of t w o definite articles in dialectal varieties. I n : Jilrgen Weissenborn and Wolfgang Klein (eds.), H�r� and Thu�. Cross-linguistic Studi�s on D�ixts andD�monstration. Benjamins, Amsterdam/Philadelphia, pp.
1 87-207. 1978: D�jiniten�ss and Ind�ftniten�ss. Croom Helm, London.
Hawkins, John A., Heim, Irene,
1982: Th� S�mantics of D�jinite and Ind�jinitr Noun Phrases. UMass Dissertation 73). Univers1ty of Konstanz,
(Schriftenre1he des Sonderforschungsbereichs 99, Linguistik, Nr.
1985: MetalingUistic negation and pragmatic ambiguity. Languag� 6 1 :
1 2 1- 1 74. Janssen, Theo M.V.,
1 984: I ndividual concepts are useful. I n : Fred Landman, Frank Veltman 1 7 1 - 1 92. Jespersen, Otto, 1 933: Essentials of English Grammar. Allan and Unwin, London. (eds.), Vari�ti�s of Formal Semanllcs. Foris, Dordrecht, pp.
Jong, Franciska de, Forthcoming. The compositional nature of (in)definiteness. I n : A . ter Meulen, E. Reuland (eds.), Th� R�pr�untation of (In)d�jiniten�ss. M . I.T. Press, Cambridge, Mass.
1 98 1 : A theory of truth and representation. In: J.A.G. Groenendijk et al. (eds.) 1 98 1 ' pp. 277-322. Keenan, Edward L. and Karen H . Ebert, 1 973: A note on marking t ransparency and opacity. Linguistic Inquiry 4: 42 1 -424. Kramsky, Jifi, 1 976: Some ways of expressing the category of determinedness. I n J. Kramsky, Papers in G�nual linguisttcs. Mouton, Den Haag, pp. 1 8 1 - 197. Link, Godehard, 1 983: The logical analysis of plurals and mass terms: a lattice-theoretical Kamp, Hans,
approach.
In:
R.
Blluerle,
C. Schwarze, A.
von Stechow (eds.), M�aning.
Intuprnation of Languag�. De Gruyter, Berlin and New York, pp.
Use. and
302-323.
Link, Godehard, forthcoming: generalized quantifiers and plural. In: P. G!lrdenfors (ed.) (forthcoming). Lilbner, Sebasuan,
1 979: Intensional� Vub�n und Funktionalb�griff�. Zur Syntax und s�mantik
von 'wuhseln' und d�n vugl�tchbar�n Vub�n d�s Deutschen. Narr, Tilbingen.
1 98 1 : I ntens1onal verbs and functional concepts: more on the "rising 1 2: 47 1 -477. Ulbner, Sebasuan, 1 986: Quantification as a major module of natural language semantics. In: J. Ulbner, Sebastian,
temperature" problem. Linguistic Inquiry
Groenendijk, M . Stokhof (eds.), Studies in Discours� R�pr�untation Th�ory and th� Th�ory of G�nualiud Quantifius. G R ASS
8. Foris, Dordrecht.
Lilbner, Sebastian, forthcoming: natural language and generalized quantifier theory. I n : P. Gllrdenfors (ed.), forthcoming. Lenning, Jan Tore, forthcoming: collective readings of defimte and mdefinite noun phrases. I n : P . Glirdenfors (ed.), Forthcoming. Lyons, Christopher,
1980: The meaning of the english definite article. I n : J. Van der Auwera 8 1 -95.
(ed.), Th� S�mantics of Dn�rminus. Croom Helm, London, pp.
1 977: Semantics. Cambridge University Press, Cambridge. 1 970a: Univ�rsal Grammar. I n : R. Montague, 1974, pp. 222-246. Montague, Richard, 1970b: Pragmatics and Intensional Lagic. I n : R. Montague, 1 974, pp. 1 1 9- 1 47. Montague, Richard, 1 973: 17r� Propu Tr�atm�nt of Quantification in Ordinary English. I n : Montague, 1974, pp. 247-270. Montague, Richard, 1 974: Formal Philosophy. Richmond Thomason (ed.), Yale Univ. Press, Lyons, John,
Montague, Richard,
New Haven and London.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Konstanz. Hom, Laurence R.,
3 26 Russell, Bertrand,
1 9 1 9: Introduction to Math�matica/ Philosophy. Allan and Unwin, London. J.H., 1 98 1 . : Distributive, collective, and cumulative quantification. I n : J . Groenendijk, M . Stokhof (eds.), 1 98 1 , pp. 483-5 1 2. Woisetschlaeger, Erich, 1 983: On the question of definiteness in "An Old Man's Book". Linguistic Inquiry 1 4: 1 37- 1 54.
Scha,
Remll:o
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Journal of Semantics
4
:
327-339
IMPLI CIT QUANTIFICATION OF TEMPORAL ADVERBIALS
H .J. VAN J AA RSVELD and R. SCHREUDER
ABSTRACT
not set clear boundaries to the time-interval between the reference-point (usually the moment of speech) and the time at which the situation referred to occurs. In the context of particular sentences these vague adverbials receive an interpretation that restricts the length of the time-interval to a more limited range of values (cf. John hasjJLSt smok�d a cigarer and John hasjJLSt marri�d). In two experiments the contextual interpretation of vague temporal adverbials was Investiga ted by means of quantification j udgments whereby subjects had t o provide numerical estimates for the t i me-interval. The frequency and the duration of everyday human acts expressed by the verbal phrases of sentences were systematically varied and were shown to have an effect on the assumed length of the t i me-i nterval. Larger esti mates were obtained when adverbials were combined with verbal phrases that expressed infrequent acts and acts with longer duration. In a final section some theoretical implications of the experimental results are presented.
An important aspect of events is their location in t ime; in many circumstances some information about the t i me at which an event occurred or will occur is relevant. Location in time in some cases even provides the only way of identifying a particular event within a series of similar events (e.g. the train leaving at 15. 00). A striking feature of locating situations in time is the varying degree of precision. I n some contexts, e.g. science or law, high de grees of precision are typically required (e.g. The suspect left the house at 9. 30). In other contexts simple location in past time or future time without any additional information is sufficient (e.g. fairy tales or science fiction). In everyday discourse, intermediate degrees of precision (e.g. Next month we are going to London) are readily observed. Speakers often willingly provide indications of when situations occur at levels of specificity that agree with Grice's (1975) conversational maxim of Quantity. According to this maxim, speakers should be informative at a level that is conversationally adequate ("Make your contribution as informative as is required" and " Do not make your contribution more informative than is required"). Other reasons for intermediate precision undoubtedly include uncertainty with respect to the exact time location or deliberate unwillingness to commit oneself to a very precise time location of situations (we will adopt Comrie's ( 1 985) use of the word "situation" to subsume processes, events, states etc.).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Vague temporal adverb1als hke soon, r�c�nt, jJLSt, which are used t o locate situations in time, do
328
{I)
John has just smoked a cigarette.
(2)
John has just got married.
For most speakers of English , the i nterpretation ofjust is likely to be different for the two sentences. Just in sentence ( I ) means something like "a few moments ago" and it would be distinctively odd when the speaker would mean by the sentence that John smoked a cigaret a few days ago. For sentence (2), however, the former interpretation will only be appropriate when John leaves city-hall as a groom. The more usual (and perfectly acceptable) inter pretation ofjust in sentence (2) will be that John in fact married a few days or a few weeks ago. This varying interpretation ofjust stands not alone. Similar examples can be given for other vague adverbials. Consider e.g. immediately which at first sight seems to be limited to the very next moments. In the sentence After the government announced higherfoodprices, riots immr.diately started in the capital, however, immediately is likely to receive an interpreta tion that will involve hours instead of minutes. Of course, vague adverbials are not the only means for 'vaguely' locating situations in time. For expressions like one of these days, within a couple of minutes, a few years ago, upper and lower boundaries for the time-interval between t0 and t5 are also difficult to specify, but the vagueness of these expressions is based on other words (a couple, a few) than temporal adver bials.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Most languages possess elaborate lexical means to express these interme diate degrees of precision. Suitable lexical i tems are called deictic temporal adverbials (Comrie 1985, Lyons 1977). Because these items locate situations i n time with respect to a reference point or deictic centre (usually the moment of speech), they (implicitly) specify a certain time-interval between the deictic centre (t J and the moment t he situation referred to occurs (ts). For purposes of precision, deictic temporal adverbials may be divided into two classes. One class of items consists of adverbials whose meaning involves well-defined time-cycles like days, weeks, or years. Examples of items in this class are tomorrow, today, last month, next year. Although each adverbial in this class leaves a certain amount of uncertainty about the 'exact' moment at which the situation referred to occurs (which will of course be l arger as the size of the relevant time-cycle increases), a clear u pper and lower boundary for the time-interval between t0 and t , can be specified. E.g. in He is going to Africa next year the exact day of departure is left in the open, but it specifies January I as the earliest and December 3 1 as the latest day. Clear boundaries are notably absent in the second class of temporal adverbials, which may be t herefore be called vague temporal adverbials. Items that intuitively belong to this class are e.g. immediately, ever, lately, recently, soon. A common characteristic of these adverbials is that they can be used for widely varying time-intervals between t 0 and t, . Consider the inter pretation of just in sentence { I ) and (2).
329
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Although it is very difficult to specify t he upper and lower boundary of the time-interval for particular adverbs in isolation, it appears from the examples above that the interpretation of these adverbs in the context of particular sentences is limited to a more restricted range of values. Our experimental research has been concerned with aspects of sentence meanings t hat bring about these more restricted interpretations. Because temporal aspects of situations are primarily expressed by the verb, main determinants of more restricted interpretations are likely to be found in the verbal phrases of sentences. I n the sentences we used as stimulus materials the frequency and the duration of the situations expressed by the verbal phrase were systemati cally varied. The results show that these two temporal aspects affect the interpretation of vague adverbials. Research of the contextually determined i nterpretation of vague temporal adverbials may provide valuable i nsights in the way "different facets of the structure of a sentence, the context, and knowledge of the world combine to enable one to make deductions about the time reference of situations" (Comrie 1985:52). From a more general psycholinguistic point of view, our research may be regarded as a case-study of the ways in which sentence contexts affect the interpretation of constituent words, an issue which has recently attracted considerable attention (Anderson & Shifrin 1 980, Foss & Ross 1 983, Johnson-Laird 1 980). It seems likely that the interpretation of different kinds of words is affected in different ways by sentence contexts. The contextually determined relevance of certain meaning aspects of content words (e.g. cf. The man played the piano and The man lifted the piano; Barclay et a/. 1 974) is to be distinguished from the instantiation that may occur for words with general meaning (e.g. cf. The fish attacked the swimmer and The fish swam in the bowl; cf. Anderson et a/. 1976). Similarly, the contextual interpretation of different kinds of vague words will be determined by quite divergent aspects of sentences. E.G. Hormann ( 1 982, 1 983) has shown that the interpretation of indefinite quantifiers like several or a few is affected by the size of the relevant objects. The study of the various ways in which sentence contexts affect the meaning of constituent words will foster our understanding of processes that u nderly sentence interpretation. The rest of this paper consists of three main parts. In the first part we present some additional characteristics of the adverbials under considera tion. The second part contains a detailed description of our experiments. I n the third a n d final section w e will present some theoretical observations that may be deduced from our results. In his discussion of grammatical categories for temporal distance , Comrie ( 1985) proposes three parameters to describe these categories. These parame ters can also be used for a more detailed description of vague temporal adverbials, as will be seen below. The first parameter Comrie discusses is the distinction between past tenses and future tenses. Some adverbials locate situations in the past time (e.g. recently), other adverbials locate situations in the future time (e.g. soon ).
330
Experiment 1: A cceptability judgments Because i ntuitions in this area are very u ncertain , we decided to support these o bservations by asking 32 subjects to j udge the acceptability of the different types of sentences on a 7-point scale. Sentences were of four types. Each sentence contained a "near" or "remote" adverb (in past or future time), which was combined with a verbal phrase that referred to a frequent or an
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Remarkably, there are some adverbials that may apply to both past and future time. E.g. t he Dutch ooit ("one day"); cf. Ooit zalde Mens op Mars Ianden ("One day man will land on Mars") and Ooit moet een grate asteroide de aarde getroffen hebben ("One day a big asteroid must have struck the eart h"). The second parameter is the reference point from which temporal distance is measured. In most cases this reference point will be the moment of speech, but the reference point itself may be situated in the past or future time (e.g. When John arrived, Peter immediately left). In the stimulus materials used for the experi ments reported below, the reference point was invariably set at the moment of speech. It should therefore be borne in mind that conclusions drawn from the experiments only apply to situations with this particular reference poi nt. The third parameter with regard to temporal distance is the n umber of distinctions. Most languages use a small n umber of grammatical distinctions like hodiemal (for "today") and pre-hodiernal ("before today") or "recent" and "non-recent". Each distinction requires the establishment of cut-off points; a specification of the time-interval that the distinction applies to (like "the present day" for hodiernal tense). Comrie notes two problems with respect to these cut-off points. One is the precise establishment of the <:: U t-off point. E.g. the d ividing line between days may be drawn at m idnight or at the time of waking up. The second problem is the rigidity with which the cut-off points are interpreted. In some languages, literally inappropriate tenses may be used to convey subjective impressions of larger or smaller temporal distance. Comrie presents an example sentence from Sotho, in which the speaker uses the recent past tense for an event in 1 824 to convey the subjective recency of the event. In other languages, e .g. Haya, the dividing lines between tenses must be strictly adhered to. As already noted , it will be extremely difficult to specify the cut-off points of vague temporal adverbials in isolation. However, a general distinction between "near" and "remote" seems to hold for vague adverbials, leading to the four-way classification of these adverbials as displayed in Table I . The distinction between "near" and "remote" i s suggested b y the observa tion that combinations of some adverbials with verbal phrases referring to 'every-day' activities tend to be u n acceptable, e.g. •John has slept recently, •John will eat in the foreseeablefuture. The reverse - combinations of "near" adverbials with expressions referring t o infrequent activities - does not lead to unacceptable sentences, although they may sound a l ittle odd (e.g. John will immediately paint his house).
33 1 TABLE
1: A four-way classification for a sample of vague temporal adverbials in Dutch. REMOTENESS
PAST
R EMOTE
pas (just now) net (just) daarstraks (a while ago) zojuist (a moment ago)
Iaatst (recently, lately) destijds (at that time, in those days) onlangs (the other day, lately)
direct (directly, instantly) ogenblikkelijk (this instant) onmiddellijk (immediately, at once) (zo) dadelijk (right away)
weldra (soon, before long) binnenkort (fairly soon, shortly) binnen afzienbare tijd (in the foreseeable future) spoedig (before long)
TENSE
FUTURE
infrequent h u man activity or act. Each adverb was classified by the two authors and a third rater as "near" or "remote" and the frequency of the human acts was estimated beforehand by scaling procedures (see below). Two groups of 1 6 subjects each were run . One group was presented with 8 "near" adverbs, the other with 8 "remote" adverbs. The same set of l 6 verbal expressions for frequent and infrequent acts was used for the two groups. Adverbs were randomly assigned to verbal expressions and these combina tions were the same for all subjects within a particular group. Table 2 presents examples of each of the four types of sentences used, together with the mean acceptability score for each type. As can be seen from inspecting the means in this table, acceptability scores were clearly Jess for combinations of"remote" adverbials and expressions for frequent acts. The interaction between kind of adverbial and frequency of act was significant (F( l ,28) = 3.59, p < .06). These results demonstrate that the classification of vague adverbials as "near" or �·remote" accounts for differ ences in acceptability between different types of sentences. As noted above, the distinction is asymmetrical. "Remote" adverbials are 'marked' ; they cannot be used in combination with expressions for frequent activities, but " near" adverbials may be combined with expressions for frequent and infre quen� activities.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
NEAR
332 TA BLE 2 : Mean acceptability scores for different combinations o f "near" a n d "remote" adverbials with expressions for frequent and infrequent activities( I = unacceptable; ? = acceptab le). NEAR ADVERBIALS FREQUENT ACTS Marie is w juist de straat overgestoken
4.4
(Mary has crossed the street a moment ago) INFREQUENT ACTS Marga gaat ogenblikkelijk een museum beroeken
5.0
(Marge is going to visit a museum this instant)
FREQUENT ACTS Richard heeft recentelijk geslapen
2.9
(Richard has recently slept) I N FREQUENT ACTS Peter gaat spoedig schaken
5.3
(Before long Peter will play chess)
Experiment 2: Quantification judgments and frequency For the second experiment we established beforehand the relative frequency and duration of the situations. To avoid very large differences in relative frequency, we purposely limited the kind of situations to be considered to ordinary h uman activities or acts. Representative examples of expressions in our stimulus materials are to open a door, to watch t. v., to write a letter, to paint a house. Fifty subjects were asked to rate both the frequency and the duration of each of 68 hu man acts on a 9-point scale. Instructions to subjects stressed that they had to base their ratings on the frequency with which the 'average' Dutch man or woman would perform the acts. On the basis of the obtained ratings, two sets of 1 4 expressions for human acts were selected. One set contained expressions for relative frequent acts (mean frequency rating was 6.8) , the other set expressions for relative infrequent acts (mean frequency rating was 2. 7). The acts in these two sets differed significantly for mean frequency (t(26) = 25. 1 4, p < . 00 1 ), but were equal for mean duration (t(26) = 1 . 64 , p = . 1 1 ). To investigate the more restricted i nterpretation of vague adverbials in sentence contexts, we adopted the following experimental procedure. Sub j ects were asked to supply a n umerical value for the time-interval between the reference point (which was said to be the moment of speech) and t., the time at which the situation referred to occurs. Thus, for each sentence in the stimulus materials the subject had to answer the question "How many minutes (hours, days) will normally or usually elapse between the utterance of the sentence and the moment at which the situation will occur?" The numerical value may
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
REMOTE ADVERBIALS
333
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
be taken as representative for the length of the time-interval the subject implicitly assumed for the adverbial. This use of quantification judgments has been applied successfully in the study of other vague expressions than temporal adverbials. Intensive adver bials (e.g. very), probabilistic adverbials (e.g. possibly) and frequency adver bials (e.g. often) have been shown to alter the meaning of the words they modify in a multiplicative function (Cliff 1 959, Howe 1 964, Lilly 1 968). By means of quantification judgments, Hormann ( 1 982) has shown that the number of items assumed for indefinite quantifiers (a few, some, several) decreases as the size of the objects involved increases. Our first hypothesis about the more restricted interpretation of vague temporal adverbials in sentence contexts is deriv�d from the observation above that "remote" adverbials cannot be combined with verbal phrases referring to 'everyday' activities. "Near" adverbials may, however, be used to locate infrequent situations as well as 'everyday' or frequent situations. The location of infrequent human activities in time will by definition on average involve longer time stretches than frequent situations. For a particular adverb the time-interval may therefore be assumed to be larger when combi ned with verbal phrases that refer to infrequent situations. We tested this hypothesis for five "near" adverbials, each of which was combined with verbal phrases referring to either frequent or infrequent situations. Three of them involved future tense (onmiddel/ijk ("immediately"), zo meteen ("right away"), straks ("presently")); the remaining two involved past tense (zo juist ("a moment ago"), pas ("just now")). Across subjects , each of the five adverbials was combined an (about) equal number of times with each of 28 expressions for human acts. Of these 28 expressions, 14 referred to frequent acts and 1 4 to infrequent acts. Combinations of adverbials and verbal expres sions were made into sentences by inserting proper names into subject positions (e.g. John hasjustplayed chess; Mary is going to a meeting immediate ly). To avoid too many estimates for a particular adverb by the same subject, sets of expressions for frequent and infrequent acts were administered to two different groups of 1 6 subjects each (each subject provided 14 numerical estimates). All subjects were students at Nijmegen University and were . randomly assigned to groups. The results of this quantifica tion experiment are presented in Table 3. These results clearly show that the assumed length of the time-interval for "near" adverbs varies with the relative frequency of the acts expressed by the verbal phrase in the sentence. For each of the five "near" adverbs, time-inter vals were estimated to be larger when combined with infrequent acts (all differences were at least significant at the 5 percent level of significance by t-test, except for straks for which a p-value of .06 was obtained). In connection with the third experiment reported below, it should be noted that the experimental results do not provide any indication of the combined effects of frequency and duration on the assumed length of the time-intervals. In this experiment the activities differed in frequency but were comparable in
334 TABLE 3: Mean estimates in minutes for time-intervals between t and t• when combined 0 with expressions for frequent and infrequent human acts.
KIND OF ACf ADVERBIALS onmiddeUijk (im mediately) zo meteen (right away) straks (presently) zo juist (a moment ago) pas (just now)
FREQUENT 6
INFREQUENT 33
20
34
1 00
130
20
41
436
1 547
Experiment 3: Quantification judgments and duration A direct measure for the amount of time involved in a situation is its duration . To limit ourselves again to human acts, some acts (e.g. to open a door) generally take less time to perform than others (e.g. to read a book). Analo gous to the effects of frequency on numerical estimations, time-intervals for vague adverbials may be assumed to be larger when combined with verbal phrases which express situations with longer duration. To test this hypothe sis, we formed 24 pairs of verb/noun combinations. One member of each pair contained a verb referring to an act which had a short duration with respect to the noun in object position. The verb in the other member of each pair expressed an act with longer duration with respect to the same noun. Verb /noun combinations were made into sen tences by inserting different human agents in subject position. A l ist of the 24 pairs of sentences is presented in the Appendix. As can be seen from this list, one verbal phrase of one member of each pair describes an act which can be conceived as momentarily (or at least with a very short duration), the other member of each pair contains a verbal phrase that describes an act with more extensive duration. In fact, we i n tended one member of each pair to be an achievement verb, the other an accomplishment verb (Vendler 1 967). Although the acts differ in relative frequ ency across the 24 pairs, the frequency of the acts expressed by the two members of each pair separately is roughly comparable. In an independent test we obtained estimates for the relative duration of the acts expressed by the 24 sentence pairs. N ine subjects indicated for each sentence pair which act had the longer duration . For all pairs the duration of the accomplishment
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
duration. In the next experiment t he activities had different durations but were matched for frequency. An exhaustive investigation of a possible inter action between frequency and duration is excluded by the fact that, as was shown i n Experiment I , combinations of "remote" adverbials with expres sions for frequent activities tend to be u nacceptable. Moreover, for the selected domain of human activities the class of frequent activi t ies with long duration is very small .
335
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
verbs was estimated to be longer. Differences in time interval estimates for adverbs in combination with achievement and accomplishment verbs may therefore be safely attributed to differences in the duration of t he acts. In this third experiment subjects were asked to provide numerical estimates for 12 different adverbs. Each adverb was randomly assigned to two pairs of sentences (i.e. subjects provided two estimates for each adverb). The 1 2 adverbs were divided into four classes of three adverbs each: "near" past, "remote" past, "near" future, and "remote" future (cf. Table 1 ). The assign ment of particular adverbs to classes was based on t he classification of three judges. Members of each pair of sentences were assigned to two different groups of 20 subjects each . One group was presented with accomplishment verbs, the other group with achievement verbs. The results clearly supported the hypo thesis. For 1 8 out of 24 pairs the time-interval was estimated larger when combined with expressions for acts with longer duration. This result was significant by sign-test (p < . 00 1 ) . To summarize our results: t he length of t h e time-interval that is assumed for vague temporal adverbials depends on the sentence context. In our stimulus materials we varied the frequency and the duration of the acts expressed by the verbal phrase. For infrequent acts and for acts with longer duration, time-intervals for vague adverbials were systematically estimated to be larger. An important question has to do with the formal representation of this contextual variation. An elaborate discussion of this complex issue (Dowty 1 979, Pinkal 1 979) is clearly beyond the scope of this paper, but two main alternatives may be sketched informally. One alternative maintains that the meaning of the adverbial is essentially constant and that it combines multiplicatively with the duration and the frequency of situations expressed by sentence verbs. Vague temporal adverbials should be considered as fixed operators on the meaning of other words (particularly verbs), analogous to intensive or frequency adverbs (cf. Cliff 1 959, Lilly 1 968). The other alterna tive proposes that the meaning of the the adverb is inherently vague and variable. I n the context of a particular sentence, its meaning will depend (partly) on the duration and frequency of the act expressed by the main verb. It will be noted that our formulations implicitly assume the second alterna tive, but that our results are compatible with both alternatives. Our results intuitively make sense because one would expect general temporal aspects of situations like frequency or duration to be of prime importance for the interpretation of vague temporal adverbials. This does not imply that they are the only determinants. In particular two other factors may have a similar effect. Dahl ( 1 984) notes that personal involvement affects subjective judgments of temporal distance. Events of direct personal concern will generally be felt as being "closer" in time. It is therefore likely that factors having to do with personal involvement may also have a systematic effect on the size of the assumed time-interval for vague adverbials. A second factor is the agent of the act referred to. The sentence in our stimulus materials that
336
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
referred to the opening of a new park by the city induced many subjects to remark ironically that that would take a very long time indeed. Some agents typically perform slowly (children, invalids, bureaucracy), while others are usually fas t (yuppies, feminists, criminals). Our results indirectly support linguistic theorizing that aims at a unified account of tense and aspect (Johnson 198 1 ). While tense is conceived as a category that is inherently deictic, i.e. concerned with the location of situa tions in time, aspect is related to the temporal structure of events. Clearly, the frequency of situations has to do with habituality or iterativity and duration of situations with the aspectual distinction between punctuality and durativi ty (Comrie 1 976). Because our results show that aspectual distinctions may affect the location of situations in time, they suggest that the dividing line between tense and aspect should not be drawn too sharply. Another linguistic issue concerns the categorisation of tense. Although tense is usually regarded as a category of the verb, it has also been claimed that it belongs to the whole sentence (Lyons 1 977). Our results favor the latter position: the time location of a situation depends not only on the verb but also on the adverb. The suggestions about a similar role of personal involve ment and agents point in the same direction. We noted above that the interpretation of vague adverbials in sentence contexts may be regarded as a case study of the ways in which sentence interpretation may affect the meaning of constituent words. Interesting questions are related to the processing of these contextualized meanings. One question concerns the temporal aspects. It has been suggested (Marslen-Wil son, Tyler & Seidenberg 1978) that sentence interpretation is an immediate process; i.e. the meanings of incoming words are directly incorporated into a preliminary sentence representation. Although immediate interpretation may generally apply to the processing of words with rather precise meanings, it can only be valid for vague words when the preceding sentence fragment contains relevant information. A more elaborated semantic processing of vague words is likely to be upheld till relevant information is presented, as would happen in case the temporal adverbial precedes the verbal phrase. A second question has to do with the kind of processing involved. Al though we used numerical estimates as indication for the assumed length of the time-interval between t0 and t5, it seems very unlikely that in everyday communication language users provide point-like estimates for vague tempo ral adverbials. Even in the contexts of particular sentences, considerable vagueness is left (but it is less than for the word in isolation). It is at present unclear exactly what kind of processes cause this reduction in vagueness, but in their outcome they bear similarity to the processing of terms with general meanings. A more restricted interpretation of vague adverbials occurs automatically, as shown by the fact that the sentence John recently slept is unacceptable, except in special situations. Similarly, the fish in the sentence The fish swam in the bowl will be restricted to particular kind of fishes.
337 Dtpartm�nt ofG�nual linguistics and lnt�rfae��lty R�search Unitfor Languag� andSpuch Catholic Univ�rsity Nijm�g�n
APPENDIX
Below are listed t h e 24 pairs o f sentences used in Experiment
3. The first verb
of each pair is an achievement verb, the second verb is an accomplishment verb. Onze voetbalclub heeft net een zware wedstrijd gewonnen/gespeeld. (Our footbal-team has just won/played a difficult match.) (The secretary has just interrupted/conducted the telephone conver sation.)
De buurvrouw heeft zo juist het geld gevonden/geteld. (A moment ago the neighbour has fou nd/counted the money.) De jongen heeft zo juist zijn knie bezeerd/verbonden. (A moment ago the boy has hurt/tied up his knee.) Klaas is (heeft) pas (voor) zijn rij-examen geslaagd/afgelegd. (Charles has just passed/taken his driving test.) De directeur heeft pas de brief ondertekend/gedicteerd. (The director has just signed/dictated the letter.) De familie/autoriteiten heeft/hebben laatst het l ijk geidentificeerd/opgegra ven. (The family/the autorities has/have recently identified/exhumed the body.)
De bibliothecaris heeft laatst een oud manuscript ontdekt/ontcijferd. (The librarian has recently discovered/deciphered an old manus cript . ) Albert Heyn heeft destijds een filiaal in Amsterdam geopend/gebouwd. (In those days Albert Heyn has opened/built a branch in Amster dam . ) D e president verraste/hield destijds (met) zijn toespraak (voor) h e t genoot schap. (In those days the president surprised the society with his speech/ gave his speech for the society . ) H e t bedrijf heeft onlangs e e n verzoek t o t belastingvermindering ingediend/ opgesteld. (Not long ago the company has submitted/framed a request for tax reduction.)
De juwelier heeft onlangs een dief (op heterdaad) betrapt/tot op straat achtervolgd. (Not long ago the jeweller has taken a thief in the very act/has chased a thief in the street.) Marie zal haar moeder onmiddellijk een aardige brief sturen/schrijven. (Mary will immediately send/write her mother a nice letter.)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
De secretaresse heeft net bet telefoongesprek onderbrokenlgevoerd.
338 Piet zal onmiddellijk een boek over economie bestellenllezen. (Peter will im mediately order/read a book about economics .) Jan zal zo meteen een sigaret opsteken/roken. (John will lit/smoke a cigaret right away.) Vader zal zo meteen de radio (op Hilversum 3) aanzetten/afstemmen. (Father will right away turn on the radio/tune the radio to H. 3.) H ij zal zijn vrouw straks het cadeau geven/het cadeau voor zijn vrouw inpakken. (Presently he will give the present to his wife/pack up the present for his wife . )
De bloeiende onderneming zal straks e e n lening afsluiten/aflossen. (The flourishing company will presently negotiate/pay off a loan.) (The expedition will soon reach/determine her final destination.) De politie zal binnenkort een aantal verdachten van de moord arreste ren/verhoren. (The police will soon arrest/interrogate some suspects ofthe murder.) De gemeente zal spoedig een nieuw park voor het publiek openstellen/aan leggen. (The city will very soon open/lay out a new park for the public.) De kunstenaar zal spoedig zijn bronzen beeld onthullen/gieten. (The artist will very soon unveal/cast his bronze statue.) De belangrij kste erfgenaam zal binnen afzienbare tijd zijn erfdeel opeisen/ opmaken. (In the not too distant future the main heir will de mand/spend his heritage . ) H e t pas getrouwde stet zal binnen afzienbare tijd h u n nieuwe wooing betrek ken/behangen. (In the not too distant future the newly weds will move into/paper their new home . )
REFERENCES
Anderson, R.C., Pichen, J. W., Goetz, E.T., Schallen, D.L., Stevens, K. V., & Trollip, S.R. 1 976: Instantiation of general terms. Journal of Verbal Learning and Verbal Behavior 1 5:667-679. Anderson, R.C., & Shifrin, Z. 1 980: The meaning of words in context. In: R. Spiro, B.C. Bruce & W.F. Brewer (eds.), Theoretical In��es in Reading CompreMnsion. Erlbaum, Hillsdale, N.J.
Barclay, J.D., Bransford, J.D., Franks, J.J., McCarrel, N.S. & Nitsch, K. 1974: Comprehension and semantic flexibility. Journal of Verbal Learning and Verbal Behavior 1 3:47 1 -48 1 .
Oiff, N . 1 959: Adverbs a s multipliers. Psychological Review 66:27-44. Comrie, B. 1 976: Aspect. Cambridge University Press , Cambridge.
Comrie, B. 1 985: Tense. Cambridge University Press , Cambridge. Dahl, 6.
1984: Temporal distance:
remoteness distinctions in tense-aspect system. In:
B. Butterworth, B . Comrie & 6. Dahl(eds.), ExplanationsofLanguage Universals. Mouton, The
Hague. Dowty, D . R . 1 979: Word Meaning and Montague Grammar. Reidel, Dordrecht.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
De expeditie zal binnenkort haar eindbestemming bereiken/uitzoeken.
339 Foss, D.J., & Ross, J . R . 1 983: Great expectations: context effects d u ring sentence processing. In: G.B. Floresd'Arcais & R.J. Jarvella (eds.}, TM Procns ofLanguag� Understanding. John Wiley, London. Grice, H .P. 1 975: Logic and conversation. In: P. Cole & J. Morgan (eds.}, Spuch Acts. Syntax and s�mant/CS 3. AcademiC Press, New York:. Howe, E.S. 1 962: Probabilistic adverbial qualifications of adjectives. Journal of Verbal uarning and Verbal &havior 1 : 225-242.
V�rbal uarning and Vubal Behavior 7 : 3 3 3-336.
Marslen-Wilson, W . O . , Tyler, L.K., & Seidenberg, M.S. 1 978: Sentence processing and the clause boundary. In: W.J.M. Level! & G .B. Flores d'Arcais(eds.), Studi�sinth�PerceptionofLanguag�. Wiley, London. Lyons, J. l977: S�mantics. Cambridge University Press, Cambridge. Pink:al, M. 1 979: How to refer with vague descriptions. In: R. Bauerle, U. Egli & A. von Stechow ( eds.), Semantics from D ifferent Points of View. Springer, Berlin. Vendler, Z. 1 967: Verbs and times. In: Z. Vendler, Linguistics in Phtlosophy. Cornell University Press, Ithaca, N.Y.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
H6rrnann, H. 1 982: Hidden determinants of understanding. In: J.-F. Le Ny & W. Kintsch (eds.), Languag� and Compr�h�nsion. Nonh Holland, Amsterdam. H6rrnann, H. 1 983: The calculaung listene r or how many are �inig�. m�hr�r� and �in paar (some , several , and a few)? In: R. B lluerle, C. Schwane & A. von Stechow (eds.), Meaning. Use. and lnttrpretation ofLanguag�. De Gruyter, Berhn. Johnson, M . R. 1 98 1 : A unified temporal theory of tense and aspect: In: P.J. Tedeschi & A. Zaenen (eds.), Tu1.u and Asp�ct. Syntax and S�mantics 14. Academic Press , New York:. Johnson-Laird, P.N. 1 980: Mental models in cognitive science. Cognitive Science 4:7 1 - 1 1 5. Lilly, R . S . 1 968: The qualification of evaluative adjectives by frequency adverbs. Journal of
Journal of Semantics 4 : 341 -370
CONTROL MECHANISMS OF ANAPHORA
LI LIANE TASMOWSKI and S. PAUL VERLUYTEN
ABSTRACT
INTRODUCTION
In a previous article that appeared in this journal ( 1 982) we put forward a unified theory of pronominal control mechanisms. This theory crucially hinges on the strong claim that 'true' pronouns, i.e. all items that are usually taken to be pronouns except deictics (in a sense we attempted to provide a clear definition for these), always involve some kind of a linguistic antecedent. 'Antecedent' obviously cannot be taken to mean here that, in every case, the pronoun is preceded within discourse by a noun with which the pronoun agrees. I n well-known cases (e.g. 'Well , he's left' uttered when there is no previous discourse; Lasnik 1 976:2) where there is no preceding noun within discourse, we claim there is nevertheless agreement of the pronoun with a controller which can only be a l i nguistic object, a lexical item , a noun. We refer to our previous publications ( 1 98 1 , 1 982) for a full argumentation for our position. The crucial examples we adduce are as follows: (1)
(John wants his pants that are on a chair a n d h e says to Mary:) Could you hand { them/* it f to me, please?
(2)
(same situation, but with a shirt) Could you hand { it/* them f to me, please?
(3)
(John is trying to stuff a large table (Ia table, fem inine) in the tru n k of his car; Mary says:) Tu n'arriveras jamais a { la/*l e f faire entrer dans Ia voiture. 'You will never manage to stuff { it (fem .)/it (masc.)f into the car.'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Taking as a starting point the hypothesis that all 'true' pronouns (in a well-defined sense) have a lingu1stic antecedent even when neither the sentence nor the text provide one explicitly, we review some recent approaches where the contrary is defended or implied. It is shown that these approaches run counter to the facts at crucial pomts. The cases discussed lead to a further development of our own views with respect to a unified approach of pronominal anaphora.
342 (4)
(same situation, but with a desk (/e bureau, masculine)) Tu n'arriveras jamais a { le!•la } faire entrer dans Ia voiture. 'You will never manage to stuff { it (masc.)/it (fern.) } into the car'
{5)
Bien que les victimes n'aient pas encore ete identifiees, on sait qu' { elles/•iis} avaient { toutes/•tous} embarque a Orly Sud 'Although the victims (fern.) have not yet been identified, it is known that they all embarked at Orly Sud'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Clearly, the real-world object PANTS does not have anything inherently p lural in it (in French, it would be rendered by a word that is singular), no more than S H I RT does. Only the grammatical number of the noun (pants, plural; shirt, singular) can explain the occurrence of them in ( I ) and of it in (2). Similarly, t he objects TABLE and DESK are neither feminine, nor mascu l ine; only the French nouns Ia table and /e bureau can explain the respective occurrence of the pronouns Ia and /e. Our conclusion is that the use of a (non-deictic) pronoun always involves linguistic control by an antecedent n ou n . This antecedent may be present within the sentence, or in surrounding d iscourse; but even if it is not, as i n ( 1 )-(4), its existence must be assumed in order to explain number and gender agreement i n the pronoun. We have called the antecedent which is not expressed i n discourse 'absentee antece dent'. As we pointed out before, it is obvious that an absentee antecedent is subject to severe recoverability conditions. The situation must involve a salient object which allows for the recovery of the absentee noun. In (3) above, for instance, the presence of the table which John is trying to stuff into his car allows for the recovery of the l inguistic item Ia table, which functions as the absentee antecedent. We call the mechanism through which a real world object 'controls' a linguistic item {the antecedent), for want of a better term , 'pragmatic control'. In an unmarked context, the recovered noun will be the basic category term which refers to the object {e.g. /a table rather than /e meuble, etc.); although the concept of "basic category term" is not always without problems, and although this term may in some cases vary from one speaker to the next, recourse to it is no doubt some kind of a 'conversational maxim ' , in order to ensure the recoverability of the nou n . But we show in our 1 982 article that, given the appropriate situation , the recovered noun may be linked to the salient object metonymously (yet not metaphorically). This also means that morphological agreement of the pronoun with an absentee controller is possible only if this controller is recoverable, along the lines suggested above, and developed in our 1982 paper. If recoverability of the absentee antecedent is not ensured, morphological agreement of the pronoun with a noun which is present in discourse remains the only possibili ty. In support of this, consider the following sentence, in a context where the identity of the victims of an accident is as yet unknown :
343 Given that the (possibly male) identity of the victims is not yet known, the pronoun can only refer to the feminine linguistic antecedent victimes. In summary, our model looks as follows: (6) SALIENT OBJECT
I
pragmatically controls
l
(ABSENTEE ANTECEDENT)
I
linguistically controls
There seems to be little doubt that the model we put forward is conceptually very simple, and that the hypotheses we appeal to are rather straightforward : (i) a salient object may (cognitively) bring about the noun which is its name; (ii) this noun is the antecedent the pronoun agrees with Our 1 982 theory presents (at least) three major characteristics in which it differs from various other proposals: (A) It is a unified theory of pronominal control mechanisms, i.e. all pronouns (except for the quite different and distinct class of deictics) are controlled in a similar manner; (B) The control mechanism of pronouns is, as is pointed out above, linguistic (language-internal, if you like) rather than 'conceptual' (in a sense we shall clarify below). . (C) More accessory perhaps, but nevertheless interesting is the fact that our 1982 proposals were primarily based on observations relating to agreement between a pronoun and a noun (the latter being possibly absent from dis course) in sentences where this noun refers to an object rather than to a human. In this respect, our initial set of data is diffe rent from that of Bosch ( 1 983) (in fact, nearly all the examples used in his book involve reference to humans rather than to objects) or of Reinhart ( 1 983) (same observation, as pointed out by Cornish 1 986). Now it is not surprising that this situation leads us to a rather different and, in our view, superior perspective on pronominal control and agreement. By limiting one's attention to examples with humans, some points relating to agreement phenomena can be completely overlooked. If the noun homme, 'man', occurs in discourse, and is followed by the pronoun i/, there is no way of determining whether the masculine singular gender/number of if is condi tioned by the occurrence of masculine singular noun homme, or else by the fact that it refers to a male h uman being. We believe that this may explain
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
!
PRONOUN
344
I. A U N I F I E D CLASS O F PRONOUNS VS. ALLEGED DISTINCTIONS
1. 1. Distinctions Based on Semantic Tests Various distinctions into pronom inal classes have been proposed in the literature. In our 1 982 paper, we examined those proposed by Evans ( 1 980) and we argued against them. Here we shall concentrate our attention upon those drawn by Peter Bosch (1983), who puts forward an interesting and convincing case for a twofold distinction within the class of pronominal words. PB first opposes syntactic vs. referential pronouns (henceforth SP and RP respectively). Basically, SPs are nothing but semantically empty phonetic forms that serve to fill a position in the sentenc_e , before the predicate for instance; they are of a purely grammatical nature, rather like e.g. verb endings. SPs presumably always agree in number and gender with an antece dent, j ust as verb endings agree with e .g. the subject (if such agreement is possible in the language examined, of course). PB's 'Syntactic Agreement Principle' (SAP; 1 983:200) states that a non-referential pronoun "may ex press agreement" with another form , and not that such a pronoun will always exhibit agreement phenomena; we shall get back to this below. Thus, for PB the pronoun is syntactic in the following sentences: (7)
Jean croit qu'il est malade. 'John thinks that he is ill.'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
how other theories reach conclusions that differ from ours. We also have reasons to think that these other accounts are mistaken , or at least biased in the wrong direction, because of their limited perspective on the full range of data. In each of the three main sections of the present article we shall develop one of the major characteristics of our theory as outlined above: the unification of pronoun classes, which will be defended against proposed divisions (Section 1 ) , the analysis of the actual noun-to-pronoun control mechanism (Section 2), and questions relating to pronoun agreement phenomena (Section 3). Our purpose in each of the three sections of the present article is twofold. First, we will develop some points that were present, explicitly or implicitly, in our 1 982 paper, but on which much more work needs to be done. Secondly, we proceed to a comparison between the three major aspects of our theory we j ust mentioned, and competing proposals put forward by various authors, in particular by Peter Bosch (henceforth PB) who has elaborated an extensive theory of pronominal control and agreement mechanisms in his book (Agree ment and A naphora, 1 983), which differs from our own account on several major points.
345 (8)
Jeanne croit qu'elle est malade. 'Jeanne thinks that she is ill.'
(9)
Lui ici! 'He here ! ' .
( 1 0)
Elle ici ! 'She here ! '
(1 1)
<;a ici ! 'That here !'
Contrary to what might appear at first glance, deictics do not exhibit gram matical agreement. Thus, in order to refer to a table (Ia table, feminine, in French) with a deictic, ( I I ) will be used, not ( 1 0). PB's other class of referential pronouns are the anaphoric pronouns. APs refer to an entity the conceptual representation of which has already been established in discourse, and hence is available to be referred to. A crucial point made here by PB is that, if some kind of agreement is observed in APs, this phenomenon must be clearly distinguished from syntactic agreement between a pronoun and an antecedent noun, as it occurs with SPs. APs do not agree with the gender of some antecedent noun, but rather with what PB calls the 'conceptual gender' of the previously established conceptual representa tion of the entity referred to. The 'conceptual gender' may often be derived from the grammatical gender of the noun that is associated with the concep tual representation of an entity in a particular context! but that is not always the case. Some rather clear-cut examples seem to prove PB right in saying that 'conceptual gender' and the grammatical gender of the noun associated with the concept are not always identical, and that the AP agrees with the former, not with the latter. Consider the Dutch sentence ( 1 2), where the second occurrence of ze is without doubt an AP, and the first occurrence perhaps also:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
RPs, on the other hand, as their name indicates, occur referentially: they refer to some referent. Strictly speaking, they do not possess an antecedent, although, of course, there may be an NP i n the sentence or in surrounding discourse which refers to the same referent as the RP (i.e. which is coreferen tial with it). Therefore, if a RP exhibits something that looks like agreement, it is agreement with properties of its referent, not with a linguistic antecedent. Within his class of referential pronouns on the other hand, PB draws a further distinction between anaphoric and deictic pronouns (henceforth AP and DP). DPs make direct reference, without any i ntervening linguistic or conceptual entity. PB and we more or less agree about this class of pronouns, although we would not say, as PB does, that deictics can only be used when the situation involves an element of unexpectedness. In some languages, such as French, deictics may be morphologically distinct from the other pronouns, at least i n certain positions. The following sentences contain the occurrence of a deictic (sentences uttered for instance when meeting a person or finding an object unexpectedly):
346 ( 1 2)
Ik zei tegen dat meisje dat ze moest meekomen; maar ze wilde helemaal n iet meekomen. 'I told this girl that she should come with me; but she didn't want to come with me at all.'
(i)
As a SP is not referential, it cannot be replaced (with conservation of the same meaning) by an expression which is clearly referential, such as a nominal anaphoric expression ('the bastard', and the like); (ii) The syntactic relation between a SP and its antecedent will be preserved even when the antecedent itself is not referential. The relation between the SP and its antecedent is the same in the following two sentences:
( 1 3)
Jean croit qu'il est malade . 'John thinks that he is ill.'
( 1 4)
Personne ne croit qu'il est malade. 'Nobody thinks that he is ill.'
For the sake of clarity, we can represent these two criteria as follows: (A) SP - [-commutable] (B) SP - nobody (read, respectively: (A) if the pronoun is a syntactic pronoun, 1t 1s not commutable with a nominal refe rential expression; (B) if the pronoun is syntactic, the antecedent may be a non-referential expression such as no body). Leaving aside deictic pronouns, and given that all non-syntactic pronouns are referential, and that all non-deictic RPs are anaphoric, (A) and (B) are equivalent to: ·
(A') not-AP - [-commutable] nobody ( B') not-AP -
and hence also equivalent to:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The pronouns ze agree with the conceptual representation of 'meisje', and hence are fe minine for obvious reasons. They do not agree with the noun meisje, which appears to be neuter in Dutch : meisje takes the neuter form of the definite article, het, and the neuter forms of the adjective. Our approach impels us to maintain that a distinction between SPs and APs is an artefact, and that a linguistic antecedent is to be provided for in both cases. So let us now examine the grounds for establishing two classes of pronouns. PB adduces two criteria for distinguishing between SPs and APs:
347 (A") [+commutable] - AP (B") not-nobody - A P In other words, such criteria do not seem to allow one to recognize SPs; only APs are positively characterized. In particular, neither non-<:ommutability nor replacement of the antecedent with nobody are said to be characteristic of SPs exclusively. If we would leave it at that, we would have to conclude that PB offers no criteria for distinguishing between what he holds to be two classes of pro nouns which are of a completely different nature . . . So let us make the reasonable conjecture (but it is nothing more than a conjecture) that {B) represents, in fact, a double implication, of the form (C): -
nobody
(i.e. a non-referential antecedent such as nodoby is possible if and only if the pronoun is a SP). If (C) is correct, we have given ourselves a way of recognizing SPs. However, it then appears that both criteria adduced by PB do not yield the same results. Consider the following examples: ( 1 5)
a. Julius' father hates him. b. Julius' father hates the poor chap. c. Nobody's father hates him . (Reinhart 1983:57)
(Unless stated explicitly, we restrict our attention to readings where antece dent and pronoun or anaphor are coreferential.) Suppose we want to apply PB's criteria in order to determine whether the pronoun him in ( 1 5a) is an AP or a SP. We therefore apply the com mutability test, and get ( l 5b); as the result is grammatical, we conclude on the basis of (A") above that the pronoun must be anaphoric. Applying the second of PBs criteria, however, brings us to the opposite conclusion: replacement of the antecedent by nobody is possible, hence, if (C) is correct, the pronoun must be a SP. Similar examples may be given in French; cf. : ( 1 6)
a. J'ai promis a Jules de le ramener a Ia maison . 'I promised Julius to bring him back home.' b. J'ai promis a Jules de ramener cette triste andouille a Ia maison . 'I promised Julius to bring that poor chap back home.' c. Je ne promettrai a personne de le ramener a· Ia maison. 'I will promise no-one to bring him back home.'
As a matter of fact, PB ( 1 983: 1 34ff. and footnote 1 1 4) acknowledges the possibility of a 'pseudo-<:onnection' between the pronoun and nobody, etc. The pronoun then would be an AP, related to objects in focus which are not
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(C) SP
348 explicitly mentioned, but introduced by inference into the discourse represen tation at hand. So we are left with the following situation : if nobody (or a similar item) is followed by a pronoun, this pronoun may be related to an inferred antecedent, in which case it is an AP, (ii) or it may be related to nobody, in which case it is an SP. (i)
( 1 7)
(the husband lost his key and knocks at the door of the house, hoping his wife is at home) Nobody there ! And { she/* he} promised to be back before mid nigh t !
( 1 8)
J'ai interroge toutes ces dames. Personne n 'a admis qu' { elle/* il } connaissait le coupable. 'I interrogated all these ladies. Nobody admitted that { she/*he} knew the culprit.'
Apart from all this, if we take for granted that indeed nobody (and similar items) itself has no referent, and if a SP has to agree with it, the item nobody should have inherent number ·and (in French at least) gender. But if we suppose French personne is singular, we are unable to make sense of utter ances with reciprocal interpretation (cf. ( 1 9)), nor are we able to explain the occurrence of the plural form of the pronoun in (20): ( 1 9)
Personne ne se connaissait. 'Nobody knew each other. '
(20)
Je les ai to us interroges. Personne n'a reconnu qu'ils se connaissaient. 'I questioned them all. Nobody admitted that they knew each other.'
If we consider it to be masculine, we cannot explain pronoun or adjective agreement in cases such as: (21)
Personne d e ces demoiselles n ' a dit qu'el/e renon�it. 'Nobody of these young ladies said she would give up.'
(22)
Personne de ces demoiselles n'est satisfaite. 'Nobody of these young girls is satisfied. '
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If this is so, however, the value ofthe nobody-test for SP-hood becomes nearly void. The pronoun as preceded by nobody may be equally well related to an inferred referent, and hence be an AP, in sentence ( 1 8) as in sentence ( 1 7) below, although in ( 1 8), the nobody-test would t�ll us it is a SP:
349 Cf. also the following sentence, found in a Dutch (Flemish) magazine (for PB, whoever or Dutch wie is non-referential, just like nobody): (23)
Wie vindt dat haar borsten te groot of te klein zij n , en daar echt problemen mee heeft, stapt beter dadelijk naar een goed plastisch chirurg. (Knack Weekend 27:66, 3 July 1 985) ' Whoever finds that her breasts are too large or too small , and really feels this to be a problem, had better go to a good plastic surgeon i mmediately.'
(24)
Cette plante a ete deplacee pour qu'elle soit davantage au soleil. 'This plant has been moved in order for it to be more exposed to the sunshine.'
(25)
?* Rien n'a ete deplace pour qu'il soit davantage au soleil. 'Nothing has been moved i n order for it to be more exposed to the sunshine.'
(26)
Ce meuble a ete liquide quand il n'a plus ete utile. 'This piece of furniture was disposed of when it was no longer useful.'
(27)
?* Rien n'a ete liquid€: quand il n 'a plus ete utile. 'Nothing was disposed of when it was no longer useful .'
(28)
Ce vase ne supporterait pas qu'il soit deplace trop souvent. 'This vase would not stand for it to be moved too often . '
(29)
?* Rien ici ne supporterait qu'il soit deplace trop souvent. 'Nothing here would stand for it to be moved too often .'
Perhaps it remains possible to define SP-hood with respect to certain syntac tic positions or semantic domains; we shall examine this possibility in 1 . 2 below. Concerning the tests for SP-hood which PB adduces, on the other hand, maybe our interpretation of (B) as (C) above is not legitimate after all. But either way there is a problem : either the criteria adduced by PB are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In summary, we observe that pronouns occurring after nobody and similar items nevertheless vary in gender and/or number, rather than exhibit agree ment with a non context-dependent and non-referential item, such as nobody is in PB's view. Furthermore, if nobody or French personne is non-referential , and can be the antecedent of the SP he or if, we expect the same to be the case for nothing or rien which is equally non-referential or perhaps, in French, even less referential than personne, so to speak , since even the idea of 'body' or 'person' is lost. Yet rien does not seem to be a possible antecedent for the SP i/:
350 insufficient to establish the distinction they are supposed to establish, or else they contradict each other. Let us now turn to (A) above. We already know that non-commutability does not necessarily mean that the pronoun tested is a SP. Indeed there exist various positions in the sentence where the pronoun is not commutable, but where nevertheless it is very questionable that the pronoun is a SP. Such positions are topicalization and anti-topicalization . PB ( 1 98 3 : 1 1 8-9) says that the 'topic-shift pronoun', as he calls it, "occurs clearly referentially, as an A P", although he also states that "it is tied i n its interpretation to the subject and cannot refer to anything else, which would be a typical SP property"; PB suggests that "SPs may arise from RPs via constructions like the topic-shift construction". Consider the following examples:
·
a. La sentinelle, elle n 'a rien entendu. 'The guard (fern.), he hasn't heard anything.' b . Elle n'a rien entendu, Ia sentinelle. ' He hasn't heard anything, the guard (fern).'
Commutability is clearly outruled in sentences such as these: (3 1 )
a.
* La sentinelle, l'andouille n'a rien entendu. 'The guard, the bastard hasn't heard anything.' b . * L'andouille n'a rien entendu, I a sentinelle. 'The bastard, the guard hasn't heard anything.'
As we have seen, non-commutability does not have to mean that the pronoun is a SP. In fact, replacement with personne, 'nobody', makes the sentences ungrammatical : (32)
a . * Personne, il n'a rien entendu. b. * II n 'a rien entendu, personne.
These data seem to point toward the conclusion that the pronouns which occu r in these positions are not SPs. It seems quite reasonable to think that they are not syntactically linked to any noun in the sentence, as their only possible antecedent noun occurs rather outside of the sentence itself. This conclusion is not without problems, however. One of PB's central ideas concerns agreement phenomena: whereas SPs presumably always agree with their syntactic antecedent, A Ps agree with the gender of a 'conceptual repre sentation' they refer to (we shall get back to this in more detail below). The gender of the conceptual representation itself may be derived from the gender of the associated noun, but need not be. In fact , in cases such as French sentinel/e, the grammatical gender of the noun is feminine, but the gender of the conceptual representation is very likely to be masculine (given that a guard is most often male). This allows for a reasonable explanation of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(30)
351 superficial 'gender disagreement' between a noun a n d a pronoun, a s in ( 1 2) . above, or as in: (33)
La sentinelle s'etait endormie, il n 'a rien vu . 'The guard fell asleep, he hasn't seen anything.'
(34)
a.
La sentinelle, elle dit qu'elle n'a rien vu . 'The guard, he says he hasn't seen anything.' b. Elle pretend qu'elle n'a 'rien vu , Ia sentinelle. 'He maintains he hasn't seen anything, the guard.'
(35)
a. b.
sentinelle, il dit qu'il n'a rien vu. pretend qu'il n'a rien vu, Ia sentinelle.
??La •n
In summary, here we have a case where the pronoun presumably is an AP, yet it can only exhibit syntactic agreement, i.e. agreement with the grammatical gender of a noun, not with the gender of a conceptual representation . This is of course detrimental to the theory under review. Indeed it m ight generally be expected that pronouns which are obviously APs can always 'agree' with the concept that is in focus at the time of processing. But take the case of an object whose name is plural although the thing itself has nothing inherently plural about it. If the AP 'agrees' with the conceptual representation, it seems reasonable to expect, in PB's perspective, that it would take singular number. Yet this is in fact completely impossible; cf.: (36)
You shouldn't w�ar these horrible pants. You know *its/their size doesn't fit you.
(37)
Ces ciseaux ne parviendront pas a couper le carton sauf { *s'il est /s'ils son t } dtlment aiguises 'These scissors will not be able to cut the cardboard unless { * it is/they are } well sharpened'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
For PB the pronoun if agrees with the conceptual gender (masculine) of 'guard', not with the noun sentinelle (feminine). Returning now to sentences such as (30a) and (30b), the problem is the following. As we have seen, there are reasons to think that the pronoun which occurs in it is an AP; yet, contrary to what PB's views seem to predict, syntactic agreement (with the gender of the 'antecedent' noun) is the only possibility here (cf. (34a) and (34b) below). Agreement with the gender of the conceptual representation (masculine) is nearly i m possible (in the case of topicalization) or completely out of the question (in the case of anti-topicali zation), as in (35a) and (35b) respectively:
352 (Cf. also section 2 below.) Conversely , instances where the pronoun must be a SP, but may disagree with its syntactic antecedent, are equally detri mental. To illustrate this, let us examine Dutch examples similar to ( 1 2) quoted above, but where some of the pronouns involved are clearly SPs. Consider ( 38a)-(38c) (asterisks indicate the u ngrammaticality of the intended, corefe rential reading): ( 38)
Examples (38a) a n d (38b) respectively show that t h e first occurrence o f ze in ( 38c) must be a SP (as it is non-commutable), whereas the second occurrence must be an AP (commutable with a nominal referential expression in this position). We find it intuitively hard to believe that these two occurrences of ze i n ( 38c) are i nstances of two completely different pronominal classes, subject to completely different control mechanisms. But more importantly, here again PB's two criteria do not correctly distinguish between his classes of syntactic and anaphoric pronouns. In cases where the Dutch noun meisje ('girl', neuter gender) is the antecedent of a SP, this pronoun, if we under stand him correctly, could only exhibit grammatical gender agreement, and would not refer to the conceptual representation 'girl'; the SP could therefore only be the neuter pronoun het. Yet, as the first pronoun occurrence in (38c) shows, ze will be used rather than het. The same observation holds for truly productive Dutch diminutives (meisje is a special case, as it is morphological ly aberrant and lexicalized). Dutch diminutives always possess neuter gender: they are accompanied by the neuter form of the defi nite article and of adj ectives. Yet, if they refer to a human being for instance, the pronoun in SP-position which they control will be hij, 'he', or ze, 'she', rather than het, 'it': {39)
Het j ongetje denkt dat { hij/•?het } ziek is. 'The little boy thinks that { he/it} is ill.'
( 40)
Het oude vrouwtje vindt dat { ze/•?het } haar kinderen niet vaak ziet. 'The little old woman feels that { she/it} does not often see her children .'
Similar examples exist in French, where words which clearly seem to possess
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
a. • Het meisj e beweert dat het domme wicht ziek is. 'The girl says that the dumb child is ill.' b. Het meisje beweert dat ze ziek is, maar het domme wicht is helemaal niet ziek. 'The girl says that she is ill, but the dumb child is not ill at all.' c. Het meisje beweert dat ze ziek is, maar ze·is helemaal niet ziek. 'The girl says that she is ill, but she is not ill at all . '
353 feminine grammatical gender, such as Ia sentinelle, 'the guard', or Ia victime, 'the victim', may be the antecedent to the masculine pronoun i/ in SP-posi tion: (4 1 )
L a sentinelle dit qu' { il/elle } n'a rien entendu. 'The guard says that { he/she } didn't hear anything.'
Notice that in sentences (39)-(4 1 ), the pronoun is clearly not referential (and hence must be a SP) according to the first test which PB proposes: it cannot be replaced by a referential expression : (42)
The pronoun is also syntactic according to the second test proposed by PB: the antecedent may be replaced by a non-referential expression such as personne, 'nobody': (43)
Personne n'aime admettre qu'il n'a rien entendu. 'Nobody likes to admit that he hasn't heard anything.'
Thus PB's theory leads to the prediction that only elle could be used in sentences such as (4 1 ) : the pronouns are syntactic, and could therefore only agree with the fem in ine grammatical gender of their antecedent. This predic tion is contrary to the facts. In the sentences (39}-(41) above, the pronoun gende·r is identical to the gender of an AP with the same antecedent; cf. : (44)
Het jongetje denkt dat hij ziek is. Maar hij is helemaal niet ziek. 'The little boy thinks that he is ill. But he is not ill at all . '
(45)
Het oude vrouwtje vindt dat ze haar kinderen niet vaak ziet. Maar ze klaagt er niet over. 'The little old woman feels that she does not often see her children. But she does not complain about it.'
(46)
La sentinelle nie categoriquement qu'il se soit endormi a son poste. ll pretend avoir veille toute Ia nuit. 'The guard denies that he fell asleep at his post. He maintains that he stayed awake all n ight long.'
We shall come back to the agreement mechanisms for APs in Section 5. At this point, we feel that there ·are no convincing data which substantiate a distinction between syntactic and anaphoric pronouns along the lines drawn by PB.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
*La sentinelle dit que )'imbecile n'a rien vu . 'The guard says that the bastard didn't see anything.'
354 1. 2. Distinctions Based on Syntactic Position
(47)
a. Fred told Mary that she won. b. Fred told Mary that he won. (=PB's ( 1 43a)-( 1 43b))
PB makes the following com ment: "No consideration of factual k nowledge or context is needed in these cases to interpret the pronouns: at least in one analysis of the sentences, they are not referential, do not link up to context, but merely disambiguate and facilitate the syntactic analysis" ( 1 983: 1 2 1 )
Thus w e may suppose that i n another analysis o f these two sentences, the pronouns involved are referential , even in such a· minimal, c-<:ommanded domain. That indeed a RP (AP) can occur in this domain becomes evident in (47c), where surely factual knowledge or context is needed to interpret the pronoun: (47)
c. Fred told John that he won.
Examples similar to these show that, in any case, SP-hood cannot be defined only through reference to a syntactic domain in any straigh tforward sense of 'domain'. In his footnote 26, PB suggests that sentences like (47c) above would have t wo syntactic analyses, and that the context may determine "which syntactic analysis will be applied in the interpretation of an actual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If tests for identifying SPs fai l , it might still be thought that, perhaps, SPs are restricted to certain syntactic positions in the sentence. We shall treat this point by referring to proposals put forward by Reinhart ( 1 983), who makes a distinction similar to PB's between two classes of pronouns; her criteria for distinguishing between the two classes are syntac tic. PB's definition of the respective domains where pronouns of each of the classes IT)ay occur, on the other hand, is cast in semantic terms. Yet in most si mple cases, as PB acknowledges ( 1 983:1 87ff. and footnote 1 47), both types of definition lead to co-extensional domains. As the syntactic definition is rather well known, we shall restrict ourselves to cases where the domain defined on a syntactic basis is identical to that of PB's semantic definition. For Reinhart,
355 utterance of the sentence"; yet it seems to us that the two syntactic analyses involved can only differ in coindexing, and that the actual syntactic domain as defined on the basis of c-<:om mand remains the same. One could perhaps still argue that only SPs are to be found in positions that are c-commanded by a coreferential NP; in other words that, given coreferen tiality, the c-commanded expression is always an SP. This is, however, certainly false in view of PB's criteria. Indeed, given the required context, the following sentence is grammatical, with coreference between Pierre and the nominal anaphoric expression cette triste andouille in c-<:om manded posi tion: (48)
Compare in this respect the following examples: (49)
Le comit.e a dit que ce ramassis de vieux crabes restera au pouvoir. 'The committee said that. this bunch of old crabs will stay in power.'
(50)
Le comite a regrette que ce ramassis de vieux crabes reste au pouvoir. 'The committee regretted that this bunch of old crabs stays in power.'
(5 1 )
L e comite s'imagine que c e ramassis de vieux crabes restera au pouvoir. 'The com mittee imagines that this bunch of old crabs will stay in power.'
(52)
Le comite a decroche que ce ramassis de vieux crabes reste au pou voir. 'The committee managed to obtain that this bunch of old crabs will stay in power.'
In (49) and (50), a coreferential interpretation will be discarded at once; not so, however, in (5 1 ) and (52). Indeed, in (49) and (50), because of the main verbs dire and regretter, one tends to consider the subject of the utterance, /e comite, as the source· of the description ce ramassis de vieux crabes; as it is extremely unlikely that the committee will qualify itself with this expression, the two N Ps will be interpreted as disjoint in reference. In (5 1 ) and (52), on the other hand, the verb of the embedding clause either indicates that, to the speaker, the committee is wrong with respect to the content of the embedded clause (s'imaginer), or else the verb may indicate a subjective appraisal of what is going on by the speaker (decrocher). Thus in these latter two cases, the utterer hi mself will be taken to be the source of the qualification ce ramassis
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Pour rna part, je crois bien que Pierre s'imagine vraiment que cette triste andouille est parvenue a tromper le Conseil. 'As to me, I believe that Peter really thinks that he managed to fool the Board .'
356
de vieux crabes as a description for the committee, and coreference is possible, although the nominal referential expression ce ramassis de vieux crabes is ccommanded by /e comite. True enough , Reinhart ( 1 98 3) only states that the bound-variable interpre tation is possible when the pronoun is c-com manded by its antecedent, not that it is necessary; in this position, she says, SPs may be identified by the sloppy identity phenomenon. Many more examples could be adduced to show that the configurations the referential expressions appear in are not the ultima ratio for the AP/SP distinction. Compare: * Nobody admitted that the poor chap was ill.
(54)
*Nobody will pretend being in good health , since the poor chap is on the verge of dying.
(55)
?Nobody will leave these premises unless the lucky bastard is a close friend of the security officer.
Here it becomes clear that nobody cannot normally be the antecedent to a nominal referential expression, simply because of the non-referentiality of the item nobody itself, and regardless of whether the referential expression occurs in a domain that is c-commanded by nobody (as in (53)) or not (as in (54)). If on the other hand the context renders the hypothesis of 'someone' more likely, sentences similar to these may become less ungrammatical (as in (55)). Indeed, in the last pages of his book ( l 98 3 :2 l l ff.) PB seems to recognize that the directly relevant domains for pronoun control mechanisms are not to be defined syntactically. He suggests that the domain in which the pronouns must be bound is in reality some kind of discursive interpretation unit, which he calls "processing unit" (PU): a SP must have its antecedent i nside the PU, whereas an AP does not have an antecedent, neither inside nor outside the PU. Recourse to an AP indeed corresponds to the choice of referring to a conceptual representation. We maintain that in either case (SP and A P) however, linguistic agreement with an intervening noun is involved, and that no true (=non-deictic) pro noun can refer to a conceptual representation without taking this noun into account, be it inside or outside a PU (cf. Section 2). We are still left with the fact that a model which treats in a completely different fashion pronoun occurrences and agreement phenomena in sen tences such as (56a) and (56b) respectively, or both pronoun occurrences in a sentence such as (56c) seem s counterintuitive as well as uneconomical: ·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(53)
357 (56)
a. Pierre dit qu'il ne reviendra pas. 'Peter says that he won't come back. b. Pierre ne reviendra pas, sauf I s'il/si le triste sire } est fauche. 'Peter won't come back, except if { he/the poor chap } is broke.' c. Pierre dit qu'il ne reviendra pas, sauf s'il est fauche. 'Peter says that he won't come back , except if he is broke.'
1. 3. A lleged Morphological Evidence for Distinct Pronoun Classes If there exists a primary distinction between SPs and RPs, and a secondary distinction within the RP-class between APs and DPs, one might perhaps expect that the major distinction, between SPs and RPs, would be most often reflected in morphology. In Indo-European languages, as PB admits ( 1 983: 106- 1 1 3), this is not the case. Various Indo-European languages (e.g. French) make a morphological distinction between deictics and other pronouns (cf. Tasmowski and Verluy ten 1982); morphological distinctions between RPs and SPs are much harder to come by. The only more or less convincing cases which could be adduced are morphologically marked reflexives. These, according to PB ( 1 983: 1 091 1 0), can only be SPs; and it is true that the tests for distinguishing SPs most often place reflexives in the SP-class. We would like to maintain, however, that reflexives are not always as clearly SPs as one might expect. Firstly, these "typical" examples of SPs often exhibit little or no agreement characteristics: singular and plural, masculine and feminine, even accusative and dative, are expressed by one and the same form in many languages (French, Italian, Spanish , Portuguese); in other languages (e.g. Russian), the reflexive pronoun is even uninflected for person. Furthermore, it is clear that the morphological distinction between re flexives and other pronouns does not i n itself correspond to the SP-AP distinction; otherwise we would expect a reflexive form in sentences such as the following, where the pronoun is a SP and where se and Jules are coreferen tial: (57)
• Jules
n'aime pas qu'on se regarde.
Yet this sentence is of course ungrammatical on this reading, since the reflexive pronoun should normally be argument-bound in its S (cf., however, Kuno 1 983). On the other hand, a non-reflexive pronoun, even when taken to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I n (56a), the pronoun would be a SP, in (56b) a AP; i n (56c), the first occurrence of if would be a SP, the second an AP. Nothing, to our knowledge, substantiates such a distinction. In summary, we believe that it is impossible to identify PB's class of SPs by syntactic position, based on the notion of c-command, or by any other definition which leads to a similar extension of the domains involved.
358 be a S P , need n o t exhibit gender agreement with its c-commanding antece dent, as was shown in examples (38a)-(38c) above. If agreement is indeed a property that is typical of SP-hood, we would be left with a rather uninterest ing result: the only SPs would then be the reflexive pronouns of some particular languages. In fact, the SP-AP distinction and the morphological reflexive-nonre flexive distinction clearly should be independent of each other. Next, consider the following sentences: II ne faut pas travailler pour soi, mais pour les autres. 'It is advisable to work not for oneself, but for the others.'
(59)
Mieux vaut se laver le soir que le matin. 'It is better to wash oneself in the evening than in the morning.'
In (58) and (59) soi and se are reflexives and hence they supposedly would be clear occurrences of a SP. Yet, this SP does not have an antecedent in the · sentence (the main clause contains an impersonal verb the subject of which is if, 'it'). This is contrary to the characteristics of SPs as PB defines them. It is totally unclear how PB's Syntactic Agreement ·Principle (SAP; cf. PB 1983: 1 99-200) might be applicable here. Recall that a SP exhibits syntactic agree ment with an antecedent. According to the SAP, this antecedent must occur within narrowly defined syntactic limits; but in the present examples there may not occur any syntactic antecedent within discourse at all. In sum mary, in Indo-European languages there are no clear-cut instances of a morphological differentiation between SPs and RPs; in particular, all A P-forms tend to equally well occur in S P-positions. This might be due, says PB, to the fact that in Indo-European the finite verb is itself marked for subject-agreement, which m akes SPs superfluous in this function ; subject pronouns therefore would naturally only arise, historically, in emphatic contexts, i.e. in contexts where R P-forms are used (PB 1 98 3 : I l l ) . The situa tion may be different in other language families. The only further examples which PB adduces belong to three related West-African languages: Yoruba, Igbo and Edo. In Yoruba, two morpholo gically distinct pronouns can be used in a sentence which may be translated as follows : (60)
Ojo thinks that he is clever.
If the pronoun is a SP, controlled by the antecedent Ojo, the form oun wi ll be used in Yoruba: if the pronoun refers to someone else, the RP-form o must be used. The examples from the two other languages are identical. PB points out, however, that the SP-forms (oun in Yoruba) are also the ordinary D P-forms in the respective languages. It seems to us that this casts serious doubts upon the analysis of these forms as SPs. PB does not offer any
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(58)
359 examples where he has applied his two tests for S P-hood to these pronouns; nor does he show that, in contexts where the pronoun would refer to the subject Ojo, but would not be in an SP-position, the RP-form o will be used, and not oun. This could be tested in a sentence such as the following: 61)
2. PRONOUNS AGREE WITH NOUNS, NOT WITH CONCEPTS
APs, it is claimed by PB, directly refer to a conceptual representation with the gender of which they agree. But we find this idea of 'conceptual gender' quite puzzling. PB is not very specific when it comes to the description of the mechanisms through which gender is transferred onto the conceptual repre sentation. He states that "in gender languages, one of the factors that influence the conceptual gender of a referent in an unmarked context, is the grammatical gender of the common noun denoting the basic category of the referent" ( 1 984:7). In our mind, this involves a category-mistake. Gender is a grammatical category, and we do not see how concepts may possess it, nor how it could be transmitted from a noun to a concept. Let us accept that the conceptual representation of a referent may vary in function of the noun which is, in a particular context, associated with the concept: the conceptual representation of the object TABLE will not be the same when the associated noun is Ia table in one context, /e meuble in another context. But it is much harder to imagine how the two different conceptual representations possess, respectively, feminine and masculine gender. More i mportantly, if the gender of the conceptual representation is indeed most often derived from the associated noun, as clearly is the case, and as PB admits, it follows that the gender of this conceptual representation is subject to the same arbitrary variations as its associated noun may be. Consider the following cases: (i)
There are languages where, for the same noun, gender varies with grammatical number. In Rumanian for instance, there is a class of nouns which are masculine in the singular (e.g. scaunul, 'the chair'), feminine in the plural (scaunele, 'the chairs'). Saying that the conceptual representa tion associated with one and the same noun is masculine in the singular, and feminine in the plural, seems to us devoid of any meaning; yet this is the claim towards which such a position will lead.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Ojo thinks that h e (=Ojo) i s clever; but h e isn't really as clever as he thinks. PB predicts that in (6 1 ), the first occurrence of he corresponds to Yoruba oun (SP), the second occurrence to o (RP). A native informant could clarify this matter, and we are curious to know his answer. As things stand, we do not feel that sufficient morphological proof for the SP-AP distinction is available; we still question the existence of a clear distinction between SPs and other pronouns.
360 (ii) There are languages where the gender of some nouns is subject to arbitrary fluctuations. This is the case, e.g., in French apres-midi, 'after noon', chromo, 'picture', entrecote, 'sirloin steak', interview, 'interview', etc. (cf. Grevisse 1 975: par. 273). One would therefore have to conclude that un apres-midi (masc. ) and une apres-midi (fern.) correspond to (be it slightly) different conceptual representations, and similarly for the other examples. Again, we completely fail to see what this m ight mean.
(62)
Quand Ia baleine sortira de l'eau, je Ia tuerai. ' When the whale shows itself, I'll kill it.' b. Quand Ia baleine sortira de l'eau , je harponnerai le monstre pour { */a//e} tuer. ' When the whale shows itself, I'll harpoon the monster to kill it.' (examples from Fauconnier I 974) a.
If our model is indeed ceteris paribus preferable for reasons of simplicity of the underlying hypotheses, the burden of proof rests upon our potential opponents; they have to show that our model is descriptively inadequate. PB attempts to do so by arguing that our claims are not descriptively adequate and that, in various instances, they make incorrect predictions concerning the shape of the pronoun that appears. We shall now attempt to show that our model can successfully explain various examples that were brought forward by PB (some of them at the talk he gave at the Belgian Linguistics Circle meeting in Brussels, December 1 984). In cases where one perceives an object which is so indistinct that it cannot be named, PB argues, one will use the singular form of the pronoun and say, for instance, (63), but not (64): ( 63)
{ Do you see } it/that?
(64)
*Do you see them?
Clearly, the 'nu mber' of the pronoun here cannot be derived through agree ment with the number of a noun since, in the situation at hand, the assump tion is that no such noun is available. In fact, examples such as these are treated in our 1982 article. T-o us, in a situation such as this, only a deictic can
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(iii) Finally, if the gender of the AP is indeed determined by conceptual gender, one would expect that the occurrence of a nominal anaphoric expression between the pronoun and the noun which corresponds to the concept could not have any influence on the pronoun's gender. Nominal anaphoric expressions usually do not add any new information to the (conceptual representation of the) referent, in particular when they are used with the definite article (cf. PB 1 98 3 : 1 50- 1 5 1 ). Yet this expectation is contrary to the facts; cf.:
36 1 be used, and not a true pronoun . We claim that deictics do not fall under the l inguistic control mechanism as outlined in (6). Deictics do not agree in grammatical gender and/or number with an antecedent noun. If they exhibit something that looks like gender and/or number, they derive it from the properties of the real-world object they refer to (we should therefore call the property 'sex' rather than gender, and use the term 'agreement' between quotes to refer to such cases); cf. also the examples (9)-( 1 1 ) above. The fact that the pronouns used in sentences such as (63) and (64) are deictics becomes clear in French, where deictics are morphologically distinct from true pro nouns in this position; cf. : Tu vois <;a? 'Do you see that?'
(66)
*Tu { lelia} vois? 'Do you see it?' (cf. Tasmowski and Verluyten 1 982:3 30)
Another type of example which PB puts forward involves the following situation : a hunter hears a noise behind the bushes, and fires his shotgun, in the hope of hitting the boar he was tracking; but in fact, he hits the forest ranger who happened to pass behind the bushes; the hunter then declares: (67)
When I fired at the ranger, I thought { itl* he } was a boar.
The sentence with he is of course grammatical, but does not have the meaning the utterer wants to convey (the same holds for (68) below with if). PB claims that the neuter gender of the pronoun can only be explained by the concep tual representation of 'the ranger' in this particular situation, i.e. the fact that the hunter thinks it is a boar; the noun ranger in itself, if it were the antecedent of the pronoun, could not explain why the sentence with he is infelicitous. But our model does not force us to say that, if a noun is present within discourse, it must necessarily be the antecedent to the pronoun which is used. In fact, here again, French examples make clear that a deictic is used, not a true pronoun: (68)
Quand j'ai tire sur le garde-classe, je croyais que { c'etait/i/ etait} un sanglier. 'When I fired at the ranger, I thought { it/he } was a boar.'
We simply have to say that in examples such as these, the noun which is present in the sentence does not function as an antecedent to the 'pronoun' (in fact a deictic). As the context makes clear, at the moment of the gunshot the speaker had not recognized the object he was firing at. Therefore no name was available to mention this object, and no true pronoun could be used; only the deictic ce is possible (cf. Tasmowski and Verluyten 1 982). Another argument put forward by PB involves what he calls 'inferred antecedents'. Consider the following examples:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(65)
362 (69)
a. The car came racing round the corner although the lights were on red. I suppose he must have overlooked them. (= example ( 1 3) in PB 1984:7). b. The driver came racing around the comer although the lights were on red. I suppose he must have overlooked them.
(70)
Whenever the driver approaches an intersection, he m ust be particu larly careful.
(7 1 )
A scientist should be modest; h e does not know the answer t o all our questions.
Sentences such as these abound everywhere, of course. (70) could be found in a preparatory course for a d river's license test. It would be far-fetched to suggest that the author of the manual necessarily conceptualized the driver as being male each time he wrote a sentence with the pronoun he referring to the driver, whereas, in fact, half of the students who study the course and take the test are female. This is not to deny that there may be some kind of 'subliminal' suggestion to the effect that a male-dominated model is i mposed upon everyone in our society. But that is due to other factors (the fact that the pronoun he is indistinctly used for reference to males as well as to humans in general may suggest an identity of those two groups). In denotation and in grammar - pace women's lib - sentences such as (70) and (7 1 ) do not (necessarily) involve the conceptual representation of a male driver, or scientist. In fact, PB's position forces him to state that the conceptualization of a referent is always necessary for the felicitous use of an AP, even in a sentence where the only possible antecedent is non-referential, such as: (72)
No-one is allowed into this building, unless he can provide a valid identification . (PB 1 984:7)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In cases such as (69a), says PB, "the referents of the pronouns are hinted at in the preceding text but are none the less clearly expectable, which implies that a conceptual representation is formed that includes a male driver of the car (. . . )" ( 1 984: 7). In our theory; we would say that the pronoun he possesses an absentee antecedent driver which is made salient metonymously through the presence of the car; we refer to our 1 982 article for further examples and argumentation. But in any case, one would have to say that the formation of a conceptual representation is equally necessary in sentences such as (69b ), where the pronoun he, just as in (69a), can only be an AP, not a SP. We seriously question the conclusion that a noun such as the driver cannot in itself function as the antecedent to a prorioun he, and that a conceptualiza tion of the situation is always necessary. Consider the following sentences:
363 The use of he in (72) i mplies, according to PB, "that a conceptual representa tion is formed that includes ( . . . ) a hypothetical visitor to a particular building (. . . )". Thus, if sentence (72) appears on a sign which is posted on the door of the building, the author of the sign, as well as everyone who happens to read it, would form a conceptualization of a visitor in his m ind, including the visitor's (male) sex? This seems to us hard to swallow. It is clear that such a sign, in reality, is equally applicable to a female who reads it, and we challenge any woman without a valid identification to walk past the guard ! Or should we say that a 'female' conceptuaJ representation may be referred to with the pronoun he in this case? But that is clearly out of the question when the antecedent is present in the sentence: A female will never be allowed into this club, unless { *he/she} is accompanied by a male escort.
Sentences such as these are, again , clear instances of the role of the interven ing noun in pronoun agreement On the other hand, PB is forced to say that, if for some reason it is impossible to form a complete conceptual representation (which must at least include the sex of the referent); an AP cannot be used felicitously. This is contrary to the facts. Sentences such as (69) and (70) are completely gramma tical and felicitous when they are used in a situation where it was impossible to see whether the driver was male or female (perhaps because the car went by too fast). In su mmary, a model along the lines of PB's appears to be descrip tively inadequate in some instances, and unnecessarily complicated in any case.
3. ON GENDER AGREEM ENT
I n ourprevious papers, we did not address the question as tow here a noun gets his gender from . In view of some of the examples brought forward above, we would like to offer some suggestions here with regard to this problem. As we said in our introduction, the fact that other authors (Reinhart , Bosch) turned their attention exclusively to examples involving humans has prevented them from becoming aware of the full range of data and has biased their conclusions. Examples involving agreement between a pronoun and a noun referring to an object are the only ones which allow for a potential test between two alternatives: (i) agreement between the pronoun and the antecedent noun, and (ii)coreference between the pronoun and the noun, which would both refer to the same object. .C onsider a part of discourse in which the French noun Ia chemise, 'the shirt', occurs. If its occurrence is then followed by the neuter pronoun ra, this might be considered as an argument in fa vourofthe hypothesis that the pronoun ra relates totheobject SHIRTratherthan tothefeminine noun chemise; if on the other hand, as indeed is the case, normally the fe minine
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(73)
364 pronoun elle will occur, this must be taken as an argument in favour of direct agreement between the pronoun and its antecedent noun (only this noun can explain the fem inine singular gender of the pronoun); cf. : (74)
a. b. a. b.
Cette chemise-la est tresjolie. Oui, mais { elle/*�} n'a pas de manches. This shirt is quite nice. Yes but { it(fem.)/it (neuter) } has no sleeves. ,.
(R RO) Use phonetic information in order to determine gender. E.g. in Ruma nian, words ending in a consonant are never fem inine. (RR 1) Use morphological information in order to determine gender. E.g. in French , words in -age, -ment are normally masculine (/'embal/age, leroulement, words in -ation feminine (Ia liberation); in Dutch , diminutives are normally neuter (poesje tafe/tje, etc.). ,
(RR2) Use semantic information to determine gender: i n Latin , names of trees are normally feminine, in French they are normally masculine; in French, names oflanguages are always masculine, names ofsciencesnearlyalwaysfeminine(/a physique, Ia biologie, etc.). (Cf. Grevisse 1 9 74, §266ff. ). (RR3a) Use pragmatic (knowledge of the world)information : nouns normally referring to male beings are masculine, nouns normally referring to female beings are feminine. E.g. the president, /e docker, latin nauta, agricola are masculine; midwife, sage-femme, vroedvrouw (Dutch) are feminine. (RR3b) Use pragmatic (situational) information : a noun referring to a male beingis masculine,a noun referring to afemale beingisfeminine. E.g. the English n oun driver may be masculine or fem inine depending on the pragmatic (situational) information available; similarly for French un/une enfant. ( RR4) In the 'elsewhere' case, [+animate], or perhaps [+human] nouns are masculine (e.g. driver, enfant when R R3b fails).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
When the refe rent is [+human] , such a test is not possible, as we pointed out above. Now let us h ave a general look at gender marking and gender agreement. The mechanisms for marking grammatical gender are, ofcourse, not necessarilythe same in all languages. Intuitively, it seems that gender marking is lexical in almost all nouns in some languages (e.g. Latin, French), much less so in others (e.g. English). But in any case, a setofredundancy rules ( R Rs) based on various strategies for recovering grammatical gender seem to be at work; some of these may be lexical, others not. The following RRsareobviously operative in various languages:
365 Generally speaking, theories such as PB's and Reinhart's only accountforcases which fall under R R 3 or R R4. Indeed they only cover those examples, e.g. involving humans, where according to the R Rs given above non-lexical information can be invoked for determining the gender of the noun. Of course, we do not claim that within one language all lexical items necessarily fall under one of these lexical redundancy rules. In some (or, in certain languages, in many) cases, gender has to be specified in the lexicon. Thus the word kat 'cat' seems to be feminine in (Southern) Dutch, even when it refers to a male cat; (RR4) seems to be restricted to [+human] nouns, and is not applicable here. The following sentence is grammatical in the situation described: (75)
The precise way i n which the R Rs are used is language-specific. Thus, i n Northern Dutch the application o f R R 3 i s often restricted t o humans, and RR4 applies to non-human animate nouns, as i n : (76)
a. Onze koe is ziek. Daarom geeft hijgeen melk meer. 'Our cow is sick. That's why he doesn't produce milk any more.'
In Southern Dutch on the other hand, RR3 is applicable to all animates: (76)
b. Onze koe isziek. Daarom geeftzegeen melk meer. 'Our cow is sick. That's why she doesn't produce milk any more.'
One RR may override another. I n Dutch, RR3 often overrides R R 1 when a diminutive noun is animate, as in: (77)
Het arme jongetje is ziek. Daarom heeft { hij/*?het } geen honger 'The poor little boy is ill. That's why { he/it} is not hungry.'
The Dutch noun meisje ('girl') is a special case: historically it isadiminutive, but synchronically it is not clearly related to a base noun, and morphologically it is slightly aberrant as a dimin utive. This may explain why it seems even harder to use the neuter pronoun het with meisje than , forinstance, in (77)above. In Latin, RR2 overrides RR 1 in a noun such as populus 'poplar' (name of tree) which is feminine although it exhibits a morphologically masculine ending; RR3 overrides RR 1 in agricola 'farmer', which is masculine although it exhibits a typically feminine morphological ending. Sometimes R Rs go back to a historical situation which has ceased to exist. Thus the English noun nurse is still felt to be feminine (historically RR3a was operative), although nowadays there are male nurses (and thus RR3b may say
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(our male cat is miawing in front ofthe door.) La at ze toch binnen ! 'Let her in ! '
366
(78)
Le nouveau ministre a promis qu' { illelle } prendrait les mesures necessaires. 'The new minister prom ised { he/she } would take the necessary mea sures'.
(79)
Le nouveau m inistre n'est pas venu a Ia reception. On dit qu'{ elle/?il etait}malade 'The new minister has not come to the reception. It is said that { she /he } is ill.'
In (78), where the pronoun is in SP-position , i/is at least possible; in (79), where the pronoun is in AP-position, e/le seems strongly preferred . This appears to corroborate PB's proposals: the SP agrees with the grammatical gender of the noun (masculine in this case), the AP with the 'gender', if one may say so, of the conceptual representation (which is feminine here). Yet other examples go against this conclusion; cf. : (80)
Le ministre des Affaires Sociales a confirme, dans une longue interview accordee au journal ' Le Monde' et publiee dans l'edition du 27 marsde ce quotidien, qu' {elle/?i l } prendra les mesures necessaires. 'The minister ofSocial Affairsconfirmed, i n a longinterviewgiven to the newspaper 'Le Monde' and published in the 27 March edition of this newspaper, that { she/he } will take the necessary measures.'
(81)
J e n 'ai plus d'entrevues avec l e m inistre, { illelle } (est trop occupe(e) pour me recevoir. 'I have no meetings anymore with the minister, { he/she} is too busy to see me.'
PB's distinction fails, here again. In (78) and (80), e/le is possible although the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
that it must be masculine). In some cases, the conflict between different RRs is unresolved in the current state of the language, and it may become the object of long discussions in normative grammar. In French, words such as docteur, 'doctor', or ministre, ' minister', are masculineon the basisofRR 1 and/ or R R3a; their masculine gender is usually kept in article and adjective agreement. Yet they may nowadays refer to a female person, which would make them feminine according to RR3b. There is disagreement as towhich pronoun , i/ore//e, is more felicitous with such nouns (assuming they refer to a female) (cf. also Mok 1 968). Notice that it may bethecasethatsomeofthe morphologicalagreementfeatures exhibit masculine gender (e.g. the article /e in /e ministre), whereas some others are undecided (the pronoun in this case) or feminine (such as the pronoun with Dutch meisje) . It might appear at first sight that PB's distinction between SP-agreement and AP-agreement is operative in sentences with words like /e ministre. Compare the following sentences:
367 pronoun is in SP-position; in ( 8 1 ) i/ is possible in A P-position. Both facts go against PB's predictions. With the Dutch noun meisje, the article and epithets exhibit neuter gender (het meisje; een mooi meisje); the preferred pronoun on the other hand is ze: (82)
Het anne meisje is ziek. Ze gaat niet naar school vandaag. 'The poor girl is ill. She doesn't go to school today.'
(83)
(standing in front ofa sophisticated deadly weapon) { Cela/�I n 'est pas a mettre entre tou tes les m ains. 'This is not to be left intojust anybody's hands. '
Th e situation is completely different with humans. On the one hand, in this case, the gender of the pronoun with absenteeantecendent (i.e. without an antecedent which is present in surrounding discourse) is without problems: it must always be i/for male h umans, el/e forfemales; cf. for example: (84)
(standing in front of a male road-accident victim , Ia victime, feminine) { 11/•elle I semble a voir lajambe fractun!e. ' { He/• she I seems to have a broken leg.'
the other hand, using deictic ce/a orra is always i mpossible with humans , at least without introducing particular connotations; compare:
On
(85)
a. b. c.
l
)
Cette femme me connaitre. Elle semble • cela, • ca ' { This woman/she/that } seems to know me.'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In fact, in (78)-(8 1 ) the preferability of using el/e increases with the distance between the pronoun and its antecedent (cf. Boel 1976:64-67; cf. also Corbett, quoted in PB 1 983, note 1 5 1 ); example (82) issimilarto these. It is hard to believe that this would be due to a change in the gender of the antecedent itself, that e.g. mi� istre would first be masculine, and then turn feminine. We think that our 1 982 proposalsabout 'absenteeantecedents'can be extended to provide a natural explanation for this phenomenon. As we pointed out then (1982, note II), the unmarked name referringto an object,asopposed to a person (i.e . , the only name that may serveasan absenteeantecedent)mayvary, and the identification of the proper name of an object i n a given situation , for a given speaker, is notalwayseasy. A fisherman may refertoa trout as/a truite(and if this name serves as an absentee antecedent, the pronoun will beel/e),a layman in the field will call the same referent/epoisson , 'the fish' (and then theequivalent pronoun will be if). Ontheotherhand,in theappropriatecontext, objects may be referred to by a deictic such as French eela or ra.
368 On the basis of observations such as these, two hypotheses seem possible to account for pronoun agreement phenomena with h umans: (i)
the pronoun is controlled by the conceptual representation of the [+ human] object, and 'agrees' with its sex, male or female; (ii) the pronoun is controlled by a linguistic antecedent; this antecedent is homme orfemme by default if no other suitable candidate is found.
(86)
Alors ton docteur, elle dit que tu n'as rien du tout? 'So your doctor, she says you haven't got anything?'
(87)
Alors elle dit que tu n'as rien du tout, ton docteur? 'So she says you haven't got anything, your doctor?'
We claimed above that items such as docteur, ministre are masculine (cf. the use of the article /e) and that the use of the pronoun elle in such cases is only possible when the distance between it and the noun is beyond a certain min imal treshold. Now it seems to be the case that with a limited class of names of e.g. common social functions in our everyday surroundings, such as docteur, dentiste and the like, the referents of which are specifically known by the discourse participants, it is the very designator of the referent that may serve as a trigger for the representation of the object. I n this particular case then, the specific saliency conditions we proposed (cf. the introduction and our 1982 paper) may be overruled, as saliency is ensured otherwise. Once the representation oft heo bject is cognitively present, the absentee antecedent mechanism, operating with the default nounfemme as outlined above, is free to operate. Again, this case is only possible with [+human] referents. Returning nowtoexamplessuch as (80), itseemsreasonableto assu methatthe search for an antecedent-with-gender which is present in discourse does not go
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As we pointed out in the introduction, weknowofnoempiricalevidence i n favor of either (i) or (ii). The predictions that are made arethesamein both cases, as the occurrence of the pronoun il or elle may be considered toexhibit 'sex agreement' with the male or female conceptual representation, or else gender agreement with a masculine resp. feminine nou n . Weoptforhypothesis(ii)on model-inter nal grounds: we have shown that linguistic control of the pronoun is the correct hypothesis in the case of objects; in the absence of empirical evidence to the contrary, we extend this hypothesis to cover those cases where human referents are involved. Before moving on, we wish to treat a possible objection to the proposals we outlined above. Recall that, as we pointed out in Section 1 . 1 , topicalization and anti-topicalization (examples (35a) and (35b) trigger grammatical gender agreement ofthe pronoun with the noun thatispresentin discourse(sentine//e in the examples we adduced) is possible. Examples such as thefollowingseem to go against this clai m :
369
(88)
Quellejolie lampe ! Je voudrais bien que tu me l la/*le } donnes. 'What a nice lamp ! I wish you would give it to me?
(89)
(looking at the lamp, Ia lampe, feminine) Tu me l la!*le} donnes? ' Would you give it to me?'
The fact that hu man beings have a special status in this respect is, ofcourse, not particularly surprising.
4. CONCLUSION
In summary, we conclude that (i)
distinctions into pronoun classes as proposed in the literature, be it on semantic, syntactic or morphological grounds, a re not tenable, and our unified theory ofpronouns stands as it is;
(ii) agreement in pronouns is with a linguistic item (the antecedent), and not with a conceptual representation, in all cases; (iii) for objects, in the case ofan absentee antecedent,agreement ofthepronoun . is with the proper name of the object, regardless of distance; there is no
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
beyond a certain distance; beyond that, the pronoun is assumed to becontrolled by an absentee antecedent, and for the reasons explained above, this pronoun can only be i/for males, e/le for females. In (80),/eministreislexically masculine, and agreement items such as the article, adjectives, etc . , which areclosetoit, will exhibit masculine gender: /e nouveau ministre, 'the new minister', etc. The chances of having the pronoun el/eoccurringincreasewith the distance between it and the NP in question , as observed before. In case el/e occurs, we assume that the distance between the possibleantecedent/eministreand the pronoun was too great to be bridged, and the speaker/hearer then takes the pronoun to be controlled by the feminine absenteeantecedentfemme. Examples suchas(78)or (8 1 ) can be explained i n the same fashion. The precise extent of this 'critical distance' remains to be defined (for some data and preliminary work, cf. Boel 1 976:64-67). The notions of distance and of acquaintance with the referent play no role whatsoever if the antecedent is an object rather than a human. As there is no default proper name for objects, only the specific proper name of the object considered can determine gender for any agreement items, regardless of distance, and regardles of whether the noun which serves as the name for the object is present in discourse or not (in the latter case, we have an absentee antecedent, as in (89)):
370 default proper name for objects. For humans, agreement is with the specific name, depending on availability and distance; or else with the default proper names homme for males andfemme for females (in French). We feel that the main lines of our 1 982 model are confirmed by the discussion of alternatives and by the evidence we adduced in this paper. Dept. ofRomance Philology, University ofAntwup (U. l.A.). Univusitdtsplein I,
B-2610 Wi/rijk. Belgtum
Boel, Else, l 976: Le genredes noms RevueRomane X I , I : 1 6-73. Bolinger, Dwight, 1 979: Pronouns in discourse. I n : T. Givon (ed.), Syntax and Semantics 1 2, Academic Press, New York, 289-309. Bosc h , Peter, 1 98 3 : Agrumem and Anaphora. A Study of the Role of Pronouns in Syntax and Discourse. Academic Press, London/New York. Bosch, Peter, 1 984: Coherence and cohesion (Comments on Roger van de Velde's paper). International Conference on Text CohesiOn, Bielefeld October 1 984 (to appear in the Procee dings). Cornish, FranCis, 1 986: Anaphora and semantic interpretation (review article on T. Reinhart's book with the same title). Ungua 68: 241-256. Fauconnier, Gilles, 1 974: LA Cor�Rrena. Seuil, Paris. Grevisse, Maurice, 1 975: u Bon Usage. Duculot, Gembloux. Kuno, Susumo, 1983: Reflexivization m English. In: L. Tasmowski and D. Willems (eds.), Problems in Syntax, Ghent, Communication and Cogn ition, Plenum, pp. 257-272. Lasmk, Howard, 1 976: Remarks on coreference. linguistic Analysis 2 . 1 : 1 -22. Mok, Q. 1 968. Contribution a rltude du Cat�gones Morphologiques du Genre tt du Nombre dans le Fram;ais Pari� A ctuel. Mouton, The Hague. Reinhart, Tanya, 1 983: Coreference and bound anaphora: a restatement of the anaphora questions. Linguistics andPhilosophy 6: 47-88. Tasmowski, Liliane, and S. Paul Verluyten, 1 98 1 : Pragmatically controlled anaphora and linguistic form. Linguistic Inquiry 1 2: 1 53-4. Tasmowski, Lihane, and S. Paul Verluyten, 1 982: Linguistic control of pronouns. Journal of Semanucs I : 323-46. .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
REFERENCES
Journal of Semantics 4 : 3 7 1 -387
DEFINITE VS. INDEFINITE I NTERPRETATION OF RUSSIAN NOUN PHRASES : A PROPOSAL OF A FORMAT FOR COMPLEX EVALUATION RULES•
CHRISTA HAUENSCHILD
ABSTRACf
I.
INTRODUCfiON
I should like to start this paper with a question that might seem trivial at first sight: why is PROCESSING REFERENCE at all an interesting topic? I think that it is extremely interesting just because, in natural-language utterances, referential properties and relations are very often (perhaps even typically) • A first version of this paper was read at the conference "PROCESSING REFERENCE" at Cleves (August, 1 985). I am obliged to the participants of that conference for many useful comments. I n the first place, however, I thank my former colleagues of the project Con 'Tra (Constance Concept of Context-Oriented Translation), Klaus-Jilrgen Engelberg, Siegfried Knllpfler and Peter Pause, for five years of really fruitful collaboration. It is nearly impossible to delimit their ideas from mine, and so ( shall not try to do so.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This paper aims at an illustration of a special type of linguistic rule that has been developed for a text understanding model Within a project of (partial) simulauon of the processes of understan ding and translatmg nat ural-language texts. The rules in question are to reflect different degrees of probability with respect to certain interpretation hypotheses on the basis of the interaction of different types of l inguistic as well as non-linguistic information relevant to a given Interpretati on hypothesis. This type of relative rule for (partial) interpretations is exemplified with a paradigm problem of machine translation: the definite vs. indefinite mterpretation of Russian noun phrases, which normally do not contain any locally una m biguous expressions of definiteness or indefiniteness (as e.g. articles are). Given the enormous complexity of the whole problem of interpreting such noun phrases (all kmds of lingUistic and extra-linguistic cntena may have to be considered), I concentrate on the case of couples of noun ph rases with identical heads. The interpretation hypothesis to be evaluated is the following: the two noun phrases in question are identical in reference, and thus the second is definite. The main task consists m determining the factors that yield relative posiuve or negative evidence for the given hypothesis. After a short discussiOn of the underlying text understandmg model and of the definitions of the basic concepts (anaphoricity, identity of reference, definiteness/indefiniteness, sentential and textual themeness), I give the list of single factors that are relevant for the interpretauon task m question. Then the corresponding single-factor rules are combined in a global evaluation rule, the functioning of which is illustrated with the aid of a simple example. Finally, I discuss the merits and shortcomings of the proposed format for complex evaluation rules as well as the potential role of such rules wuhin a machine translation system.
372
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
expressed in an i mplicit way. F o r example, w e do n o t say: "There was a man walking down the street. The man I j ust mentioned . . . " or " . . . that I mention ed in my first sentence wore a raincoat and a brown hat . . . " ; but we continue: " He wore a raincoat and a brown hat . . . " ; for we are sure that, in every normal situation , our addressee will be able to infer the intended interpretation, n amely that a man and he are to refer to the same individual in the given sequence of sentences. It is well k nown t hat in many languages even such basic referential proper ties of noun phrases as definiteness and indefiniteness are normally not expressed in an explicit way (i.e. by locally unambiguous signs). Latin is an example of this type, as wel l as most Slavic languages. • This situation leads t o a far less trivial question : how can the strategies of i nterpretation that enable h earers and readers to find out which referential properties and relations are intended by the producer of a natural-language text be adequately described in linguistic rules? It might be asked whether it is reall y the l i nguist who should do the job; whether it would not be wiser to l eave the task of describing s uch strategies to psycholinguistics or to A rtificial I n telligence. To be sure, both disciplines have i mportant contributions to make to the topic, but there i s still much left to do for the linguist, as I shall try to show in this paper. To this end, I want to propose a format of linguistic rules which I suppose t o be appropriate for the description of well-defined sub-tasks within the i nterpretation process. These rules have essentially the following properties: - they reflect d ifferent degrees of certainty with respect to a given interpretation hypothesis for diffe rent cases of application ; - they consist of sets of simpler rules for single factors of several types (referring to different kinds oflinguistic as well as non-linguistic knowled ge that are made available by different components of the underlying text u nderstanding model); - each single-factor rule yields relative evidence for or against the interpre tation hypothesis in question; - the i nteraction of the relevant single factors is described systematically by indicating their relative (positive or negative) weights in the global evalu ation ; - the application of the complex rule to a suitable piece of text yields a global evaluation of the interpretation hypothesis in question . This rule format has been developed in the framework of a project aiming a t a (partial) simulation of the processes of understanding and translating natu ral-language texts (see Section 2). I n its present form it has been arrived at by a double approach with deductive as well as inductive aspects: - deductively, we had to take into account the possibilities and requirements of the conception of a text understanding model that was to conform with our assumpt ions about a satisfactory theory of text understanding and translation as discussed in Pause ( 1 98 3); - inductively, we considered the characteristics of specific sub-tasks within
373
(1)
V politiceskoj zizni arabskogo mira proizos lo In political life Arabian (gen.) world (gen.) took-place sobytie, kotoroe . . . , - gosudarstvennyj perevorot v Sudane. event that . . . - political revolution in S udan.
Perevorot byl bystrym Revolution was quick
i prakticeski beskrovnym. and practically bloodless.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the interpretation of referential properties and relations that are left implicit in the given texts. The sub-problems we analyzed in detail are the following: 1. the search for the actual antecedents of anaphoric personal and possessive pronouns in German, French and English texts (see Hauenschild and Pause 1 983, and Pause 1 986a); 2. the interpretation of noun phrases in Russian texts, which normally do not contain any explicit expression of definiteness or indefiniteness (see Hau enschild 1 985). I want to illustrate my proposal of a global evaluation rule on the basis of the second sub-problem. Given the enormous complexity of the whole pro blem of interpreting Russian noun phrases (where all sorts of linguistic as well as extra-linguistic information may, in principle, have to be considered), I chose an even more restricted domain of application for the sample rule: I concentrate on the question of the conditions that yield identity of reference between two noun phrases with lexically identical heads in a Russian text. This case is relevant for the referential interpretation of Russian noun phrases because identity of reference leads to a definite interpretation of the second noun phrase. According to the definition I assumed in Hauenschild ( 1 985), this is the simplest case of defi niteness, namely anaphoric definiteness with identity of reference between antecedent and anaphor, where the referential relation is indicated by the lexical identity of head nouns (for a short discussi on of the definition of definiteness, see Section 3). It is, however, important to notice that identity of head nouns does not guarantee identity of reference: if e.g. the modifiers of the two noun phrases in question lead to contradictory properties of the assumed shared referent (the red book and the black book, 2 identity of reference is i mpossible. There are other types of counter-evidence to identity of reference (see Section 4). It is time now to give an example of the phenomenon described in this paper. The following piece of text is taken from a Soviet periodical of foreign affairs (I took about 600 examples of repeated nouns from that periodical in its Russian and German versions as a basis for my analysis of relevant criteria for identity of reference).' I give the example with an interlinear word-by word translation (in order to indicate its structural properties) and with a "normal" translation (in order to indicate the content).
374 [In the political life of the Arabian world there has taken place an event that. . . : a coup d'etat i n the Sudan. It was quick and practically bloodless.] (from "Novoe vremja" No. 30 ( 1 971), page 1 0)
2. THE UNDERLYING TEXT UNDERSTA NDING MODEL
The text understanding model underlying my proposal for complex evaluati on rules is Con3Tra (= Constance Concept of Context-Oriented Translation, see Engelberg et al. 1 984 and Pause 1 986b), which has been developed by the project " Ubersetzungsbezogene Kontexttheorie" ("Context theory for tran slation") of the Sonderforschungsbereich 99 "Grammatik und sprachliche Prozesse" ("Gra mmar and language processes") at the University of Con stance. I give a scheme of the text u nderstanding model of Con3Tra in Fig. 1 .
�
KNOWLEDGE MODEL
I
LINGUI STIC REPRE SENTATION synta c t i c s entence s truc ture s emant ic- thematic sentence s truc ture themat i c text s tructure
Figure 1
I
knowledge . of the lexicon : wor ld
grammar
XT
tl TEXT
MODEL
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In the given constellation, it is absolutely clear that the two occurrences of perevorot a re identical in reference and thus the second occurrence is interpre ted as a definite noun phrase. Things would be quite different if there was a modifier within the second noun phrase, e.g. perevorot v Egipte (revolution in Egypt). In that case, the two noun phrases could obviously not be identical in reference because the modifiers would yield contradictory properties of the potential shared referent, which we infer from our geographical knowledge. I shall come back to this example in Section 5, where I shall discuss the relevant global evaluation rule.
375
3. DEFINITIONS OF BAS I C CONCEPTS
The following stat.e ments are not definitions in the full sense of the word, they are j ust hints to my understanding of the underlying concepts, sometimes even j ust indications of the way these concepts are reconstructed in the text understanding model of Con3Tra (for a more exhaustive discussion see Hauenschild 1983 and 1 985).
Anaphoricity An expression in a text is anaphoric if its interpretation depends on an antecedent in that text. I assume a very wide notion of anaphoricity (comparable to the notion of "substitution" in Harweg 1 968): the antecedent may be a simple noun phrase as well as a whole paragraph (the proper delimitation of what can count as an antecedent is still to be defined; at this point it is important that the antece dent m ust be found in the text; in this respect my definition differs from that of Stenning 1 978). The referential relation between antecedents and anaphors is not restricted to identity of reference.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
One of the main characteristics of this text understanding model is the double representation of the source text : - on the one hand, there is the Linguistic Representation indicating the struc tural properties of the input text at the different levels of linguistic descripti on; it consists of: - the syntactic sentence structures (dependency structures); - the semantica/-thematic sentence structures (representing simultaneously the functor-argument structure and the thematic structure of sentences; it is similar to the approach in Sgall et al. 1 973); - the thematic text structure (representing the hierarchical and conceptual relations between textual themes and sub-themes); - on the other hand, there is the Text Mode/ for the information conveyed by the text (representing the content of the input text in an object-oriented way, because it is necessary - e.g. in the case of anaphora resolution - to find all . the information given by the preceding text about the referent of the potential antecedent; our approach is strongly influenced by the concept of a discourse model in Webber 1 979). The single factors of our single-factor rules for the referential interpreta tion of Russian noun phrases are to be tested with the aid of these compo nents of the text understanding model (more precisely with the aid of the "actors" that are responsible fer the construction of these components and, most importantly, with the aid of the "arbiter" , whose task is the administra tion and evaluation of interpretation hypotheses in Con3Tra; see Engelberg et al. 1984, Pause 1 986b).
376
Definiteness/Indefiniteness A noun phrase in a text is definite if the verbal behavior of the sender expresses the hypothesis (possibly implicitly) that the text or knowledge model of the receiver contains a suitable (unique) object satisfying the description of the NP or that the receiver is able to infer or construct a suitable (unique) object on the basis of facts and rules available to him. A noun phrase in a text is indefinite if the verbal behavior of the sender expresses the hypothesis (possibly implicitly) that the text model and the knowledge model of the receiver do not contain a suitable object or that they contain more than one suitable object ("object" in this definition is hyper onymous to "reference object" and "knowledge object"; objects may be individuals or sets of different sorts). Different Cases ofDefiniteness Covered by the Above Definition I . Anaphoric definiteness with identity of reference between antecedent and anaphor (e.g. this man, the man, he preceded by a/the man in a text). 2. Anaphoric defin iteness without identity of reference between antecedent and anaphor (e.g. the author preceded by a/the book in a text). 3. Non-anaphoric definiteness (e.g. the present Queen ofEngland or thefirst woman who will receive the Nobel Prizefor linguistics without prior mention of the respective referents). There are some similarities between my defi nition and the approach of Hawkins ( 1 978): his "anaphoric use" corresponds to my notion of "ana phone definiteness with identity of reference", his "associative anaphoric use" to my "anaphoric definiteness without identity of reference". However, I have no notions corresponding directly to h is "immediate situation use" and "larger situation use", because I am not so much interested in deictic uses. My main point is the possibility of constructing new objects that can be
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Identity of Reference Two expressions in a text are identical in reference if they refer to the same entity in the text model. This is just a statement of how "identity of reference" is reconstructed in the text understanding model of Con 3Tra. The main point is that I want to avoid some of the difficulties that are usually connected with the notion of "coreference" (see Bach and Partee 1 980). The most important thing is that the relation of"identity of reference" between linguistic entities is not necessarily a direct reflex of the identity relation between objects of the real world. The producer of a text may, consciously or unconsciously, deviate from the state of affairs in the real world (provided he is speaking of real-world objects). Two expressions are to be interpreted as identical in reference if the producer intends the addressee to connect the information given with respect to the second expression directly with the information given with respect to the first one. This is reconstructed in Con3Tra by representing both expressions with the aid of one and the same reference object in the text model.
377 referred to by definite (anaphoric as well as non-anaphoric) noun phrases. My notions of definiteness and indefiniteness have overlapping domains of application, which I hold to be appropriate: I assume that in the sphere of constructing new suitable objects and corresponding referring expressions, the producer of a natural-language text has some freedom of expression without much i mpact on the content of his utterance. The following statements on sentential themes and textual themes are rather rudimentary for such a complex topic.
Textual Themes Textual themes are the referents which the text is about (static view, after a (partial) result of the understanding process has been reached). There are different degrees of textual themeness: textual themes may be more or less prominent (depending on several criteria, e.g. how often a referent is mentioned). The textual themes are coded in the thematic text structure (see Fig. 1 ), which represents their hierarchical as well as their conceptual inte"ela tions. The most important thing about sentential themes and textual themes in the given context is their contribution to the interpretation of intersentential anaphoric relations: if we search for the antecedent of a given anaphoric expression in the sentence we are just analyzing (where it will normally be located in a relatively thematic position), we have to look for the textual themes in the parts of the text already analyzed as candidates. This result of our informal analysis of different texts is reflected by the following statement on The Prototypical Anaphoric Relation The antecedent refers to a (relatively) prominent textual theme; the anaphor is (relatively) thematic in its sentence. For more details (e.g. criteria for sentential and textual themes, interaction between sentential and textual themes in the thematic structuring of the text) readers are referred to Hauenschild ( 1 985). I shall come back to the definitions in my discussion of the merits and shortcomi.ngs of the global evaluation rule. At this point, I just want to point out that my notions of "sentential theme" and "textual theme" do not directly correspond to van Dijk's ( 1 977) "sentence topic" and "discourse topic". My main criterion of delimitation is that of the dynamic vs. static
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Sentential Themes Sentential themes are the referents of those elements of a sentence that are processed first in the course of analyzing the sentence (dynamic view). The thematic ordering of elements (which is primarily conceived as a scalar one) is reconstructed by the ordering of constituents in the semantic-thematic sentence structure (see Fig. 1 ) , which reflects the order in which the constituents are considered in the construction of the text model.
378 aspect of the thematic structuring whereas van Dijk's notions refer to two different types of entities: "sentence topics" are referents like Eva or Prague; "discourse topics" are propositions.
4. SINGLE-FACTOR RULES FOR TH E INTERPRETATION OF RUSSIAN NOUN PHRASES WITH RESPECT TO IDENTITY O F REFERENCE AND DEFINITENESS
Basic Rule of the Interrelation Between Identity of Reference and Definiteness:
If NP1 and N P2 are noun phrases with identical heads and ifNP1 precedes NP2 in the text and i f N P 1 and NP2 are identical in reference, then NP2 is definite. Morphological Rules:
Rule of Number (or Cardinality) Disagreement If NP1 and N P2 differ in number, then they cannot be identical i n reference· unless one occurrence of the nominal head is a singular form conditioned by a numeral that denotes a cardinality > I and the other occurrence is a plural form without a numeral (in such cases identity of reference is possible in spite of the formal difference in n umber). Rule of Genitive Subjects and Genitive Direct Objects I f NP2 is a subject or a direct object in the genitive case, then identity of reference with NP 1 is relatively improbable. "Configurational" Constraints4
= Absolute(?) Syntactic Rules:
Rule of Elements of One Coordination or One Comparison If NP1 and NP2 are elements of one coordination (fulfilling identical syntac-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I do not want to comment m uch on the details of the single-factor rules (lovers of the Russian language may consult Hauenschild 1 985). I j ust want to state the rules in their full version in order to show what the abbreviations in the global evaluation rule (to be presented in the following section) really mean, and to give an idea of the complexity and diversity of the criteria to be considered. The Basic Rule describes the simplest case of definiteness (according to my definition of the term). For all the other rules it is presupposed that the first two conditions of the Basic Rule are fulfilled (if they were not fulfilled, definiteness would of course not be excluded; for there are other, more complex types of definiteness; but the rules would have to be modified in some respects). There are other restrictions by now: in their present version, the rules do not apply to noun phrases in generic or idiomatic use; most continuous terms (mass terms, abstract terms) are also excluded.3
379 tic and semantic functions) or of one comparison, then identity of reference between them is close to impossible.
Rule of A rguments of One Predication If NP1 and N P2 are (obligatory) arguments of one predication (in the seman tic sense), then identity of reference between them is h ighly improbable u nless they are the arguments of an identity statement. Rule of Components of One Clause If NP1 and N P2 are located i n the same clause, then identity of reference between them is relatively i mprobable unless they are the arguments of an identity statement.
Rule of Complements of Copular Verbs and Implicit Predication If either NP 1 or NP2 is the complement of a copular verb i n a predicative statement (i.e. not in an identity statement) or the complement of an implicit predication (e.g. an apposition or a phrase with kak (as)), then identity of reference between NP1 and NP2 is next to impossible (unless one of them is a complement as indicated and the other is a generic NP).
Rule of Subjects5 If N P2 is a subject and if it is not in the genitive case and if it stands in initial position or at least before the inflected verb and if it fulfils the conditions of the rule of modification (see below) and if the distance between NP1 and N P2 is a small as possible (with a sentence boundary between them) or NP1 denotes a (relatively) prominent textual theme, then identity of reference between NP1 and N P2 (and hence definiteness of NP2) is highly probable. Rule of Genitive A ttributes If N P2 is a genitive attribute and if it fulfils the conditions of the rule of modification (see below) and if it is part of a (relatively) thematic constituent of its sentence or clause or NP1 denotes a (relatively) prominent textual theme, then identity of reference between NP1 and N P2 (and hence definiteness of NP2} is highly probable. Rule of A dverbials IfNP2 is the head of an adverbial or of a prepositional attribute (esp. of a local or temporal PP)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Relativized Syntactic Rules:
380 and if it fulfils the conditions of the rule of modification (see below) and if it is (part of) a (relatively) thematic constituent in its sentence or clause or NP1 denotes a (relatively) prominent textual theme, then identity of reference between NP 1 and NP2 (and hence definiteness of NP2) is highly probable. Semantic-Pragmatic Rules:
Rule of Uniquely Refe"ing Expressions If NP 1 and NP 2 are occurrences of an expression that has a u nique reference in normal contexts (esp. of a proper name) and if none of them contains additional restrictive modifiers, then identity of reference between them and definiteness of both are highly probable. There are other authors who have stated similar regularities for the definite vs. indefinite interpretation of Russian noun phrases (e.g. Birkenmaier 1 979, G ladrow 1 979, Obst 198 1 ; for a discussion of them see Hauenschild 1 985). A mong the critical points there are two i mportant ones: I . none of them analyzes the role of factual knowledge in a systematic way; 2. they only give a partial description of the interaction of the single regularities they refer to. I am not in a position to present a final solution to these two points, but I hope that I have been able to push the solution some steps further: ad 1 . : In my rules, it is at least clear where factual knowledge has to play its role: all the rules that mention textual themeness have to rely partially on factual knowledge because of the corresponding criteria for textual t hemes; the rule of modification refers to factual knowledge in all cases where compatibility of modifiers is n ot just a question of antonymous vs. synony mous lexical entities; even the rule of uniquely referring expressions will often have to resort to factual knowledge. The organization and manipulation of factual knowledge, however, seems to be located beyond the borderline of linguistics proper - I am glad to leave that task to Artificial I ntelligence. ad 2.: My description of factor interaction is still partial, but nevertheless somewhat more systematic than the corresponding statements in the litera ture cited above. I shall come back to this problem in the following two sections.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Rule of Modification6 If there are modifiers in NP2 that do not occur in NP1, then the additional modification of N P2 has to be compatible with what has been said so far in the text about the referent of NP1 (esp. with the modification of NP1) and the "new" modification of NP2 must not yield a true restriction of the contextual extension of NP1; otherwise identity of reference between NP1 and NP2 is i mpossible.
38 1 5.
SCHEME OF A GLOBAL EVALUATION RULE
Scheme of a global evaluation rule Condition : NP 1 and NP2 have identical head nouns; NP 1 precedes N P2• Question: Are NP, and NP2 identical in reference? (Is NP2 definite?) N Negative evidence (the maximum value of negative evidence is - 1 ) N 1 Difference i n number (or cardinality) N2 Conditions of the single-factor rule of modification not fulfilled N3. 1 NP1 and NP2 are elements o f one coordination or comparison N3.2 NP1 and NP2 are arguments of one predication (this condition is tested iff N3. 1 does not hold) N3.3 NP1 and NP2 are components of one clause (this condition is tested iff N3. 1 and N3.2 do not hold) N 4 NP1 o r (excl.) N P 2 i s the complement o f a copular verb N5 If NP2 i s a subject: NP2 follows after the inflected verb in its clause N6 If NP2 is a subject or a direct object: the surface case of NP2 is genitive
-I -1 -0.9 -0.8 -0. 6 -0. 9 -0.9 -0. 6
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The main point in my proposal of a global evaluation rule for the criteria that make two noun phrases with identical heads identical in reference is the following: the rule is meant to give a systematic description of the interaction of all the relevant single factors. To this end, I consider different types of i nteraction, most i mportantly positive and negative interaction. Positive interaction or "factor cumulation" means that different criteria enforce one another in order to yield stronger evidence for or against the interpretation hypothesis in question. Negative interaction or "factor rivalry" means that different criteria contradict one another, whereby the evidence derived from one criterion is weakened by the conflicting evidence derived from the other one. In the texts that served as a basis for my analysis of the criteria relevant for the i nterpretation of couples of noun phrases with lexically identical heads with respect to identity of reference, cases of "factor cumulation" are much more frequent than cases of "factor rivalry". This is plausible because the texts are of the "informative" type, where cooperative strategies of text production are legitimately presupposed. I shall return to the questions of "factor cumulation" and "factor rivalry" in the next section, where I discuss the merits and shortcomings of my proposal of a format for global evaluation rules. But first Jet me give a sample rule in that format. The rule corresponds to the single-factor rules given in the previous section.
382
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
P Positive evidence (the maximum value of positive evidence is +0.8) +0.8 P I NP1 and NP2 are uniquely referring expressions P2 NP2 is a subject, a genitive attribute, a prepositional +0.4 attribute, or a free local or temporal adverbial +O 2 P3 NP2 is any other adverbial +0.2 P4 Conditions of the single-factor rule of modification fulfilled P5 If NP2 is a subject: the distance between NP 1 and NP2 is as �mall as possible or NP 1 denotes a (relatively) prominent textual theme +0.2 P6 If NP2 has any other syntactic function: N P2 is (relatively) thematic in its clause or NP1 denotes a (relatively) prominent textual theme +0.2 In this scheme of a global evaluation rule, the lines indicating the different types of positive or negative evidence as well as the numerical probability values are to be understood as abbreviations for (parts of) the single-factor rules given i n Section 4 and for the relative degrees of certainty respectively. In order to illustrate the functioning of the above evaluation rule, it is now applied to the example ( I ) given in Section I. The relevant criteria clearly yield a positive evaluation: - P2 (NP2 is a subject) gives the positive value of +0.4; - P4 (NP2 has no modifier, thus the rule of modification is observed) gives +0.2; - P5 (first condition: the distance between N P 1 and NP2 is as small as possible) gives +0.2; - P2 in combination with P4 and P5 yields a global positive evaluation of +0. 8 (which is the maximally possible positive value, see Section 6) for the interpretation hypothesis in question, namely identity of reference be tween the first and the second occurrence of perevorot (revolution) in the given piece of text, and thus definiteness of the second. If there had been supplementary modifiers in the second NP, they would have to be tested for compatibility and restrictiveness.7 If the second NP had been e.g. perevorot v Egipte (revolution in Egypt), the geographical part of the k nowledge model would have to give the information that v Sudane (in the Sudan) and v Egipte (in Egypt) are incompatible as properties of one referent. I n that case, N2 would have led to an unequivocal negative evaluation of the hypothesis that NP1 and NP2 are identical in reference. The maximally possible negative value of -I would then "win" over the positive value, which in that case would be only +0.6 because of the non-applicability of P4. This would constitute one of the relatively rare cases of factor rivalry, whereas the realistic example in Section I is much more typical in that it is a clear-cut case of factor cumulation.
383 6 . MERITS A N D S HORTCOMI NGS O F TH E PROPOSED RULE FORMAT
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As has been announced in the introductory section, the main question to be dealt with in this paper is the following: how can the strategies of interpreta tion that enable addressees of natural-language texts to find out which referential properties and relations are intended by the producers of those texts be adequately described in terms of linguistic rules? The rule format proposed in this paper is by no means a final answer to this question, but it may be a first step in a promising direction. It allows a systematic description of the interaction of different criteria, each of which yields but a relative degree of certainty for the interpretation hypothesis in question. This seems to be a typical situation with such complex interpretation tasks as the i nterpretation of referential properties and relations that are left i mplicit by the producer of a text. Considering the texts I analyzed in order to find out the factors that make two noun phrases with identical heads in a Russian text identical in reference, I gained the impression that all the partial regularities I detected ought to be interpreted as tendencies (and not as strict linguistic rules). Therefore, I tried to find a rule format that would allow us to express such relative rules that refer to different levels of the linguistic description. One ofthe advantages ofthe format ofcomplex evaluation rules as proposed in this paper consists in the possibility to describe positive as well as negative interaction of different factors (in the sense of "factor cumulation" vs. "factor rivalry" - as far as I know, the latter has never been considered systematically in linguistic descriptions). In this respect, my proposal is similar to the concept of a two-dimensional "evidence space" in Rollinger ( 1 984), where all kinds of hypotheses (not specifically linguistic in terpretation hypotheses) are evalua ted with the aid of a combination of positive and negative evidence. values. negative evidence values. However, the abbreviation of the relative weights of the different crite"r ia with the aid of simple numerical values (or pairs of such values) is rather problematic. In the case of my sample rule given in Section 5, the correspon ding values have been arrived at by analyzing a large number of examples and they ought to be tested on more suitable pieces of Russian texts; but they can never reach a final state of "correctness" because their theoretical status is undefined. However, I do not know of any viable alternative. Another shortcoming of the proposed rule format is the following: the single factors relevant for the interpretation hypothesis in question are not really independent of each other. The interdependencies, however, are only partially reflected by the global evaluation rule (e.g. in the case of N 3 . 1 to N 3 . 3 , where only the strongest applying condition is tested). It would, of course, be desirable to have those interdependencies explicitly expressed in the rule. The whole multitude of factors pro and contra identity of reference is, after all, implausible for the reconstruction of the processes of text understanding
384
7.
CONCLUSION: TH E POTENTIAL ROLE OF COMPLEX EVALUATION RULES I N A MACHINE TRANSLATION SYSTEM
One of the best criteria of adequacy for linguistic rules (apart, of course, from their theoretical plausibility) consists in testing their functioning within the framework of a complex system of text processing. The rule format that I proposed here has been developed in the context of a theoretically oriented computational model of text understanding and translation (where the com putational realization is still only partial; but there exists a component that is able to handle evaluation rules referring to syntactic factors). Complex evaluation rules, analogous to what I presented in this paper, have been
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
unless the different types of evidence and counter-evidence are conceived of as manifestations of some underlying principle(s). Only if such principles exist is it sensible to assume that the different factors are connected with different components of the underlying text understanding model. 8 In the case of interpreting Russian noun phrases, I found the following basic principles to be responsible for the whole spectrum of different factors (for a more exhaustive discussion see Hauenschild 1 985, Chapter 3 . 2): - the negative factors can be reduced to the Gricean "Cooperative Princi ple" (Grice 1 975) - if the noun phrases in question were assumed to be identical in reference, the resulting interpretation would be inconsistent, or some presupposition(s) would be violated; - the positive factors are related to the principle of the thematic structuring of sentences or texts in the sense of our concept of a "prototypical anaphoric relation" (see the definition of sentential and textual themes in Section 3), where the antecedent is a (relatively) prominent textual theme and the anaphor is (relatively) thematic in its sentence or clause. The connection between the two underlying principles mentioned so far consists in the fact that the principle of the thematic structuring of sentences and texts may be viewed as a component of the Cooperative Principle: if the producer of a text wants to be cooperative, he will use strategies that enable the addressee to find out, in a straightforward way, the pivotal points of an intended anaphoric relation, which is best done by placing the antecedent as well as the anaphor in a (relatively) prominent position within the corresponding thematic structures. 9 The principle of "factor cumulation", which I mentioned in the previous section, can also be considered a component of the Cooperative Principle: it is easy for the addressee to find out the intended interpretation if there is no conflicting evidence (for further arguments see Pause 1 984). All these interre lations and interactions of criteria and underlying principles ought to be expressible in an ideal format for complex interpretation rules. Although my proposal of a rule format is still far from reaching this ideal, I should like to discuss a potential application of such rules in the final section of this paper.
385
Fachberetch 20 lnformatik Technlsche UniversittJt Berlin Franklmstr. 28-29 D-1000 Berlin 10 Germany
NOTES I. So far, I have excluded generic uses of noun phrases from consideration, although they constitute a very interesting phenomenon. Even in languages with definite and indefinite articles, generic interpretations of noun phrases are normally left to be inferred from the context or the background knowledge. 2. The criterion of modifier compatibility is, of course, dependent on special types of texts. I only constder texts of the "informative" type, where the main aim of communication is a reliable transfer of information. 3. It will probably be possible to include mass terms and abstract terms after further analyses on the basis of suitable texts. By now, the material basis is simply insufficient: there is only a small number of examples of repeated mass terms and abstract terms in the texts I have analyzed so far. 4. I have put "configurational" in quotes, although the three following constraints show some similarities with Reinhart's approach of "c-command domains" (Reinhart I 98 1 ). In view of the counter-examples in Bolinger (1971), I hold that these constraints d o not constitute absolutely strict rules, but have to be more or less relativized by semantic (or pragmatic) criteria. 5. In this rule as well as in the following two ones, the last two conditions are to be understood as alternative prerequisites of a maximally positive evaluation. The relative positive weight is equal in cases where one of them is fulfilled and in cases where both are. 6. According to this rule, the following cases get a positive evaluation: - NP, does not contain any modifiers; - NP2 contains the same modifiers as NP1;
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
developed for other s u b-problems within the process of text u nderstanding (see Pause 1 9 86a for an evaluation rule for the search of the actual antecedent for an anaphoric personal pronoun in an English text). It would be highly revealing for the adequacy of the given rule format to find out whether i t m ight be applicable in other environments. A t present, it seems to me rather probable that analogous rules for the interpretation of referential properties and relations will be useful for t he analysis of the source text in the theoretically oriented machine translation system that is being 1 developed in the project KIT/NASEV 0 at the Technical University of Berlin (see Hauenschild 1 986). In that system , too, anaphora resolution and similar problems of referential interpretation will be the crucial points for a correct analysis. The main characteristics of the rule format proposed in this paper, i.e. the systematic description of the interaction of different criteria referring to different components of the text understanding model will in principle be applicable to the analysis part of KIT/NASEV, t oo , because we plan a quasi-modular implementation of the system for independent reasons. Such a quasi-modular simulation of the processes of text understanding and translation gives us at least some hope of being able to cope with the enormous complexity of these processes.
386 the modifiers of NP, are a subset of the modifiers of N P , ; t h e supplementary modifiers o f N P 2 are compatible with what h a s already been said about the referent of N P 1 and they do not yield a true extensional re!triction of that referent in the given context. The test for compatibility and restrictiveness may be very complicated in special cases. I do
7.
not pretend to give a solution for all of them. But still, the concept of a text model in Con1Tra (interacting with the knowledge model) offers a practicable first approach. Of course, this raises the question of modularity and interaction of modules that has been a
8.
crucial point in recent linguistic and psycholinguistic debates. I should argue for a concept of "quasi-m odularity" as it is reflected in the text understanding model ofCon1Tra, where there are different components that interact in nearly all the sub-processes of text understanding. I ought to comment on the fact that the negative factors have a higher maximal value than
9.
the positive factors. This situation can be explained in the following way: the positive criteria
unequivocal negative evidence (e.g. antonymous m odifiers) may lead to an interpretation with an indirect anaphoric relation, but without identity of reference.
1 0.
KIT/NASEV ("Neue Analyse- und Syntheseverfahren zur maschinellen Obersetzung" =
"New Procedures of Analysis and Synthesis in Machine Translation") is sponsored by the Bundesministerium filr Forschung und Technologic as a supplementary, theoretically oriented research project in the context of EUROTRA-D, which is the German pan of the machine translation project of the European Community.
R E FERENCES Bach, E., and Panee, B .
1 980: Anaphoraandsemanticstructure. In:J. Kreiman & A . E . Ojeda(eds.):
Papers from the Paraussion on PronoUIIS and Anaphora. Chocago Linguostocs Society, Chicago, pp.
1 -28.
Birkenmaier, W .
1 979: Artikelfunktionen in einu artikelloun Sprache. Studien zur nomina/en
Determination im Russisclren. Fink, Milnchen. Bolinger, D.
1 977: Pronouns and Repeated Nouns. Indiana University Linguistics Club,
Bloomington. Engelberg, K.-J., Hauenschild, C., Kniiptler, S., and Pause, P.E.
1 984: Con1Tra: einproudurales 99,
Moddl des Textverstehens fur die Obuutzung. Papers of the Sonderforschungsbereoch Konstanz. Glad row, W.
1 979: Die Determination des Substantivs im Russischen undDeutschen. Enzyklop!!die,
Leopzig.
1 975: Logic and conversation. In: P. Cole and J. Morgan (eds.): Syntax andSemantics 3: Spuch A cts. Academic Press , New York, pp. 4 1 -58. H arweg, R . 1 968: Pronomina und Text konstitution. Fink, Milnchen. Hauenschold, C. 1 98 3 : Ans!!tze zu einer dynamischen Semantik im Obersetzungsmodell Con1Tra. Conceptus X V I I (No. 40/4 1 , Special Issue on "Theory of Meaning"): 1 53- 1 64. H auenschild, C. 1 985: Zur Interpretation russisclru Nominalgruppen: Anaplrorisclre &ziige und
Grice, H . P.
thematische Strukturen im Satz und im Text. Sagner, Milnchen. Hauenschild, C.
1 986: K I T/NASEV oder die Problematik des Transfers bei der maschinellen
Obersetzung. In: I . S . Batori & H.J. Weber (eds.): Neue Anslitu in maschineller Oberutzung: Wisunsreprliuntation und Textbnug. Niemeyer, Tubingen, pp.
1 67- 1 95. 1 983: Faktoren-Analyse zur Modelherung des Textverstehens. Linguist/scire Berichte 8 8 : I 0 1 - 1 20. Hawkins, J . A . 1 978: Definiteness and Indefiniteness. A Study in Reference and Grammaticality Hauenschild, C . , and Pause, P.E.
Prediction. Croom Hel m , London. Obst,
U.,
198 1 : Studien zur zweidimensionalen syntagmatischen Substitution in modernen
Prosatuten des Russischen. Aschendorff, MOnster.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(which .are connected with the thematic structuring) give evidence for the existence of an anaphoric relation, but not necessarily of an anaphoric relation with itkntity of referena; thus
387 Pause, P . E . 1 9 8 3 : Context and translation. In: R . Muerle, C . Schwarze, a n d A . v. Stechow(eds.): Mea11ing, Use and Interpretation of Language. De Gruyter, Berlin: 384-399. Pause, P.E. 1 984: Das Kumulauonsprinzip - eme Grundlage filr die Rekonstruktion von Textverstehen und Textverstlindlichkeit. Zeitschriftfiir Liuratu,...,•issenschaft undLinguistik 55: 38-56. Pause, P. E. 1 986a: Anaphor Resolution: Focus A sstgnment vs. A ccumulating Evidence. Papers ofthe Sonderforschungsbereich 99, Konstanz. Pause, P.E. 1986b: Zur Modellierung des O bersetzungsprozesses. ln: J.S. Batori and H .J. Weber (eds.): Neue Anstltu in maschmeller Obersetzung: Wissensreprlisentation und Textbezug. Niemeyer, Tiibingen, pp. 45-74. Reinhart, T. 1 98 1 : Definite NP anaphora and c-command domains. Linguistic Inquiry 1 2 : 60�35. Rollinger, C.-R. 1 984: Evidenzbewertungen filr Fakten und Regeln. ln: Rolhnger (ed.): Probleme
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
des (Text-) Versuhe/IS. Anstltu der Kilnstlichen lnulligenz. Niemeyer, Tiibmgen, pp. 209-219. Sgall, P., Hajieova, E., and Bene�ova, E . 1 973: Topic, Focus and Generalive Semantics Scriptor, Kronberg/Taunus. Stenning, K . 1 978: Anaphora as an approach to pragmatics. In: M. Halle, J . Bresnan and G.A. Miller (eds.): Linguistic Theory and Psychologtcal Reality. Cambridge Umv. Press, Cambridge, pp. 1 62-20 1 . Van Dijk, T.A. 1 977: Sentence topic and discourse topic. Papers in Slavic Philology 1 :49-6 1 . Webber, B.L. 1 979: A Formal Approach to Discourse Anaphora. Garland, London.
Journal of Semantics 4
:
389-390
BOOK NOTE
G.E. Hughes and M.J. Cresswell, A Companion to Modal Logic. Methuen, London and New York, Pp. XVI I I + 203. Ulf Friedrichsdorf and Peter Schroeder-Heister
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The title of this book might suggest that it is primarily a continuation of the authors' well-known "Introduction to Modal Logic" ( 1 968). However, this characterization would be strongly misleading. The book is a self-contained presentation of modal logic from a different point of view. Whereas in "Introduction to Modal Logic" emphasis was put on syntactical features and decision algorithms, the new book is written from the semantical point of view and works entirely with Kripke-style model theory. More i mportantly, it tries to treat a variety of modal systems in a uniform manner. Thus the · authors place special emphasis on the technique of canonical models for completeness proofs and on the characterization of logical systems by classes of models or frames. Naturally, they confine themselves to normal systems and mainly to the propositional case. Many recent results, in particular those of_van Benthem, Fine and Segerberg, are included . The first two chapters i ntroduce the standard normal propositional sys tems K, T, B; S4, SS. and their semantics, and prove compl�teness using canonical models. Chapter 3 presents some recent results about the characte rization of systems with classes of models whose accessibility relation fulfills certain conditions. Although the authors' treatment does not incorporate the most advanced results that can be found in the literature, the presentation of what is also called the "correspondence theory" is entirely satisfactory. Chapter 4 introduces a notion of completeness in the absolute sense as opposed to the completeness with respect to a class of models. A system is called complete if it can be characterized by a class of frames (that each normal system can be characterized by a class of models is trivial - simply take the class which only contains the canonical model). As an example of a theory incomplete i n this sense, the authors present the extension of K by the axiom VB: MLp v L(L(Lq :> q) :> q) ("VB" for "van Benthem"). In Chapter 5 pseudo-epi morphisms between models and frames are defin ed which, roughly speaking, represent the accessibility relation of one model or frame in another. In addition, frames generated by one world are i ntroduc ed. This leads to the result that certain (classes of) models or frames can play the role of others, e.g. that if a formula is valid on a frame it is valid on every generated subframe of that frame, or that any complete modal system is characterized by any class of frames for this system that contains all the generated frames for it. Chapter 6 is devoted to frames of canonical systems. The rule of disjunc tion is established as sufficient for the frame of the canonical model o f S being strongly generated, and a certain amalgamation property is shown to be
390
Univ�rsitdt Konstanz Fakultdt fiJr Math�matik Postfach 5560 D- 7750 Konstanz F�d. R�p. G�rmany
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
sufficient for the rule of disj unction. It is proved that the system KW (also known as G, resulting from K by addition of L(Lp :J p) :J Lp), is complete but not canonical in the sense that the frame of its canonical model is a frame for it. Chapter 7 discusses the use of subordination frames to obtain complete ness results for standard systems without using canonical models and, in particular, without being forced to work with non-countable sets of worlds (remember that with canonical models, worlds are construed as sets of formulas). The method is first used for systems containing D (i.e. systems having Lp :J Mp or M(p :J p) as a theorem) and then extended to other systems. Chapter 8 investigates finite models and proves the finite model property for various systems using the method of filtrations. This method is used to prove the completeness of KW(=G). Systems lacking the finite model prop erty are also discussed in passing. Finally, Chapter 9 gives an outlook on modal predicate logic. The book contains a bibliography of relevant literature, a glossary of technical terms, a list of axioms for propositional systems and an index. I t is also attractively printed and bound. Unfortunately, as the authors realized themselves, not all of the proofs in the book are correct. For this reason they wrote "A Companion to Modal Logic - Some Corrections" (Logique et Analyse 29 ( 1 9 86), 4 1 -5 1 .). The corrections concern primarily the compactness ofK4.3W and ofS4.3. l . , and , more importantly, the completeness of KW. Although the theorems in the book are right, this article shows that repairing the proofs will require a quite substantial rewriting. In spite of this shortcoming, our j udgment of the book is generally positive. It is very well written, and experts in modal logic as well as students with some general background in logic can profit from it. We used this text in a course on modal logic and found it excellent for this purpose. We strongly recom mend "A Companion to Modal Logic" as an introduction to the subject.
Journal of Semantics 4 : 39 1 -393
BOOK NOTE
Gunter Saile, Sprache und Handlung: Eine sprachwissenschaftliche Untersu chung von Handhabe- Verben, Orts- und Richtungsadverbialen am Beispiel von Gebrauchsanweisungen (= Schriften zur Linguistik , 1 0). Vieweg, Braunsch weig und Wiesbaden, 1 984. Pp. 285. DM 48.00 Ad Foolen Th is book is an essay in descriptive semantics. Saile analyses German sen tences which express action by hand and which contain a locative or directional adjunct, cf. ( l ): Frank stellt die Tasche in die Ecke. [Frank puts the bag in the corner.]
Saile concentrates on two aspects of this sentence type, on the one hand on the verbs that express action by hand, the so called verba manupellendi , or, in German, 'Handhabe-Verben', like setzen (put), werfen (throw), giessen (pour), fallen lassen (drop), and on the other hand on the prepositions that occur in the adjuncts, like in (in) and gegen (against). The core of the book gives an extensive semantic analysis of the above mentioned verb class and a more or less complete semantic analysis of the directional and locative prepositions in German . Other aspects of the sen tence type under discussion, like the problem of intentionality and the semantic roles involved, are only marginally discussed. At first sight verbs and prepositions might seem quite different aspects of sentence structure. Saile shows, however, that it makes sense to discuss them in relation to each other. Actions implicate processes, and processes i mpli cate states, as is well known from action theory. States fit in well with locative and processes with directional adjuncts . From these relations a natural organization of the book is deduced: Chapter 2 concentrates on locative prepositions as they occur in locative adj uncts in state descriptions, as in (2): (2)
Die Decke ist auf dem Bett . [The blanket is on the bed .]
In Chapter 3 the directional prepositions are treated as they occur in directio nal adjuncts in process descriptions, as in (3): (3)
Die Flasche fallt auf den Boden. [The bottle falls on the floor.]
Chapter 4 focuses on the verba manupellendi as they occur in action sen tences accompanied by a directional or locative adjunct.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( I)
392
(4)
The cigar is in the mouth.
does not have to mean that most of the cigar is inside the mouth? 3. Due attention is given to the sense-semantic dimension, i.e. to the meaning relations among the different prepositions and among the verbs. 4. The author uses many pictures, schemas, diagrams, and formalisms to make his analysis more explicit. The book is clearly oriented towards logical semantics, and wherever possible the results of the analysis are formulated in higher order predicate logic. 5. Each of the three chapters ends with a section on the syntax of the sentences that are discussed in that chapter. This t akes the form of a set of generative phrase structure rules, linked with a lexicon . Apart from these system atic aspects the three chapters contain many unsystematic, but nevertheless interesting, excursions on diachrony, dialec tology, language acquisition , and contrastive linguistics, which enliven the description of the prepositions and verbs. The book ends with three shorter chapters (50 pages in sum) which have a more applied outlook. Saile analyses a text which gives instructions for assembling a mincing machine. The text contains six illustrations, showing which actions must be taken successively to put the mach ine together. The text contains of course a large nu mber of action sentences with locative and directional adjuncts, which tell the user of the machine where the different parts are and where they should be put . This text offers a good sample of language use in which the action sentences as treated in the theoretical part can be studied in t heir functioning in a n atural environment. Saile examines the accuracy of the instructions in relation to the pictures and he compares the German version with an English and a French translation of it. The book ends with suggestions for how linguists and psycholinguists could be consult ed in formulating and testing such texts.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
../
Each of these chapters is internall y organized in more or less the same way. The following aspects are systematically t reated in Chapters 2 to 4: I. Semantic subclassification. The 26 locative prepositions are subclassified into 4 groups (elementary, vertical, horizontal, neutral), and the 25 directio nal prepositions subclassify i n to three groups (start, path, goal). The verbs are divided into three groups, the first of which contains verbs like stel/en (put), in which an agent controls the whole process. The second class contains . verbs like werfen (throw), in which only the start of the process is under the control of an agent, and finally there is a class of verbs like fallen lassen (drop), which implicate that an agent does not prevent a state or process from con tinuing. 2. A referent ial semantic description of each of the prepositions and, in a schematic way, of the verbs under discussion. This means that for each lexical item the precise conditions for its 'true' application in a possible world are given. For example, how can in (in) be defined in such a way that a sentence like (4),
393 The author has reached a sound balance between theory, description and application. H is strong involvement with the subject matter can be noticed in several places in the book, which makes it a pleasure to read. I can recom mend this study to everyone who is interested in the semantics of German prepositions and action verbs. Univ�rsity of Ntjmtgtn Dtpartm�nl of Linguistics Erasmuspl�in I 6525 GG Nljm�g�n Th� N�thulands
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011