JOURNAL OF SEMANTICS AN INTERNATIONAL jOURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE
MANAGING EDITOR: PETER BoscH (IBM Germany} REV IEW EDITOR: TIBOR K1ss (IBM Germany} EDITOR IAL BOARD: N.AsHER (Universiry of Texas,Austin}
R. BARTSCH (Universiry of Amsterdam) J. VAN BENTHEM (Universiry of Amsterdam) B. BoGURAEV (IBM Research, Yorktown Heights) D. S. BREE (Universiry of Manchester)
S.LEVINSON (MPI Nijmegen)
S. L6BNER (Universiry of Diisseldor� SIR JOHN LYONS (Universiry of Cambridge} A. MANASTER-RAMER (Wayne State Universiry) W. MARSLEN-WILSON (MRC, Cambridge)
B. GEURTS (Universiry of Tilburg} M. HERWEG (Universiry of Hamburg)
J. McCAWLEY (Universiry of Chicago) L. M.G. NoORDMAN (Universiry ofTilburg} R.A. VANDER SANDT (Universiry of Nijmegen) T. SANFORD (Universiry of Glasgow) R. ScHA (Universiry of Amsterdam) H. ScHNELLE (Universiry of Bochum)
P. N. JoHNSON-LAIRD (Princeton Universiry} H. KAMP (Universiry of Stuttgart} E. LANG (Universiry of Wuppertal)
W. W AHLSTER (DFKI Saarbriicken) B. WEBBER (Universiry of Pennsylvania) H. ZEEVAT (Universiry of Amsterdam)
H. BaEKLE (Universiry of Regensburg} G. BROWN (Universiry of Cambridge) 0. DAH L (Universiry of Stockholm) S. C. GARROD (Universiry of Glasgow)
P. HoPPER (Carnegie Mellon Universiry) L. R. HoRN (Yale Universiry) S. IsARD (Universiry of Edinburgh)
P. A. M. SEUREN (Universiry ofNijmegen) A. VON STECHOW (Universiry ofTiibingen) M. STEEDMAN (Universiry of Pennsylvania)
EDITORIAL ADDRE SS: Journal of Semantics, IBM Germany Scientific Centre, IWBS 7ooo-75, Postfach 8oo8 8o,D-7000 Stuttgart 8o, Germany. Phone: (49-711-) 6695-559. Telefax: (49-711) 6695-500. BITNET: bosch @ds0lilog.
New Subscribers to the Journal of Semantics should apply to the Journals Subscription Department, Oxford Universiry Press, Walton Street, Oxford, OX2 6DP. For further informacion see the inside back cover. Volumes 1 -
© Copyright by NIS Foundation All righ!S reserved; no parr of this publication may be reproduced, stored in a retrieval system, or rransmi!!ed in any form or by any means, elec!!onic, mechanical, photocopying. recording. or otherwise without either the prior wri!!en �rmission of the Publishers, or a licence �rmi!!ing remicred copying issued in the UK by the Copyright Licensing Agency Ltd, 90 Tottenham Court Road, London W1 P 9HE, or in the USA by the Copyright Clearance Center, 27 Congress Street, Salem, Mass o1970. The Journal of Semantics is published quarterly by Oxford University Press. Subscription is $117 �r year. Second class paid at Newark, New Jersey. ISSN 0167-5133. POSTMASTER: send address corrections to The Journal of Semantics, c/o Virgin Mailing and Distribution, Building 1 so, Newark International Airport, Newark, New Jersey 071 14, USA.
JOURNAL Volume
10
Number
OF
SEMANTICS
1
CONTENTS CAROLA ESCHENBACH
Semantics of Number REINHARD BLUTNER
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
33
j AN vAN VooRST A Localisr Model for Event Semantics
65
joumal ofSemalltics 10:
© N.I.S. Foundarion (1993)
I-31
Semantics of Number
CAROLAESCHENBACH University oJHambur�
Abstract
'Diese Ansicht, den Gawmgsbegr
ijj" gewissermassen
als ausser der Kategorie des Numerus liegend zu
betracltten, 1md von ibm durch Beugung Sin� . 11iaris und Pluralis zu 1mterscbeiden, ist unlaugbar eine sehr philosopbische, deren Entbehrung andere Sprachen zu anderen Hii
)
/fsmitteln
zwingt' (Wilhelm V<>n
Humboldt .
t
I N T RO D U C T I O N
In the last two decades, the question of how plural can be represented in a logical framework has been widely discussed. Starting with the assumption that the difference between singular and plural should be reflected by differences in the logical types of the referents (individuals vs. sets of individuals; Bartsch 1 973; Hausser 1 974; Bennett 1 97 5), the discussion turned to the assumption that an appropriate representation of plural should be based on a model theory with a structured universe of discourse (cf e.g. Massey 1 976; Scha 1 98 1 ; Verkuyl 1 9X 1 ; Link 1 983, 199 1 ; Hoeksema 1 983, 1 988; Bunt 1 9H 5; Krifka 1 9X9a; L0nning 1 989; Landman 1 989; Dolling 1 990; Barker 1 992). While the discussion mainly focused on what the structure looks like and how the differ ence between distributive and collective sentences should be represented, little has been said about how the singular-plural distinction of nouns should be treated. This question will be tackled in the present article, based on the assumption of a structured model theoretic universe which is presented in Section S· The approach presented in Section X meets the intuition that a singular noun phrase
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This paper presents an analysis of how number can be represented in a logical framework based on a semi-lattice universe. The features singular and plural of count nouns are treated in a uniform way, assuming rhar rhe meaning of nouns should generally be represented as independent of number. This opposes the assumption, quire common in rhe current discussion of plural, that p lural should be analysed as an operator on the meaning of the singular form of count nouns. Based on a discussion of relational nouns as son, sister, and owner it is shown that the predicate-operator approach is not able to handle the number phenomena systematically. In addition, an analysis of some determiners and quantifiers is presented in the proposed framework to illustrate how to cope with their systematic interaction and restrictions with respect to number.
2
Semantics of Number
2 T Y PE - D E P E N D E N T N O U N C L A S S I F I C A T I O N : PR E D I C AT I VE A N D R E L A T I O N A L N O U N S
One (more or less) implicit assumption underlying the current discussion of plural is that the class of nouns is uniform with respect to semantic type. Nouns are always assumed to correspond to unary predicates, and the class of nouns that is better represented by relations between objects is not discussed. 1 Many nouns which are derived from verbs (destru ction , visit , owner), as well as nouns which express relations between people ( mother, brother, aunt , chi ef) and nouns like king , capit al, beginning , end , link , relation , and member, are relational nouns. In order to distinguish the two argument places, I will refer to one as the 'referential argument place' and to the other one as the 'internal
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
such as the boy refers to one (undivided) individual, while the referent of the boys has more than one atomic part. Although this intuition seems to be shared by other researchers working in this field, the. methods they employ in their analysis are more complex and sometimes fail to meet it. This will be elaborated in Section 6 by a discussion of the approach of Link ( 1 98 3), which is, on the one hand, the most influential one in the current discussion of plural and can, on the other hand, serve as a paradigmatic example of approaches neglecting the existence of relational nouns such as son and owner. In the study of semantic phenomena connected to number, it is quite common to stipulate determiners like the , which allow for combination with both singular and plural nouns, to be ambiguous even if they do not exhibit any difference in form in the different contexts. Such treatment is exemplified in the discussion of the approach due to Krifka ( 1 9H9a) in Section 7. In contrast to this, the approach presented here gives a sound basis for a uniform analysis of such determiners as will be shown in Sections 9 to 1 1 . I will follow Lobner ( 1 987) with respect to the logical type of noun phrases in distinguishing between definite (type: e), indefinite (type: ( e, c)), and quantificational noun phrases (type: ( (e, t), t)). Link ( 1 9X6) studies how to deal with structure domains in the framework of the generalized quantifier theory (Barwise & Cooper 1 9X 1). The main discussion as outlined needs some preparation. In the first two sections, I describe two ways of classifying nouns with respect to type differences (predicative vs. relational nouns) and structural differences (count, mass, and group nouns). While the second classification is often dealt with in theories on the plural, the first one has not been considered in this area. In Section 4, I will raise the question of whether every occurrence of number should be represented semantically or not.
Carola Eschenbach 3
.
( 1 ) predicative nouns (boy , circle , squ are , group . . .) a. The girl entered the room. A girl entered the room. b. The girls entered the room. Four girls entered the room.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
argument place' (cf Bierwisch 1 9X9). Predicative nouns such as boy and beer differ from relational ones in having only a referential argument place. Syntactically, the internal argument is always optional. The internal argument can be explicitly specified by several possessive constructions, syntactically occurring as specifier or modifier (John 's mother , my brother , the destru ction o ftlze city, German: der Vater desJungen ), or in special cases by other prepositional phrases (the relation between jo lm and Mary, German: die Eintrittsk arte fur das Kon zert ). In addition to the cases of pure predicative nouns and pure relational nouns, there are many nouns which cannot be classified out of context within this dichotomy. Nouns such as child and picture can occur in both uses. There are nouns which obviously have distinct denotations in their predicative and relational uses (e.g. the child ifEdith may refer to an adult). To handle such cases, we could assume two different lexical entries. But in most cases, it is not easy to determine whether or not a noun is relational and what kind of argument(s) should be assumed, e.g. table , house , car , clut ch (cf Lohner 1 9X 5 ; Krifka 1 9X9b; Fraurud 1 990). The lack of clarity on this point in the area of artefacts can be seen as due to the interaction of the functionality of objects (which sets them in relation to someone or something) and the properties objects have to have to ful fil this functionality. But since these problems will not affect the question of how the singular and plural features of nouns should be semantically described, I will concentrate on clear cases, i.e. nouns which refer to persons and relate persons to each other. Relational nouns have not been studied in the plural discussion, and even if they appear in examples, their semantic type is ignored (cf e.g. Scha 1 9X 1 : the sides o frectangle 1 run p arallel to the sides o frectangle 2 ) Sometimes it has been mentioned that to be roo m-mates/brothers are collective predicates, but the fact that the semantic basis of these verb phrases is given by a relational noun has been ignored (cf Link 1 9X 3; Lenning 1 9X9). The differences in the referential behaviour of predicative and relational nouns with respect to number can be seen in examples ( 1 )-( 3). (A •c• indicates that one needs contextual knowledge to identify unmentioned individuals in order to interpret the sentence.) The class of relational nouns can be split into two subclasses, one containing nouns which specify anti-symmetric relations (2), while the other holds those nouns which allow for symmetric inter pretations ( 3).
4
Semantics of Number
( 2)
relational antisymmetric nouns (referee , member, teacher, side , . . .) a. C'fhe daughter entered the room. cA daughter entered the room. b. C'fhe daughters entered the room. cFour daughters entered the room. (3) relational non-antisymmetric nouns (p artner, opponent , relative , . . .) a. C'fhe sister entered the room. cA sister entered the room. b. The sisters entered the room. Four sisters entered the room.
(4) a. Karin's daughter entered the room. b. Michael's daughters entered the room. c. Ina's sister entered the room. d. Oliver's sisters entered the room. The crucial point illustrated by these examples is that it is not necessary to specify an argument of the plural relational noun if it occurs reciprocally as in (3b). In this case, the plural noun is used as a predicate on complex objects: It is interpreted as meaning that all atomic parts of the referent are connected pairwise to each other by this relation. The reciprocal use of relational nouns is possible for all relational nouns that do not exclude symmetric constellations. It should be possible to describe it systematically on the basis of the representation of the nouns in the lexicon.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The relevant noun phrases in ( r ) give information about age and sex of the referents, while the noun phrases in (2) give information about their sex and specify that they stand in a special relation (child -of) to some person not mentioned in the sentences. To interpret these sentences properly, we need contextual information about at least one of their parents; that is, the context has to specify whose daughter(s) is talked about. Similar problems arise with (3a), while (3b) allows for two interpretations. Either we already know one sibling of the referents as in (3a) (this reading should also be marked with •c'), or we interpret the noun phrase reciprocally: the referents stand in the sister-of relation to each other. I will call a noun phrase with this interpretation a 'reciprocal occurrence' or 'reciprocal use' of the relational noun. In (4), the i nformation needed to interpret (2 ), (3a), and the non-reciprocal reading of (3 b) properly is given explicitly.
Carola Eschenbach S
3 S T R U C T U RE - D E P E N D E N T N O U N C L A S S I F I C A T I O N : MASS, COUNT, AND GROUP NOUNS
=
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
W hile the differences i n type are not analysed i n the plural discussion, much has been said about noun classification with respect to structural properties of the universe underlying the denotations of nouns. The distinction between mass nouns and count nouns has a long tradition in philosophy and semantics (cf Strawson 1 95 9; Quine 1 960; Link 1 983; Bunt 1 98 5 ; Krifka 1 989a; Pelletier 1 979; Pelletier & Schubert 1 989, and literature given there). Typical mass nouns are water, g old, sugar, while man , car, event are typical cases of count nouns. The nouns of these classes are said to exhibit different behaviour in morphology, syntax, and semantics. A morphological distinction between mass nouns and count nouns is that mass nouns do not allow for both a singular and plural form. Many mass nouns in English and German are singularia tantum, i.e. do not allow for pluralization. Syntactically, mass nouns co-occur with mu ch , little , and unstressed so me ( sm), while count nouns have numerals, a( n ), each , every, many , several, jew , and stressed s ome as determiners. But there are also some determiners such as the and no which are not sensitive to the mass-count distinction. In contrast to singular count nouns, mass nouns and plural predicative count nouns have the property of cumulative reference (cf Quine 1 960; Link 1 98 3). This property can be described in the scheme: if P applies to A and P applies to B , then P applies to A and B . In the case of reciprocal readings of relational nouns, the situation is more complicated. From Claudi a and Angela are sisters and In a and Friederike are sisters it does not follow that Claudia, Angela, In a, and Friederike are sisters. But we can obtain Claudi a and Angela, and In a and Friederike are sisters . Thus the cumulative reference property does not hold in the case of associative use of and (cf Link 1 99 1 ; Landman 1 989; Hoeksema 1 988; and Krifka 1 990 on the treatment of this kind of phenomena). As observed by Allen ( 1 980), these classes are not as clearcut as they seem to be. The various criteria for the count-mass distinction will yield different divisions of the class of nouns. So it seems that what we have are some clear examples such as water , gold , lightning vs. boy , car, tree and many in-between cases such as apple, cattle, and admiration . Pelletier & Schubert ( 1989) give an elaborate discussion of which linguistic level (syntax, semantics, or pragmatics) is sensitive to the mass-count distinction, and whether this is a distinction between nouns, noun occurrences, or full noun phrases. They come to the conclusion that the most promising approach is to ascribe the difference to full noun phrases and treat it as a semantic distinction. If a (singular) noun is not preceded by a determiner that allows for one interpretation only, the larger context may determine whether it occurs as mass
6
Semantics of Number
The second distinction between nouns in the discussion of plural is the distinction between nouns denoting simple individuals (boy , circle , thing) and nouns denoting groups (group , family , co mmittee , couple), which are tradition ally included in the set of collective nouns. The class of group nouns is interesting in the discussion of number because some predicates are only applicable to plural phrases and to singular group-denoting noun phrases, but not to singular individual-denoting noun phrases. (s) a. The girls meet. b. *The girl meets. c. The group meets. This distinction is mainly a semantic one.2 In German, which is more strict with respect to number agreement than British English (cf The co mmittee meet.), a singular group noun can only appear in syntactic contexts which permit singular nouns (cf (6)): group nouns do not agree with plural verbs, must not be antecedents of reciprocal pronouns, and cannot be preceded by determiners that exclude singular count nouns. (6) a. *Die [the b. *Die [the
Gruppe group-SG Gruppe group-SG
trafen met- PL traf met-SG
sich. REFL.] einander. each other.]
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
or count. It should also be mentioned that there may be contexts in which a decision between the two readings is irrelevant (The apple in the sal ad makes it sweet). A noun that can occur in mass contexts as well as in count contexts is usually interpreted differently in the different contexts. The count reading can be related to the mass reading by 'kind of' or 'portion of' (e.g. Yesterday I t as te d two winds. I'd like a beer.), while the relation o f the mass to the count version can often be described as 'material of' (e.g. She puts so me apple into the s alad). In order to handle this in the semantic representation, one could describe the difference between mass and count readings by applying operations to the basic noun representations, yielding proper count or mass nouns (cf Link 1 98 3; Pelletier & Schubert 1 989). But since the question of how this could be described properly is not my point, I will concentrate on clear cases of mass and count noun occurrences. In the discussion of the semantics of number, I will mainly concentrate on count noun occurrences and describe the difference between their singular and plural forms. As Allen ( 1 980: 542) pointed out, it is not a linguistic universal that mass nouns are classified as singulars. And even in languages such as English, a few nouns that meet the criteria for mass nouns are morphologically plural (e.g. news , me asles, linguistics ). At the end of the paper I will describe how mass noun occurrences can be integrated into this framework.
Carola Eschenbach 7
c. *Alle Gruppe traf sich. [all group-SG met-SG REFL.] In the literature on plural, the main question discussed with respect to this distinction is whether the xroup and the members o fthe group should or can have the same denotation (cf the discussion in Link 1 98 3, L0nning 1 989, and Landman 1 989). The analysis of number which I will present at the end of the paper depends on the assumption that the referents of thegroup and the members efthegroup are distinct. On this basis, the difference between group and groups is analysed in the same manner as the difference between girl and girls . As a con sequence, member is interpreted as a relation between group-objects and other objects (cf Dolling 1 990 and Barker 1 992). S Y N T A C T I C V S . SE M A N T I C N U MB E R
Number is a syntactic feature which is subject to agreement between various parts of a phrase or a sentence. The extent to which this agreement takes place differs from language to language. I will analyse number on the semantic level as a feature of noun phrases and consider the agreement occurrence of number in items such as verbs as a purely syntactic feature, even if the number of the noun phrase is obvious only from its agreement behaviour.3 Although this is not always mentioned in the current discussion of plural, this is done by other researchers in the same way. However, it is seldom said explicitly to which operators number features are mapped in the semantic representation, and how and when such a mapping can be done. The only approach which interprets singular corresponding to plural by assigning semantic operators to both of them is the one presented by Verkuyl ( 1 98 1 ). In the other approaches, the singular form of a noun is assumed to correspond to the semantic core of the noun, to which the plural operator can be applied (I am not sure whether to count Krifka's approach, which will be discussed in Section 7, as a proper exception to this)! Connected to the analysis of the feature singular is the question as to whether the singular form of mass nouns and count nouns should be analysed in the same way. Krifka ( 1 9X9a) assumes mass nouns to be transnumeral, i.e. the feature singular is not interpreted at all. I will assume that we have three semantic number categories: mass, (count-)singular, and plural and that the syntactic feature singular is (at least in German and English) ambiguous between mass and (count-)singular. At the end of the paper I will show what is comon to mass and (count-)singular.5 The next problem to be regarded here is whether the number of a noun occurrence is a consequence of agreement with another constituent of the noun phrase. This question was raised by Verkuyl ( 1 9X 1 ) and Krifka ( 1 9X9a) in the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4
X
Semantics of Number
A determiner which does not restrict the number of the noun (such as the and all) is given a uniform representation which combines with the result of applying the number operator to the core noun meaning. A determiner which restricts the number of the noun (as a ( n ) and bo th ) is applied to the core noun meaning. (Its representation might give a hint as to why it restricts the number of the following noun in its specific way.) Another problem arises with questions and negations. It has often been noticed that a question like (7a) cannot honestly be answered in the negative if the addressee has only eaten one apple or parts of one apple (sec e.g. Krifka 1 9X9a: 72; Lonning 1 9X9). A similar problem arises with (7b, c); e.g. (7c) is false even if John only ate one apple or parts of one. (7) a. Did you eat apples? b. John did not invite girls to his parry. c. John ate no apples.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
context of numeral constructions as three apples and 1.0 apples . They assume that the choice of the plural form of the noun is a consequence of agreemcllt berween the numeral and the noun. Regarding other determiners which are restricted i n combination with nouns like each , every , a (restricted to singular count nouns) and &o th , many (restricted to plural count nouns), I will follow them for the purpose of the currenr discussion by assuming that these restrictions are due to the semantic contellt of the determiners, and thus consider the number of the corresponding nouns as a pure agreement pheno menon that is not interpreted semantically. So the main question that has to be answered is: how should the number feature of a noun be represented if this is not fixed by another constituent of the enclosing noun phrase? While the approaches ofVerkuyl and Krifka depend upon their treatment of number in the context of numerals, the approach presented in this paper is basically independem of whether number is interpreted in the context of the items mentioned. It is mainly the plural form co-occurring with 1.0 which will be a problem in the general interpretation of the number feature. The question of whether the number feature of a noun should be interpreted at the semantic level or mainly be described as an agreement phenomenon has led to a lack of clarity with respect to the interaction between number and determiners. Several approaches assume determiners that can be combined with singular and plural nouns as being ambiguous. In addition, the different readings ascribed to a determiner such as the are often not systematically derivable by the combination of a shared semantic basis and the meaning of the number features (cf e.g. Link 1 9X 3; Scha 1 9X I; Bunt 1 9X s). In contrast to this, I will give representations of determiners which meet the following requirements:
Carola Eschenbach 9
As
s
THE FOitMA L LA NGUA GE A ND THE STRUCTURE O F THE UNIVEitSE
Before entering an analysis of Link's representation of plural nouns, I will give a short presentation of the formal framework for the discussion of plural noun phrases. As usual in formal semantics, I will have a formal language named 'Logic of Plural Nouns' (LPN) as an intermediate level between natural language and the denotational level. LPN is based on first-order predicate logic with equaliry, lambda abstraction, and a proper treatment of l -descriptions. Although I will use predicate-valued variables to describe several operators in LPN, the quamifiers V and 3 will never bind them (cf L01 ming 19X9: 1 sX ff on the completeness of first-order languages with second-order relation and function symbols). In the model - theoretic part, which yields the semantics for LPN, I assume a structured domain of rcfen:nce.6 This structure is reflected in LPN by additional basic symbols e and at and dw defined symbols �. <,La. b. ctbl, cmplx, and mass. The symbol e, which is a binary operation symbol, i s meant to represent the conjunction of terms as in john an d Mary , brea d an d hu ller. or pa per an d a pencil. at is a unary predicate symbol. A n LPN-strucrure is a triple (D, A, u ), where uis a binary operation on rhe set D and A � D. For simpliciry, I will use rhe symbol u as representing an opera tion on subsets of D and D X D roo: if S, N � D, then SuN : {s u nJs E S 1\ n E NJ. If S, N � D >: D, rhm SuN:� {(s 1 un 1 , s� un�)J(s 1 , s�) E S 1\ (n 1 1 n�) E NJ. A subset S of D (or D X D) is closed with respect to u iff S u S � S. The smallest subset of D (or D X D) which is closed with respect to u and includes a set S is called the u-closure ofS (written: su). su always exists and irs definition does not depend on whether D is complete with respect to u.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
a consequence, Krifka and L0nning do not analyse plural strictly: the denotations of plural nouns are allowed to include atomic objects. Krifka describes the choice between the singular and the plural form of a noun as purely pragmatic. So the interpretation of phrases such as the girls as referring to a complex object is only determined by pragmatic inferences and not on the semantic level (see Krifka 1989a: 72 f). L0nning explicitly excludes the problem of associating the morphological feature with the semantic operator from his analysis (cf L0nning 1 9H9: 10, fn. H). To handle (7) in the context of the plural discussion, I think it is more reasonable to re-analyse negation and question phrases, or to determine whether number should be represented semantically at all in these cases. At the end of the paper I will give a description of no which is not sensitive to the number distinction and meets these semantic requirements.
1o
Semantics of Number
An LPN-model Jf- ('7/, [.]) consists of an LPN-srrucrure ?/• - (D, A, u ) and rhe interpretation function [.], which maps LPN-expressions ro 01. The ser D is rhe domain of rhe standard quantifiers V and 3. In LPN rhe symbols E9 and at corresponds ro u and A, respectively, i.e. [EEl] u and [at] A. The defined symbols � and < reprcsenr rhe improper and proper parr relations, while rhc disjoinrness of rwo objects is symbolized by l The sy1nbols.; and # will be used ro abbreviate [�] and [l], respectively. =
ID�] [D<J IDU
=
]a � b � a E9 b - b] [a< b � a� h 1\ a i' b] ]a lb � --.3 c ]c � a 1\ c � b]]
V a, b V a, h V a, b
[A E9 1] [A E9 2] [A E9 3] [A E9 4] [A EEls] [A E9 6] [Aat]
V a, b 3 c Va V a, b V a, b, c V a, b \fa, b \fa
[a E9 b- c] [a E9 a- a ] [ a E9 b- b E9 a] [a E9 (b E9 c)- (a E9 b) E9 c ] [-.(a� b)=> 3 c [c �a 1\ clb]] [-.(aLb) => 3 c V d [ d �c � (d � a 1\ d � b)]] [at(a) => -.3 x [x
According ro ]Aar], A is a subser of rhe aroms of (D, u ). This ser is meanr ro hold rhc denorarions of singular count noun phrases. lr corresponds ro rhe domain of reference in standard approaches of model-rheoreric semanrics which are nor concemed wirh number or rhe counr-mass distinction. The domain of reference I) may include mass-enriries as well as atomic and complex enriries. Mass-enriries are rhose enriries which do nor haw any parr included in A ([mass]- M - D\(A u D)). I will nor borhcr about whar rhe parr of D which holds rhe mass-entities looks like. One could assume rhar every mass-enriry has a proper parr, bur rhis is nor necessary in rhis approach. The only imporranr rhing ro norice is rhar we do nor force mass-entities ro be builr up from aroms in D . Counrabk entities are rhose which do nor have any mass enriry as parr, which means char they are exclusively built up from elements of A ([ctbl] - C - D\(D u M) ).x Complex enriries are finally those which are countable bur nor elements of A. I Dmass] V b [Dcrbl] Vb ]D cmplx] V b
]mass(b) � -.3 a ](a � b) 1\ at(a)Jl [ctbl(b) � V c [c � b => 3 a [a � c 1\ at(a) ] ]] ]cmplx(b) � ctbl(b) 1\ -. at(b)]
The operation u is assumed nor ro be sensitive wirh respect ro the mass-counr distinction in order to allow objects in D which are neither proper masses nor
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The properties ofu and A are specified wirh help of rhe axioms ]A E9 1]-]A E9 (>] and ]Aar].7
Carola Eschenbach I I
countable objects (cf. Link ryXJ; Krifka 1yXya). Bum ( 1 yX 5 ) gave an elaborate analysis of such 'ensembles'. The inclusion of rhe subset A of D is due to the assumption of an objective atomicity criterion, i.e. a criterion that judges which objects are atomic individuals, and which are not. In the latter discussion I will need two partial functions, mapping su bsets of D on elemenrs of D. max,; is rhe partial function defined by: _
max,;(S) :=
{x,
ifx E S and 't/y E S[y�xj . un d e f'med , orI 1erwise ·
If S only comains one element, e.g. S (xJ, we get x � max,;(S). There are two ways for max,;(S) ro yield no value: S is empty or S has more than one elemenr, ' bur it does nor conrain any elemenr which has all the other elemenrs of S as parts. The generalized sum -operation is given by: =
·= ·
z E D [z { undefined, otherwise x
,
if't/
#
x
� 't/ y E S [y
#
z]]
Since sum is not defined on0 at least, it is a partial function. To assume sum to be a total function is exactly to assume that D is a complete Boolean algebra. Bur as long as we apply this function to non-empty finite subsets of D, we will always get a well-defined result. The function sum coincides with max" in the domain on which both are defined. As long as S is finite, the condition sum(S) max,. (Su) holds. Bur if S is not finite it might be the case that max" (Su) is undefined while sum(S) is defined (cf. L0nning 1 989). Since max,. and sum are operators on sets, the corresponding conditions can only be given by definition-schemata in first-order predicate logic. The symbols I will use in LPN as counterparts to these operators are 6 and o, respectively. =
[ DS6) [ DSo]
6(P) =LX jP(x) 1\'t/y [ P(y) � y � x]j o(P) =LX 't/ z [x lz � 't/ y [P(y) � z ly]]
Nouns are represented by unary predicate symbols and relational symbols in LPN, which are mapped by [.] to subsets of D and D X D (or on) respectively. In contrast to mereological attempts to represent mass nouns, I assume that (predicative) mass nouns denote sets of mass-entities, rather than mass-entities. This allows a uniform treatment of mass nouns and counr nouns (Link 1 98 J). A proper predicative count noun will have a subset of C as irs denotation, while a proper predicative mass noun has a subset of M as its denotation. Nouns which allow for both occurrences may have elements of M and C in their denotation (cf. Link 1 9X 3; Pelletier & Schubert 1 989). The objects that are referred to by group nouns are included in A and C in the same way as the other individuals. The represenration of determiners will be discussed later.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
sum(S)
12
Semantics of Number
The structures presented for the plural interpretation are similar enough to the structure Link introduced in Link ( 19R 3) to discuss his suggestion for the meaning of plural with respect to nouns on this basis.
6 SE M A N T I C S O F N U MB E R : L I N K
[D*] (0181]
[*P] [181P]
= =
(sum(X)IX � [ P] [*P]\A
A
X# 0}
The plural operators are only defined on unary predicates. From chis and the observation in Section 2, the question arises as to whether it is possibk to define corresponding operators applicable to relations. We can reformulate Link's description of the operators • and 181 in our formalism as in (Sa) and (8c). I will discuss the operators on relational nouns chat correspond to *, because this operator is always introduced as che·representation of plural in approaches following Link's suggestions (e.g. L0nning 1 9R9). (R) a. b. c. d.
* /..P/..x [ctbl(x) 1\ V z � x ( at(z)=> P(z)]] *GIRL f..x [ctbl(x) 1\ V z � x [at(z)=> GIRL(z)]] 181 /..Ph [cmplx(x) 1\ V z < x [at(z) => P(z)]] 181GIRL f..x [cmplx(x) 1\ V z < x (at(z) => GIRL(z)]] =
=
=
=
The corresponding operators applicable to relational nouns should allow us co represent the phrases (9a-e), which give all possible combinations of singular and plural head nouns and singular and plural complements. The difference berween (9d) and (9e) is not syntactic or semantic but due to world knowledge. While persons of different sex may have common children, chis is not possible
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
There are essentially two different approaches to the semantic representation of plural based on a lattice-structured universe. The first one is presented by Link ( 1 9R3) and will be described in this section. The second approach is given by Krifka ( 1 989a) and will be oudined in the next section. Link's basic assumption is that the core meaning of a count noun is given by the meaning of the singular form of that noun. As a consequence, there is no operator in his logic which corresponds to the meaning of singular. In his analysis, nouns such as boy are mapped to predicates which have only atomic objects in their denotation. He defines two operators, applicable to such predicates, called 'the plural operator' (*) and 'the proper plural operator' (181). The plural predicate is derived from the basic predicate by applying * or 181 to the singular predicate. The semantics of* is the function that maps a set of atoms in D to its complete u -closure (i.e. a closure operator based on sum); 181 excludes the atoms from chis set, allowing only complex objects in the denotation
Carola Eschenbach
'1
3
for persons of the same sex. The semantic representation of these sentences should allow both interpretations. (9) a. b. c. d. e.
son of Peter sons ofjohn son of Peter and Ann sons ofjohn and Mary sons ofjohn and Peter
( 1 0) A.x[SON(PETER)(x)] The obvious transformation of the plural operator to the relational nouns yields ( 1 1 a), which is of use in interpreting (9b) as in ( 1 1 b). ( 1 1 ) a. * = ARAyAx [ctbl(x) 1\ 'tJ z � x [at(z) => R(y)(z)]] b. Ax [ · *SONQOHN)(x)] = Ax [ctbl(x) 1\ V z �x [at(z) => SONQOHN)(z)]] ·
In order to represent the sentences (9c-e), we have to assume an additional operator to allow for complex arguments. This operator is not j ustified by syntactical or morphological features in natural language. It has to be applied if the internal argument of the relation is specified by a plural noun phrase. Such an operator must be applicable to the basic (singular) relation in order to get the right interpretation of (9c). The obvious interpretation in this framework is ( 1 2a). The translation of (9c) is given in ( 1 2b). ( 12) a. * · = ARAyh [ctbl(y) 1\ 'tJ w � y [at(w)=> R(w)(x)]] b. Ax [* · SON(PETER$ ANN)(x)] = Ax [ctbl(PETER$ANN) 1\ 'tJ z �PETER$ ANN [at(z) => SON(z)(x)]] - h [SON(PETER)(x) 1\ SON(ANN)(x)J9 This operator is guitc strong if we consider the examples (9d, e), in which the noun is plural, and will only yield the right representation for example (9d), as in ( 1 .2c), assuming that the common children ofjohn and Mary are referred to. ( 1 2) c. Ax [*· ( · *SON)QOHN$ MARY)(x)] Ax [ctblQOHN$ MARY) 1\ V z �JOHN$ MARY [at(x) =>
·
*SON(z)(x)]J
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As stated above, Link assumes that the basic meaning of the nouns is only defined on atomic objects. This restriction is meant to reflect the distributive nature of predicative nouns: each of some boys is a boy. To make the problem of this assumption as clear as possible, I take the strict position that all argument places of a relational noun such as son , member, riferee , teacher orfriend should be restricted in the same way (i.e. [SON] � A X A). With respect to these nouns, this assumption is justified since, for example, each of the sons ofjohn is a son ofjohn, and a son ofPeter and Ann is a son of either of them. Such a basic form can directly be used to represent (9a) as in ( 1 0).
1 4 ·Semantics
of Number
Ax [ · *SONQOHN)(x) 1\ ( · *SON)(MA RY)(x)] Ax [ctbl(x) 1\ V z �x [at(z) => SONQOHN)(z)] 1\ V z �x [at(z) => SON(MARY)(z)]] Ax [ctbl(x) 1\ V z �x [at(z) => SONQOHN)(z) 1\ SON(MARY)(z)]] To represent (9e) as in ( 1 3b) we would like to have an operator ** as in ( 1 3a), which cannot be the result of combining * · and · *. J..IUyh [ctbl(x) 1\ V z �x [at(z) => 3 w �y [R(w)(z)]] /\ ctbl(y) 1\ V w �y [ at(w) => 3 z �x [R(w)(z)lJ b. A.x [**SONQOHNEBPETE R)(x)] - Ax [ctbl(x) 1\ V z �x [at(z) => 3 w �JOHN$ PETER [SON(w)(z)]] 1\ ctblQOHNEBPETE R) 1\ V w �JOHNEBPETER [at(w) => 3 z �x [SON(w)(z)]]] - A.x [ctbl(x) 1\ V z �x [at(z) => SONQOHN)(z) V SON(PETER)(z)] 1\ 3 z �x [SONQOHN)(z)] 1\ 3 z �x [SON(PETER)(z)]]
( 1 3) a.
••-
·
( 14) a. b. ( 1 s) a. b.
George is a member of the committee. George is one of the members of the committee. George is an owner of this house. George is one of the owners of this house.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
If we were to assume that the proper counterpart to (X) is in fact **, and * and · * arc the consequences of restricting one of the arguments to atomic objects, (e.g. • · R - A.yA.x [**R(y)(x) 1\ at(x)] = A.y/..x [ctbl(y) 1\ at(x) 1\ V w � y [at(w) => R(w)(x)]] ), we need to assume that singular has a semantic counterpart, which is only used for relational nouns. If we considered the operators which correspond to Link's proper plural operator 0 (i.e. · 0, 0 · and 00) this possibility would not exist. One might now, as one of the referees did, argue that it is not necessary in Link's approach to assume that the basic meaning of a relational noun is restricted to A X A. In this case, one might assume that the relation SON holds berween any male person and his parents. As a consequence it is necessary to state the disrributivity of this relation separately, and the operator ** is still needed to be able to represent (9b, e) appropriately. Thus, plurals would be represented by * in the case of a predicative noun and by ** in the case of a relational one. However, there are still some relational nouns to be considered which behave differently from son and the others mentioned above. While ( 1 4a) is the sentence preferred to describe the relationship berween George and the committee, ( 1 sa) seems a bit odd in contrast to ( 1 sb). Thus, ownership might be ascribed to a collection of people, from which the relation berween one of them and the property is derived.
Carola Eschenbach
1s
(3) b. The sisters entered the room. Four sisters entered the room. ( 1 6) 1-RJ..x [ctbl(x) 1\ 'V z, w �x [(at(z) 1\ at(w)=> (R(w)(z) � (w "# z))]] Assuming with Link that the basic meaning of count nouns is restricted to atoms, one would need several operators to get all possible interpretations of relational nouns and one would have problems with the proper representation of nouns like owner. Thus it appears that Link's plural operators are not the representation of the morphological feature plural. What he defines is, rather, a way of obtaining a cumulative predicate in a structured domain on the basis of a non-cumulative one. It is justified if we consider how we may evaluate a plural predicate with respect to a collection of objects, i.e. on the conceptual level of language understanding. But these operators cannot correspond to the feature plural, as long as there are collective plural noun phrases such as the sisters , the murderers qfSa muel Edward Ratchett or plural nouns without a corresponding singular form such as people and German: Leute , Eltern .
7
SEMANTICS OF NUMBER: KRIFKA
Krifka ( 1 989a) assumes that the meaning of a count noun should be represented by a complex structure that is based on the core meaning of the noun which does not depend on number. I will use GIRL to represent the core meaning of the noun girl. I ts denotation is assumed to be closed with respect to the join operation. The difference berween mass nouns and count nouns is reflected in Krifka's approach by a difference in type. While a mass noun is represented by a unary predicate on mass-entities, a count noun is mapped to a more complex structure: a relation berween individuals and rationals. Girl is represented by
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The noun owner, like invent or, auth or, and murderer, might in irs plural form behave collectively with respect to the referential argument place. In Link's framework, these nouns and their behaviour with respect to number would have to be treated differently from predicative nouns and relational nouns like son and member . The integration of relational nouns is also a serious challenge to the approach ofVerkuyl ( 1 98 1 ). Since it is based on pure distributive interpretations of count nouns, phrases as sons ofPeter andjolm and owners ofthis h ouse cannot directly be integrated. It would either need revision with respect to the type of the basic meaning of the nouns or additional operators to handle number and numerals. To get the full paradigm of possible plural noun uses, we would in both frameworks have to assume an additional operator to get the reciprocal reading of(3b). This can be formulated as ( 1 6).
16 Semantics of Number
( I 7), where NE(GIRL) is a function which maps an object to the number of (atomic) girl -parts it has. ( I 7) 1-r, x [GI RL(x) 1\ NE(GI RL)(x)
=
r]
Krifka ( I 989a: 72) specifies the semantic plural -operator as depicted in ( I 8), 1 0 and girls is represented as in ( I 9). The plural-operator is not used to represent phrases like three girls (represented by (2o) ), in which he judges the plural as a purely syntactic occurrence. So this approach seems to explain why only singu lar count nouns cannot appear as bare forms in German and English.
=
Krifka does not introduce a semantic operator to represent singular. As a consequence, he has to assume that all determiners (definite article, quantifiers, possessive pronouns, etc.) have at least two different semantic representations, one applying to singular count nouns and one applying to mass nouns and plural count nouns. While plural (which is not determined by agreement) is interpreted at the noun itself, and the singular of mass nouns is not reflected at all, the representation of the singular feature belonging to a count noun is only forced by the correspond{ng determiner (c£ (2 I ) ; Krifka I 989a: 76 f£). 1 1 In a way, Krifka assumes that the difference between singular and plural should be represented by a difference in type. He sees the difference not between the referents of such phrases (as with Bartsch I 973; Hausser I 974; and Bennett I 97 5), but between the denotations of the nouns. ( 21 ) Det.
the 1 10 each
pl. mass
sg
1-X[MAX(X)] 1-Y , X[Y n X= 0]
1-Z[MAX(Z( I ))] 1-Z, X[Z( I ) n X= 0] 1-Z, X [Z( I ) � X] 1-Z, x [Z( I )(x)j
a
E
1-X, x [X(x)]
Krifka cannot assume that the plural-operator he specifies corresponds to the plural morpheme of the noun. This becomes obvious if we regard pluralia tan tum as Leute (people , folk), Eltern (parents), and Gesclzwister (siblings), which denote complex objects and do not have (at least in the German examples) a sin gular form. It seems to be most natural to assume that the meaning of plural is fixed within their lexical entries. According to ( I 9) it seems reasonable to assume the representation (22) as the lexical entry for Leute. The existential quantifier must be included in the representation, because there is no plural morpheme which could supply it.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( I 8) /-Z, x 3 r [Z(r)(x) 1\ r � I j ( I 9) AX 3 r [ GIRL(x) 1\ NE(GIRL)(x) = r 1\ r � I ] (2o) h [GIRL(x) 1\ NE(GI RL)(x) 3]
Carola Eschenbach
(22)
17
h 3 r [LEUTE(x) 1\ NE(LEU TE)(x) = r 1\ r � 1 ]
X A N A LT E R N A T I V E A P P R O A C H TO T H E S E M A N T I C S O F N U MB E R
Although Link and Kritka introduced a structure set as the domain of reference, their analysis of number is not directly based on the embedding of the referents in this domain, bur on predicates with different behaviour with respect to it. The analysis that I present here is based purely on the embedding of the referent of the noun phrase into the domain of reference. In addition, I assume that, even if number is not always interpreted semantically, singular should have a semantic counterpart, too. In this section, I will concentrate on count nouns and not discuss the question of how to define a semantic singular operator that allows a semantic interpretation of singular in mass noun occurrences. I will use the terms 'count singular' and 'count plural' to refer to the singular and plural features of count nouns. In Section 1 2 I will generalize my analysis to include mass nouns.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Based on this representation, we could not combine Leute with a numeral as in drei Leute . To allow for the representation of drei Leute in a compositional manner, Kritka has to assume that there is a number- independent representa tion of Leute in the lexicon, whose singular form is not realized in German. The same holds for Eltern and Geschwister, and there would be no explanation for the lack of the corresponding singular forms. (Morphologically, all forms would be regular if they were derived from the singular forms Leut , Elter, and Gesc lzwist or Gesc lzwister, but these singulars simply do not exist in German.) According to this, Kritka (pers. comm.) assumes that the plural morpheme has no meaning at all and associates the plural operator with the determiners (in bare plurals not occurring directly in the surface structure) which trigger plural agreement. But according to this claim, he would have to assume that determiners like t lze and n o are ambiguous in three ways (singular, plural, mass) and E as doubly ambiguous (plural and mass), e.g. the,g: /..Z [MAX(Z( 1 ))], tlze pi: AZ [MAX(/..x 3 r [Z(r)(x)])], t lzemass: AX [MAX(X)] . So this assumption is not consistent with the description of the determiners given in Krifka ( 1 989a). I am not convinced by Krifka's analysis, which leaves the semantics of number not to the noun but to the determiners. And even his weak interpretation of plural (including r � 1 instead of r > 1 in the representation) proves inadequate for pluralia tantum. If the plural predicates include the atoms in their denotation and if it were a purely pragmatic choice (as he argues) that we use the singular form when we refer to atomic individuals, it should always be possible to use these pluralia tantum to refer to atomic individuals which is not the case.
Ig
Semantics of Number
I do not assume that the semantic core of a count noun is restricted to the atomic part of the universe as Link ( 1 98 3) did for (distributive) count nouns and L0nning ( 1 989) did more or less implicitly. I assume, rather, that the core meaning of all nouns are closed with respect to u as described in (2 3) and (24) (c£ Allen 1 980: S 54; Scha 1 9X 1 ; Link 1 9X 3 with respect to mass nouns; Krifka 1 9X9a). Krifka ( 1 9X9a, b) suggested that verbs should be analysed in a corres ponding way. (2 3) predicative: GIRL(e, t) [APJ P(a) A P(b) � P(aE9b)
As for the problems of Link's and Krifka's approach to the semantics of plural, it is obvious that the main role of number is to specify the complexity of the referential argument of the noun (i.e. the referent of the noun phrase if it refers). Thus the semantic core of number should be given by a unary predicate as in (2 5 ). (2 5 ) Count plural: cmplx Count singular: at
As we have seen above, at and cmplx involve countability of the argument, which is a result of my restriction to count nouns. To fit this into a compositional approach to semantics which takes care of the morphological structure, one can specifY the number operators, which are applied to the numberless representation of the noun phrase, as in (26). The combination of the number operators with relations is carried out by functional composition.13 at and cmplx are not necessarily realized as special morphemes attached to a noun. They may, as we will see later on, be included in the meaning of determiners, quantifiers, or nouns (e.g. pluralia tantum). My main claim is that wherever we interpret count singular or plural at all, it should be done by using these operators. The operators defined here restrict the denotations of the noun to special parts of the domain of reference. (26) Count plural: ct-pl :- A.PA.x [cmplx(x) A P(x)] Count singular: ct-sg :- A.Ph [at(x) A P(x)] This leads to translations from English to LPN as given in (27). Whether a phrase like the sons ofPe ter an d Ann refers to their common sons or to the sum ofPeter's sons and Ann's sons is not a matter of different readings of the phrase, but only a question of the kinds of models in which this phrase is evaluated. Based on the analysis given, it is possible to have lexical entries of Leu te ,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(24) relational: SON(e, (e, t)) 12 [ARJ R(a)(b) A R(c)(d) � R(aE9c)(bE!1d)
Carola Eschenbach 1 9
NL expression singular form
plural form
girl(s) group(s) son(s) son(s) of Peter son(s) of P'ercr and Ann
Ax [cmplx(x) 1\ GIRL(x)] Ax lcmplx(x) 1\ GROUP(x)] 1-yh [cmplx(x) 1\ SO N (y)(x)] Ax lcmplx(x) 1\ SON(P)(x)] h [cmplx(x) 1\ SON(PEB A)(x)]
h [at(x) 1\ G I RL(x)] Ax [at(x) 1\ GROUP(x)] 1-yh [at(x) 1\ SON(y)(x)] Ax [at(x) 1\ SON(P)(x)] h [at(x) 1\ SON(PEB A)(x)]
To describe rhe reciprocal use of relational nouns, I will use an operation rec , which maps binary relations to predicates (i.e. an operation of type ((e, (e, t)), (c, t))). It is defined by [Dr•c]. The result of applying rec to the relation SISTER is given in (2X)_I4 [Dr•c] rec : /.RA.z [cmplx(z) 1\ V x , y < z [x l y � R(x)(y)]] (2X) S I STER •c - Az [cmplx(z) 1\ V x, y < z [x L y � SISTER(x)(y)]] -
'
The possible interpretations of relational nouns that are nor anti -symmetric are given in (29). The plural form of these nouns is ambiguous between the (2 t)) NL expression
singular form
plural form
sisrer(s)
/.yh [at(x) 1\ S I STE R(y)(x) ]
1-yh[cmplx(x) 1\ S ISTE H. (y)(x) ] Ax [SISTEw•c(x) ] h [cmplx(x) 1\ SI STER(P)(x) ]
sisrer(s) of Peter Ax [at(x) 1\ S ISTER(P)(x)]
relational and rhe reciprocal reading. The only nouns I know that do not fulfil this strict condition in the reciprocal occurrence are nouns like cousin and neighbour. For example, if we have three people, two of whom live together and the third is their neighbour, then we might refer to them as the ne��hbours. Bur I think that this should not lead to a weaker interpretation of the reciprocal use, but be seen as due to the fact that live together specifies a stronger relation than ne��hbour. The need for the strong reciprocal reading is obvious in that the neighbours cannot be used to refer to the inhabitants of a city, even if there is a chain with respect to the neighbour relation between each pair of inhabitants. As pointed out by a referee, the twins in (3o) docs not obey the strict reciprocal interpretacion. But it seems to me that in this case twin is used predicatively. I
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Eltern , and Ceschwister which already include rhc meaning of plural in rhc semantic representation.
20 Semantics of Number
would not be surprised if, of some twin-pairs, only one attended the meeting, or if triplets, etc. were expected to come, too. (3o) The twins gathered in the big stadium for their annual twin -festival.
[Drec 1 ] rec : = /.R/..z [:3 x [x < z] 1\ 'V x, y < z [x l y => R(x)(y)JJ [Drec2] rec : = f.R/..z['t/3 x, y < z [x l y => R(x)(y)ll
Though some languages have a more elaborate number system, I will not go further into their analysis. The approach presented here can be extended to describe other number categories. The only additional category that may be useful in the analysis of English and German is the dual. The expressions both , either, neither, and pair may be seen as a lexicalized version of this category, which appears in some other languages as a morphological category (e.g. Greek; cf. Humboldt 1 X27). If we reconsider what was said in Section 4, we see that expressions that agree in number with a noun or noun phrase usually take the referential argument of the noun to fill the external argument position. The approaches of Verkuyl, Krifka, and Link do not provide a basis for explaining the number feature of verbs. Although the subject-verb agreement has nearly become a pure syntactic restriction in languages such as German and English, its semantic basis is due to the double specification of the complexity of the referent of the subject noun phrase. This background is still obvious in coordination (29a), the treatment of group nouns in British English (29b), and pseudo-quantifiers like the German eine Menge or a lot (29c) (cf. Hoeksema 1 9X 3 ; Barker 1 992; Pollard & Sag 1 99 1 ).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
We might assume that the operation rec represents the semantic core of a lexical rule of reciprocals that is applicable to all relational entries, i.e. not only to relational nouns but also to transitive verbs and prepositions. 1 5 It is not necessary to restrict it to those relations which allow for symmetry. Anti-sym metric relations are excluded because they do not allow for a proper clique. But we could also mark anti-symmetric relations in the lexicon to avoid using the operation in these cases. On the other hand, this operator is meant to describe how reciprocal uses of relational nouns are interpreted. Whether it has to be assumed as being applied to relational nouns on the semantic level mainly depends on the general framework of meaning composition assumed. The reciprocal use of relational nouns is restricted to complex objects (i.e. in the plural form). A similar restriction applies to reciprocal forms of verbs. I regard this as a consequence of the semantics of reciprocals: the quantification has existential impact, i.e. is not allowed to yield truth because of emptiness of the domain. We can reformulate [Drec] as [Drec l J or [Drecz], where 't/3 is a general quantifier with existential impact.
Carola Eschenbach 2 1
(3 1 ) a. John and Mary walk home. Tea and coffee i s in the kitchen. The committee is old. b. The committee are old. c. Eine Menge Kinder spielen im Ho( Eine Menge Sekt ist getrunken worden. A lot of champagne has been [A lot of children play in the yard. drunk.]
9 THE M E A N I N G O F THE D E F I N ITE ARTICLE Some natural language determiners can occur in front of both mass nouns and count nouns. It seems reasonable to represent them on the semantic level as a single operator on nouns. One of these determiners is the definite article (c( Lobner 1 9X s: 2Xo). Lobner described the behaviour of the definite article as follows: The children refers ro the entire complex object ro which children applies the child to the entire obj ect to which c/,ifd applies (which is necessarily only one child); and the snow to the entire object to which snow applies. (Li:ibner 1 yX5: 2X2).
We have already seen that Krifka ( 1 9X9a) had to assume two or three different representations of the definite article, one which combines with singular count nouns and one which combines with plural nouns and mass nouns. Link ( 1 9X 3, 1 99 1 ) assumes four different semantic operators to represent the definite article: singular (l), plural (o), proper plural (o"), and mass (J.A.). In Link ( 1 99 1 ) he defined l and o " as restrictions of o. L0nning ( 1 9X9) discusses Link's o operator and its consequences with respect to the model theory. On the semantic level, they associate this operator with the generalized sum operation. Thus it corresponds in the main respects to the operator o as defined here. There are mainly two problems with this analysis. The function sum allows for sum(S) � S. It will always be the case that P"(o(P)) holds (if o(P) refers), bur in this approach it is not possible to handle predicates which do not have the cumulative reference properry (c( Krifka 1 9X9a: 7 4 f(). The second problem with this analysis is that the plural defi n ite article is not
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As a consequence, it does not matter whether we assume that a number feature which occurs as a consequence of agreement (with referring noun phrases) is represented or not: representing it semantically yields redundancy, bur no conflict. Though the determiner of a noun phrase influences the status of the referential argument of the noun, it may determine whether number should be interpreted or not. This will be elaborated in Section 1 1 , · where I discuss quantificational determiners.
22 Semantics
of Number
applied to one of the plural predicates *P or 0P, but to the singular predicate P. This does not in any way correspond to the facts in languages like English, where there is only one definite article, which is not sensitive to countability and number, and the nouns are the expressions which bear the number feature.
(32) [b(P)] = max,.([P])
Assuming that we represent the definite article by 0 and the number distinction as described above, the representation of some definite noun phrases are listed in ( 3 3). The evaluation in the last two columns is with respect to the models M I and M2. M I = (((a, b, d, e, f, g, h, j . . .) u, ( a, b, d, e, f, g, h, j), u), [.] M 1) with [JOHN]M 1 = j, [BOY]M 1 = {a, b, a u b), [SON]M 1 = {U. a), U, b), U, a u b)), [SISTER] M 1 = ( U, e), U. h), (h, e), (e, h), (f, g), (g, fjju and from this [SISTERrec] M , = {e u h, fu g) M2 = (((a, d, e, f, j . . .) u, ( a, d, e, f, j), u), [.] M2) with [JOHN] M2 - j, [BOY] M 2 - {a), [SON] M 2 = (U, a)), [SISTER] M2 = {(d, e), (e, d), U, fjju and from this [SISTERrec]M2 = (d u e) . (3 3) Expression
LPN
all
the boy
b(ct-sg(BOY)) b(ct-pl(UOY)) b(ct-pl(SISTEH'«)) b(ct-sg(SONUOHN))) b(ct-pl(SONUOHN))) b(ct-sg(SISTERUOHN))) b(ct-pl(SISTEI�UOHN)))
maxd[ BOY] n A) max5 ([uov] n (C:\A)) max5 ( SISTEw••] n (C\A)) max., ( SON] ([JOHN] ) n A) max6 ([SON] ([JOHN] ) n (C\A)) maxs ([SISTER] ([JOHN]) n A) max,. ([SISTER] ([JOHN] ) n (C\A))
the boys the sistl'rs the so11 of John rhe so11s of Johll the· sister of John the sisters of John
models
[
undef a aub
undef
undef d u e· undef a aub
undef
undef cuh
undef
If we want to describe definite mass noun phrases in this framework, we have primarily two possibilities. Either we assume that the domain of masses is structured by the same relation �; (as I did in the description of the LPN strucmres), or we assume that the ordering relation in max., is a parameter
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Regarding the closure properties of noun denotations as described above, the most natural way to get the 'entire object' interpretation relative to a predicate in a semi- lattice structure is to use the maximum function relative to the underlying ordering relation �; (c( Krifka I 989a: 7 5 , who also used max., but defined it on the basis of sum, and Kadmon I 990). Correspondingly, I will use the operator 0 to represent the definite article. The logical type of this operator is ((e, t), e) (i.e. it maps predicates to entities); its model theoretic interpretation is given in (32).1 6
Carola Eschenbach
23
which is contextually specified: the sortal properties of the elements of the set determines which ordering relation is used. 1 7
10 I N D E F I N ITES A N D NUMERALS
It has been widely recognized that only singular count nouns allow the indefinite singular article while plural count nouns and singular mass nouns allow for bare forms. If a noun (like apple) can occur as a mass noun and as a count noun, an occurrence of the indefinite article indicates that it should be interpreted as a count noun, while a bare form indicates that it should be interpreted as a mass noun. Krifka's suggestions on the meaning of count nouns give an explanation for this difference but get into difficulties with all unspecific determiners, as I showed above. Because the indefinite article a ( n ) only allows singular nouns, I assume that the singular feature of the noun in this context is due to agreement with the article and not interpreted. The basis of the indefinite singular article is at (as in (3 s) ) i.e. the meaning of singular in connection with count nouns. This explains why mass nouns do not occur with the indefinite article and nouns which allow for count and mass occurrences are interpreted as count occurrences in connection with it.'8 ,
(3 5) a (n )
/.N/.x !at(x) 1\ N(x)]
If we assumed that the effect of bare forms is semantically determined, we would have to assume that there is an empty determiner with special semantic effects as in ( 36). This assumption is quite problematic because there is no syntactical or morphological counterpart to this operator (c( Lobner 1 9H6).
(36)
E
/.Nh [-.at(x) 1\ N(x)J
It is also not clear in which contexts this restriction on bare occurrences really holds. Regarding bare nouns in prepositional phrases like out of town/bed , at
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I n English (and German), indefinites can appear as bare forms as i n (pa, b), or with a determiner from a certain class, which includes the indefinite articles a ( n ) and some (pc-f), numerals (pg, h), and a few more. (34) a. Mary puts water into the soup. b. Bill likes books . c. John owns a donkey . d. Some man entered the pub. e . Mary puts some water into the soup. f Some book belongs to me. g. One book belongs to me. h. Three boys entered the pub.
24 Semantics of Number
(37) some
t.N/..x [N(x)J
The combination of a numeral with a count noun usually yields a stronger restriction on the complexity of the object referred to. It has become quite common to analyse them as generalized quantifiers or to map them directly to integers or rationals (cf Krifka 1 989a). I will analyse numerals as operators with behaviour similar to the number operators (38) (cf Verkuyl 1 98 1 ; Bunt 1 98 5). (38) one three
t.N/..x [ctbl(x) 1\ quantity(x) - 1 1\ N(x)] I.NI.x [ctbl(x) 1\ quantity(x) - 3 1\ N(x)]
The function quantity, which maps entities to natural numbers, can be defined as in (39). 1t has some similarity to Krifka's NE function, and to attain a more elaborate analysis of various numeral constructions (e.g. measure and classifier constructions) one should also include the possibility of modifYing it with respect to various criteria of counting or measuring (cf Eschenbach 1 99 1 ). The main difference between my approach and Krifka's is that he assumed NE to be a part of the noun meaning, while I include quantity in the semantics of the numerals. (39) [quantity] :-
{ C --- IN
x ..... l {a E A ! a �; xJ I
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
home/sclzool!work , in bed!sclzool/church!prison , in part , in coordinating structures like kitchen and bath , week after week , day in, day out , or in complex phrases such as in time of war, in point/case of, sort ofthing/person , kind ofsin/situation , it is not clear whether these bare nouns should be regarded as mass nouns. It is rather a question of whether the bare nouns in these contexts refer to entities at all. I think that the restrictions on bare forms have to be analysed with respect to their referential behaviour and discourse function, but such an analysis is beyond the scope of my discussion. Allen ( 1 980) gave an explanation of the effect of bare forms as referring expressions on a pragmatic basis: 'If his listeners do not already know the countability of the NP reference, the speaker must make it known to them' (Allen 1 980: 542). This rule is only applied in connection to indefinite noun phrases. Accordingly, using a bare singular form signals that the referent is not countable. The main difference between the indefinite noun phrases with the determiners a ( n ), some or without any determiner corresponds to the extent to which these phrases are used referentially, and I assume that this should be handled on a level different from semantics. So I take (3 5) as the semantic representation of a (n) and (37) as the semantic representation of some.
Carola Eschenbach 2 5
Numerals can also occur in partitive constructions as three '?[the boys . Assuming (4o) as the semantic counterpart of the partitive of, we get regular inter pretations of such phrases as in (4 1 ). (4o) of (4 1 ) '?[the boys three ofthe boys
1-.y/..x [x :::;; y] 1-. [x :::;; o(ct-pl(BOY) )] Ax [ctbl(x) 1\ quantity(x) �
I I
3 1\
X :::;; o(ct-pl(BOY))]
Q U A N T I F I E RS
(.p )
N no
every each hoth all each of' both ,1· all 'l
def NP
+
+ + + +
smg. counr
pl. counr
mass
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
+
Due to their fle xibility in use, I present two lexical entries of all and both . This ambiguity is due to different logical types (((e, t), ((e, t), t)) and (e, ((e, t), r))). In both cases the first entry can be derived by functional composirion of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
To conclude my remarks on number, I will give a brief sketch of how quantifiers can be handled in my framework. I will concentrate on each , every , all , both , and no , and restrict myself to the analysis of their occurrence as determiners of nouns and in combination with definite noun phrases (both and all can appear either before or after a definite noun phrase; cf all the man , weal/ , they both ). (42) describes the differences in disrribution of these quantifiers. The first column indicates whether the quantifier can be combined (with bare) nouns and the second column indicates whether it can be combined with definite noun phrases. Columns 3-5 describe their restrictions with respect to the structural basis. This table shows that the resemblance of plural count nouns and mass nouns is greater than the resemblance of singular count nouns and mass nouns. There seems to be no determiner that excludes the plural count nouns alone. This justifies the assumption that the number of mass nouns should be handled differently from the singular of count nouns.
2.6 Semantics of Number
(43) no each , every each of all
all of
A.NA.P V y, x [(N(y) 1\ x �y) � --. P(x)] A.NA.P V x [at(x) 1\ N(x) � P(x)] A.yA.P V x [at(x) 1\ x < y � P(x)] A.NA.P [P(b(N)) 1\ V3 x < b(N)[P(x)]] ( � A.NA.P [3 x [x < b(N)] /\ V x �b(N)[P(x)]]) A.yA.P [P(y) 1\ V3 x < y [P(x)]] ( A.yA.P (3 x [x < yj /\ V x � y [P(x)]]) A.NA.P [ctbl(b(N )) 1\ quantity(b(N)) = 2 1\ V x < b(N) [P(x)]] A.yA.P [ctbl(y) 1\ quantity(y) = 2 1\ V x < y [P(x)]] =
both both (of)
Since I am not a native speaker of English, I am not aware of any clearly semantic difference berween each and every . In the Oxford Advanced Learner's Dictionary of Current English ( 1 974), the difference is described as: 'When every is used . . ., attention is directed to the whole; when each is used, attention is directed to the unit or individual.' If this is correct, one could assume another representation of every , which lies in berween each and each of and shows a great resemblance to all , too. (44) every
A.NA.P V x [at(x) 1\ x < b (N) � P(x)]
In this cse, both turns out to be a more specific version of every as well as all . This might explain the fact that neither o f them i s used if there are only rwo (atomic) individuals in the range of the quantifier. However, I do not . assume the given analysis of the English quantifiers to be complete. A much finer analysis has to be made in order to caprure all the differences berween their uses (cf Vendler 1 967).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
second entry with b. Correspondingly, the representation of each of can be regularly derived from the partitive of and each In contrst to no and all , the determiners each , every and both determine the number of the following noun, so I assume that the number of the noun is a result of agreement in these cases and will not be interpreted. The different behaviour of the quantifiers can be seen as a reflection of their semantic structure (43). Each and every quantity over the atomic domain of N, which is reflected as the restriction to singular count nouns. The representations of all (of) include the general quantifier with existential impact, and will yield no truth-value if the respective domain is empty. Consequently, they are not applicable to atomic individuals (i.e. entities that have no proper part). All N makes a statement about the N and quantifiers over the (proper) parts of the N; thus all is only applicable to objects with proper parts, not allowing singular count nouns. This representation takes care of the ambivalent behaviour of all , which can i n many cases replace the definite article. Though I did not include a special operator to describe the dual, my description of both includes ctbl(.) 1\ quantity(.) - 2, which ought to be replaced by a basic predicate dual.
Carola Eschenbach 12
2.7
W H A T I S C O M M O N TO M A S S A N D S I N G U LAR COUNT N O U N S ?
at(x) cr- sg: -.3 y � x [at(y)] mass: sg � cr-sg V mass: at(x) V -.3 y � x [at(y)] - at(x) V (-.at(x) 1\ 3 y < x [at(y)]) - at(x) V -.3 y < x [at(y)] - -.3 y < x [at(y)] -.
Consequently, the approach described here is open to the assumption that singular is ambiguous between ct-sg and mass as well as to the assumption that singular is unambiguous but underdetermined.
1 3 CONCLUSION I have described a semantic treannent o fnumber which is based on the assump tion that number always specifies which properties the referent of a noun phrase has with respect to the structure of the domain of referents. Singular and plural of count nouns are both represented as operators on the core meaning of rhe noun, which is assumed to be indifferent with respect to the complexity of the referents. The basis for the representation is given by assuming a basic pre dicate at, which holds for the proper individual atoms in the domain of refer ents. While this predicate directly corresponds to the restriction that singular imposes on count nouns, the representation of plural is more complex. This corresponds to the observation that in many languages plural is marked with respect to singular. Taking an objective criterion of atomicity for granted, the analysis outlined in this article presents a simplified view of the phenomena of number and numerals. It is neither capable of handling units of measurement such as litre , and classifiers like head in three head ofcattle , nor can it systematically cope with
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
As we have seen, the behaviour of plural count nouns and mass nouns is much more similar than the behaviour of singular count nouns and mass nouns. Therefore it seems quire strange that in languages such as German and English mass nouns are syntactically singulars (i.e. agree with singular forms of verbs). Bur in the presented framework we can recognize what is common to the referents of the syntactically singular noun phrases (general singular): they do not allow for proper atomic (individual) parts. The difference lies in how this condition is fulfilled: count singulars do not allow for any proper parts and mass nouns do not allow any (even improper) atomic individual parts.
2B
Semantics of Number
count noun occurrences of mass nouns like three wines . But this is another story, which has to be told another time. CAROLA ESCHENBACH FB Informatik
Received: 3 1 - 1 0-9 1 Revised version received: 28-7-92
Universitiit Hamburg Bodenstedtstr. 1 6 D-2000 Hamburg 50 Germany
Acknowledgements
N OT E S 1
2
�
C( Lohner ( 1<.)X5), who gives an elaborate analysis of referential properties of various noun classes. He uses the term 'sorral' for what I call 'predicative'. For ease of the further discussion, I will concemrate on those relational nouns which correspond to binary relations berween objects, bur the suggestion presented on meaning of number will not depend on this point. Since it is nor relevant for my discussion, I will ignore the situational argument, which Lohner assumes for every notlll, and his distinc tion berween functional and relational nouns. C( Kuhn ( 1 ()Xl) on the relation berween mass nouns, count nouns, and collective nouns. C( Dressler ( 1 ()6X) on semantic categories of verbal number. Link ( 1 ()X3: 306) states: 'First of all, let us rake seriously the morphological change in pluralization, which is present in many natural languages. and introduce an operator working on • -place predi·•·.
cares P, which generates all rhe individual sums of members of the extension of P.' The operator of Link is his (improper) plural operator. What he did not consider is that in some languages, such as Larin, there is no such morphological distinc tion berween singular and plural: both forms are derived by suffixes on a number-independenr stem. The differ ence berween languages which do or do nor have a number distinction (e.g. English vs. Japanese) seems to be de scribed appropriately by assuming rhar the noun meaning in both cases respects rhe same structural restrictions, and that languages with number distinction have additional operators applicable to rhe noun meaning. A few examples of plural mass nouns such as news and measles seem to hint that plural, too, should be regarded as sys tematically ambiguous. But since they seem to be exceptions in the languages under consideration, I hope to cover them by special lexical entries. ·•·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The research reported in this paper was supported by the Deutsche Forschungsgemeinschafr (DFG) in rhe project 'Gruppierungs- und Abgrenzungsprozesse'. Thanks to Kari Fraurud, Christopher Habel, Michael Herweg, Wolfgang Heydrich, Manfred Krifka, Claudia Maienborn, Geoff Simmons, Dieter Wunderlich, and three anonymous referees for comments and discussions about spoken and written pre-versions of this paper, although I was nor able to react to all their poims. l am also grateful to Geoff for his help in my struggle with rhe English language. Needless to say . . .
Carola Eschenbach
�
�
13
I .J.
only a few ways ro weaken this; cf John is his best adviser and my mother's daughter . Bur ir would be quire strange to refer to John by usingjo/m's best adviser, and even my mother's daughter yields a special effect. The question is whether this phenome non should be retlecred on the semanrical level by restricting the denotation of relational nouns ro [IJ If we do rhis, no relational noun rhar allows symmetry could semantically be classified as speci fying a transitive relation. If we do nor treat rhe disjoinrness restriction by re stricting the denotation of relational nouns, we could describe it as a general restncnon on referring expressions, which has to be handled at another level. and describe the closure conditions as is done here. If� is of rype (H. C) and 11 is of rype (A, H). then � ' 11 is of rype (A, B). I f � is A.x a(x) and 11 is A.y f3(y), then � ' 11 is A.y a(f3{y)). (A.PA.x [at(x) 1\ P(x)l) ' (A.yl..x [ SON{y){x)J) - A.yA.x Jat(x) 1\ SON(y)(x)J As long as objects which are built up from finitely many atoms are concerned. rhe denotation of rhe derived predicate can be described as [ RREc] (max.;(Mu) IM s;;; A is a proper clique w.r.r. [ R]). have borrowed rhe term 'clique' from graph theory. A clique with respect ro a relation is a set in which every pair of different individuals is in this relation. A proper clique is a clique wirh more than one elemenr. lr has been observed rhar not all reci procal phrases force the srricr imerprera rion described above. Fiengo & Lasnik ( 1 97 3 ) and Langendoen ( 1 97X) discuss examples such as: (i) The men and the IV
1
5
anMher.
{ii) The boxes are nested inside ,me another. However. Fiengo &· Lasnik ( 197 3) observed rhat reciprocal descriptions of states are usually interpreted strictly {excluding stares as described in (ii)). The reciprocal occurrence of relational nouns
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Cf Link { 19X 3, 1 99 1 ); Lmming { 1 9X9); Krifka ( 1 9X9a). The basic structure ourlined here is in its mereological part based on a slightly weakened classical system as described by Simons ( 1 9X7: Ch. 1 , 'Minimal Extensional Mereology plus Binary Sum Principle'). The reason nor to employ rhe classical system (which is complete with respect to the sum of arbitrary non-empry subsets of the domain) is rhar I do nor need irs power. 7 I f a structure (D, u ) includes an element n E D, such rhar X 1\ II U X 't/ X E [) lx U n xJ, rhen n is called a 'neurral elemenr' of (D, u ). If D holds only one elemem, rhen rhis is necessarily rhe neutral element. The existence of such an enriry in all other cases is excluded by JA EB 5 J. Cf Link ( 1 9X 3 ), who uses the neurral element as referent of non-denoting expressions, and Lmming ( 19X9), Bencivenga (1 9M), and Eschenbach ( 1 ()X9) for a discussion of such assumptions. X Accordi ng to rhe definition given above, Au s;;; C holds. If the definition were C : � Au, C would only include objects built up from finitely many atoms. 9 This and some of rhe following equations are valid under rhe presupposition rhar rhe names used refer to elements of A. 1 o Later on he replaced ( 1 X) by AZ, x 3 r [Z{r)(x)J (cf Krifka 1 9X9a: 22. 3 ). 1 1 Since Krifka did nor analyse English bur German, I translated his suggestions. I think rhar my poinr comes our, even if rhe correspondence between German and English is nor so srricr, E represenrs rhe empry determiner, i.r. rhe possibility of using bare forms. 1 2 A relational noun always establishes a relation between two disjoint individuals. The phrase Mary's sisters will never refer ro a collection of persons including Mary. The same holds if rhe internal argument is complex: ir is nor the case rhat Mary and Susan are sisters of Mary and Eve. even if Mary, Susan, and Eve are sisters. There are (>
29
30 Semantics of Number
rcc
•.
.
We
17
also have to assume that a maximal elemenr a l ways exists. One argumem char favours a flexible rrcatnH'Il[ nf dcfinircs in general is the
fact char supeilarive expressions do mainly occur in definite expressions, although referems of plural superlatives cannot be uniquely determined. In some languages, the superlative is expressed by the combination of the definite article and an adjective in irs comparative form. To analyse these phrases. it does nor seem to be appropriate co asume the same analysis of the definite article as above, bur to assume char the adjective deter mines the specification of the orderingparameter [ the tallest man ] max,_,� 1,. ,([ ct-sg(m an )]), [ tfu· tallest llll'll ] max.,11,., ([ ct-pl(man)]). I follow Lohner ( 1 yX7), who argues for �
r
X
treating indefinites as predicates.
RE FERENCES Al len, K. ( 1 <)Xo), 'Nouns and countability', Language , 56, 541-fl7. Barker, C. ( 1 yy2), 'Group terms in English: represeming group as acoms', Journal ,1· Semalllics . 9, 69-9 3 . Bartsch, R. ( 1 <)73), The semamics and symax of number and numbers', in J. Kimball (ed.). Syntax and Semamics , vol. 2, Seminar Press, New York, 5 1 -<) 3· Barwise, J. & Cooper, R. ( 1 yX 1 ), 'Generalized quamifiers and natural language', Lin.!.!uis tics and Philosophy, 4, 1 5<)-2 1'). Bencivenga, E. ( 1 9H6}, 'Free logics', in D. Gabbay & F. Guenthner (eds), Handbook <{ l'hilosoplrical L<�!.!iC . vol. I l l , Re id el .
Dordrechr, 37 3--P(>. M. R. ( 1 <)7 5 ). 'Some extensions of a Momague fragmem of English', thesis, dis rribured by the Indiana University Linguis tics Club. Bierwisch, M. ( 1 yXy), 'Evem-nominaliza rions: proposals and problems'. in Linxuis Bennett,
tische Studien Reihe A
194 .
Akademie
Verlag, Berlin, 1-73. · Bu m, H. C. ( 1 9H 5}. Mass Terms and Model Theoretic Semallfics . Cambridge University Press, Cambridge.
Dol l ing, J. ( 1 990), 'Plurals and collective nouns: an outlook on the semamics of group expressions', in L. Kalman & L. Polos (eds), Papers from the Second Sylllpos ium on L<��ic and Langua,\!e , Budapest, Akademiai Kiad6, r H y-2o 1 . Dressler, W. ( 1 968), Studien zur verba/en Pluralitiit: lterativum, Distributivum, lntensi vum in deralfgemeinen Grammatik, im Latein ischen und Hethitischen , Sitzungsberichr der
Osterreichischen Akademie der Wissen schaften, Philosophisch-Historische Klasse, Bohlau, Vienna. Eschenbach, C. ( 1 989), 'Komplexe I ndividuen in Referenriellen Nerzen', in D. Merzing (ed.), GWAl-89 , Springer, Berlin, 244-53. Eschenbach, C. ( 1 99 1 ), 'Symaktische und semamische Aspekte quamitativer Nomi nalgruppen', MS, Universitat Hamburg {GAP-AP 22). Fiengo, R. & Lasnik, H. (1<)73). The logical structure of reciprocal semences in Eng lish·, Fou ndations ,1-Language , 9, -1-4 7-6 H . Fraurud, K. ( 1 yyo), 'Definiteness and the processing of noun phrases in narural discourse', journal 4 Semamics . 7, 3<)54 3 3-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
seems to fir imo this category. Krifka ( 1
Carola Eschenbach 3 1
konstitution: Zur Scmantik ""'' Massentcrmen, Pluraltcrmen und Aspektklassen , Finke,
Generalized Quantiflm: Linguistic and L<\l(i cal Approaches , SLAP 3 1 , D. Reidel, Boston,
1 5 1 -Xo. Lohner, S. { 1 9X 5). 'Definites', Journal ·if· Semantics, 4, .2 79-J.26. Lohner, S. ( 1 9X6), 'In Sachen Nullartikel', Linguistische Berichte, 1 0 1 , 64-5. Lohner, S. ( 1 9X7), 'Natural language and gen eralized quanrifier theory', in P. Garden fors (ed.), Generalized Quant[{icrs , Reidel. Dordrecht, 1 X 1-2.0 1 . Lmming, J . T. ( 1 9X9), 'Some aspects o f the logic of plural noun phrases', thesis, Department of Mathematics, University of Oslo (COSMOS-Report Nr. 1 1). Massey, G. J. ( 1 976), Tom, Dick, and Harry, and all the king's men', American Philo sophical Quarterly , 1 3 , .2 , X<)- 107. Pelletier, F. J. & Schubert, L. K. { 1 9X9), 'Mass expressions', in D. Gabbay & F. Guenthner (eds), HandbMk ·if· Philosophical L.\s:ic , vol. IV, Reidel, Dordrecht, 3.27-407. Pelletier, F. J. (ed.) ( 1 979). Mass Terms: S,mll' Philosophical Problems , Reidel, Dordrechr. Pollard, C. & Sag, I. ( 1 99 1 ), 'Information based syntax and semantics, vol. .2 : Agree ment, binding and control', MS. Quine, W. v. 0. {I 91io), w,,d and Object, . MIT Press, Cambridge. Mass. Scha, R. { 1 9!! 1 ), 'Distributive, collective, and cumulative quantification ', in J. Groenen dij k, Th. Janssen & M. Stockhof {eds), Truth, Interpretation and In.fimnation , Foris. Dordrecht (GRASS .2), 1 3 1 -5X. Simons, P. (1 9X7), Parts: A Study in Ontology . Clarendon Press, Oxford. Strawson, P. F. ( 1 959). hrdi11iduals , Methuen. London. Vendler, Z. ( 1 967). 'Each and every, any and all', in Z. Vendler, Liny.uistics in Philosophy. Cornell University Press, Ithaca, NY, 70-
Munich. Krifka, M. ( 1 9X9b), Four Tlwusand Ships Passed Through the Lock , SNS-Bericht X9-41i, Ttibingen. Krifka, M. {1 990), 'How to get rid of groups, using DRT: a case for discourse-orienred semamics', MS, Austin, Texas. Kuhn, W. ( 1 9X.2), 'Kollektiva und die Tech nik KOLLEKTION am Beispiel des Deutschen', in H. Seiler & Ch. Lehmann (eds), Apprehension: Das spracldiche E.rfm�>en von Gegenstanden , Part 1 , Narr, Tubingen, X4-97. Landman, F. ( 1 9X9), 'Groups I & II', Lin,l!uistics and Philosophy, I l , 5 59-Ms , 7.23 -44. Langendoen, D. T. ( 1 97X), 'The logic of reciprocity', Linguistic Inquiry, 9, 1 77-97· Link, G. { 1 <)X 3), The logical analysis of plurals and mass terms: a lattice theoretical approach'. in R. Bauerle, Ch. Schwarze & A. von Stechow (eds), Meaninx, Use, and Interpretation of Language , de <)6. Verkuyl, H. J. ( 1 9X 1 ), 'Numerals and quanti Gmyter, Berlin. 30.2-.23. fiers in X-bar-syntax and their semantic Link, G. ( 1 99 1 ), 'Plural', in A. von Stechow & i nrerpretation ', in J. Groenendijk, Th. D. Wunderlich (eds), Semantik . de Gmy Janssen & M. Stokhof(eds), Formal Methods ter, Berlin, 4 1 X-40. in the Study .if' Lanxua,s:e, vol. 2., Mathe Link, G. { 1 9X6), 'Generalized quanrifiers and matisch Cenrrum, Amsterdam, 5(>7-99plurals', CSLI Report No. CSLJ-X(>-67, Stanford. Also in P. Gaerdenfors (ed.),
B
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Hausser, R. { 1 974), 'Symax and semantics of plural', Papers from the 1oth Regional Meeting of the Chicago Linguistic Society. 234-47· Hoeksema, J. ( 1 <)X 3), 'Plurality and coi�unc tion', in A. G. B. ter Meulen (ed.), Studies ir1 Modeltluwetic Semantics , Foris Publica tions, Dordrecht, 6 3-X 3· Hoeksema, J. { 1 9XX), The semantics of non Boolean "and" ·, Journal •!{Semantics , 6, 1 940. Humboldt, W. von ( 1 X.1.7), 'Ueber den Dualis', in W. von Humboldt, Ober die Sprache ( 1 9Xs) (ed. by J. Trabanr), dtv Munich. Kadmon, N. ( 1 990), 'Uniqueness', Linguistics and Philosophy, 13, .273-3.24. Krifka, M. ( 1 9X9a), Nominalreferenz und Zeit
jtmmal oJScmamics
ro:
© N.I.S. Foundarion (1993)
3 3-6�
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
RE I N HARD B L U T NER Max Planck Gesellsc/zafi, Berlin
Abstract
I
I NTRO D UCTION
There are two rival theoretic approaches which have elicited promising new lines of research in natural language semantics. The first one is the Generalized Quantifier Theory (GQT) initiated with Barwise & Cooper ( I 98 I ), which has continued the approach of Montague ( I 974) . This line of theorizing aims at a uniform treatment of all NPs as generalized quantifiers, denoting sets of properties of individuals. GQT has been especially useful in clarifying the algebraic properties of natural language quantifiers. The second of the two rival approaches is represented by semantic theories such as Kamp's ( I 98 I ) Discourse Representation Theory, Seuren's ( I 986) Discourse Semantics, or Heim's ( I 982) File Change Semantics. In the present paper, let me take Kamp's ( I 98 I ) Discourse Representation Theory (DRT) as the prototypical example of this family of paradigms, and let me restrict myself to reflections on this theory. DR f (as well as the related theories just mentioned) explicitly rejects the uniform treatment of all NPs as generalized quantifiers and instead proposes a division of labour between (i) a quantificational NP-semantics (quantifiers as restricted + unselective binders) and (ii) a nonquantificational discourse semantics (which translates indefinite NPs as discourse markers + conditions imposed on them).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The cenrral topic to be discussed in this paper is the definiteness restriction in !here -insertion contexts. Various attempts to explain this definiteness restriction using the standard algebraic framework are discussed (l:larwise & Cooper 1 <)X 1 ; Keenan 1 <.i7X; Milsark 1 <)7�; Higginbottham 1 <) X7; Lappin 1 <)XX) and the shortcomi ngs of these attempts are demonstrated. Finally, a new approach to the inrerpretation of existential Ihere he -sentences is developed within the framework of Groenendijk & Stokhof's ( 1 990) Dynamic Monrague Grammar. This approach makes use of a varianr of Partee's ( 1 <)XI\) 'type-shifting'-operator l:lE and it overcomes the shortcomings of the rival analyses. The general conclusion is that Dynamic Montague Grammar has applications other than those which prompted it and advantages other than those Gmenendij k & Stokhof claim for it.
34
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
2 E X I S TE N T I A L THER E BE- SE N TE N C E S The study of the grammatical and semantic properties of existential sentences in English such as ( 1 ) There are ghosts .
(2) There is a man in the garden . ( 3 ) There is a man standing on the corner. has fascinated the transformational linguistics, formal semantics, and discourse pragmatics in various periods of their development. In this section I will highlight some recent results which are of relevance for questioning the general validity of the standard algebraic framework as used in GQT.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The recent unification of DRT and GQT within the framework of dynamic intensional logic (Groenendij k & Stokhof 1 990) has explicitly challenged this division of labour. The 'fall back' of dynamic intensional logic to a uniform treatment of all NPs (NPs as so-called dynamic generalized quantifiers) and the emergence of a compositional semantics for discourse may seem conservative. For several reasons I think that this is not the case. Moreover, I am convinced and would like to argue that Groenendij k & Stokhof's dynamic GQT is notj ust a formal exercise on compositionality and compatibility of frameworks, but may very well have empirical import particularly in domains where neither DRT nor GQT are successful if taken per se . The phenomenon to be discussed in some detail is the definiteness restriction in contexts of there-insertion. The results support the claim that the theory of dynamic generalized quantifiers forms a new kind of approach to recalcitrant problems of natural language semantics. The paper is organized as follows. Section 2 gives a review of several attempts to explain the definiteness restriction of existential there be -sentences using the standard algebraic framework and it demonstrates the shortcomings of these attempts. Section 3 introduces Groenendij k & Stokhof's ( 1 990) dynamic GQT and it presents the translation of a small fragment of English into the dynamic intensional logic. Finally, in the constructive Section 4 a dynamic approach to the interpretation of existential there be -sentences is provided. I hope to demonstrate that this account gives a true explanation of the definiteness restriction involved in this construction and overcomes the shortcomings of the rival attempts discussed in Section 2.
Reinhard Blumer
3S
2. 1 The difiniteness restriction in there-insertion contexts
It is a long- standing observation that indefinite NPs narurally occur in existen tial there -contexts, whereas definite expressions (and quantificational ones in the narrower sense) are only marginally acceptable.
(4) (a)
There is
(b) *There is
the boy everyone my brother
) )
hungry .
hungry .
Generally, it can be said that contexts of there -insertion represent one of the linguistic environments exhibiting a definiteness restriction (DR), i.e. in there insertion contexts indefinite expressions are typically preferred} Since (in)definiteness is at least partly a semantic property of expressions it seems appropriate to explain the DR involved in terms of a semantic theory for narural language. The rather extensive semantic literature on (in)definiteness appears to provide a fairly clearcut characterization of the distributional restrictions. For example, Milsark ( 1 974) introduced the terms 'weak' and 'strong' as labels for two complementary classes of NPs, roughly those that can and those that cannot appear in the postcopular subject position of existential there be - sentences. Furthermore, he proposed a semantic characterization of these classes: weak NPs are those specified by a non-cardinal, quantificational one. Although Milsark ( 1 974) gives only an informal treatment of cardinality, he makes quite clear what is the main motivation for providing such a semantic characterization. It is precisely the possibility of reducing the DR with the help of these semantic features to a semantic deviance. In this vein, Milsark suggests reducing the deviance of there be -sentences with strong subjects to a case of vacuous quantification, which exhibits a well -known type of semantic devi ance. There is a widespread inclination to follow Milsark in his two basic methodological insights: (i) to describe the distributional classes involved in the DR in terms of concepts provided by an appropriate semantic theory for narural language; (ii) to explain the DR by reducing this restriction-with the help of those concepts-to some common type of semantic (or pragmatic) deviance. Furthermore, the general methodological background suggests following the Fregean notion of compositionality. (iii) by analysing the meaning of existential there be -sentences within a com positional framework of semantic interpretation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
j j
a boy someone no donkey
36
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
2.2 The standard algebraicframework and there-insertion contexts
There are rwo major points of disagreement among existing analyses of there insertion sentences. The first concerns the question as to whether the there be part of existential there -sentences does attribute any meaning at all to this construction. Some authors claim it does not (e.g. Higginbotham 1 98 7; Lappin 1 988; Keenan 1 987), others claim it does: Milsark ( 1 974) suggested an existential quantifier and Barwise & Cooper ( 1 98 1 ) the universal property THING. The second point of disagreement concerns the structure of the coda (the nominal element and the optional predicate governed by the existential verb). Three structures have been proposed for sentences like ( s), indicated in (6):
( s) There is /are Det student (s) hungry . (6) (a) There is/are [ N r Det student (s)] [xr hungry] (b) There is/are [N P D et [ N . student (s) hungry ]] (c) There is/are [5c[Det student (s)] [hungry]] In (6a) NP and XP do not form a constituent but arc both sisters of the verb (the 'NP-XP' analysis, cf. Milsark 1 974). According to (6b) the coda is a structure of postnomnial modification (the 'bare N P' analysis, cf. Jenkins 1 97 s; Williams I 984) . According to (6c) the coda is a 'small clause' in the sense of Williams (I 975 ) (the clausal analysis, cf. Stowell I 98 I ; Sa fir I 982 ). Taking into consideration these major points of disagreement let us discuss now some problems chat arise if the standard algebraic picture of Barwisc &
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Although alternative methodologies are conceivable-Sa fir ( 1 982, 1 987), for instance, insists chat at least part of the DR in there -insertion contexts must be explained with a property chat is relevant at S-scruccure only2-let us assume that Milsark's view of the status of the DR is basically correct. Now the question arises as to the appropriate semantic theory by means of which (i) the classes of NPs involved may be semantically defined and (ii) the DR can be explained as a case of semantic (pragmatic) deviance. It is common practice to recommend model-theoretic semantics as the appropriate cool. In fact, model-theoretic semantics, especially GQT, has suggested several defini tions which are successful in picking out the intended distributional classes quire accurately. It is certainly significant that, for instance, the line between weak and strong NPs can be drawn by a simple model- theoretic definition. But does it contribute anything towards an explanation of the DR? Can GQT really help to explain the DR as the basic intuition (ii) suggests? In the next subsection let me assess several approaches that use the standard algebraic framework (Barwise & Cooper 1 98 1 ) in order to explain the DR for existential there be sentences.
H.einhard Blumer
37
Cooper ( 1 9X 1) is taken as the basic semantic tool to explain the DR for existen tial there be -sentences. According to this picture, each determiner D, let it be definite, universal, or indefinite, is interpreted as a function from ordered pairs of subsets of the interpretation domain E to truth values: (7) [ D](¢ )( l/J ) - {T, FJ , where ¢ � E, 2.2. I Barwise & Cooper
ljJ � E.
(I 98 I )
Barwise & Cooper themselves propose an analysis which is temptingly simple. The authors adopt the �bare NP' analysis of there -insertion structures and interpret sentences like ( 5 ) as indicated in (R).
THING
D
STUD n HUNGRY
D(STUD n HUNGRY) D(STUD n HUNGRY) (THING) Barwise & Cooper ( 1 9X 1 ) give the following definition of what they understand as the explication of Milsark's terms 'strong' and 'weak': (9) A determiner D is positive strong1 1c (or negative strongBu respectively) if for every model M - (E, [ ]) and every ¢ � E, i f [ D](¢ ) is defined then [ D](¢ )(¢ ) - T (or F, respectively). If D is not (positive or negative) strong1 1< , then D is weakB< :·
When the determiner D of the postcopular NP is strong110 then the translation D(STUD n HUNGRY)(THING) is logically true or logically false.3 Barwise & Cooper conclude from this fact that strong11c determiners are excluded from there - insertion sentences because they give rise to tautologies or contradictions which are not informative. Although Barwise & Cooper's filter condition ( 1 o) The determiners are naturally excluded from existential there- sentences as j ust the (positive or negative) strong ones is quite correct for the most common determiners! there explanation for this condition is questionable. Note that by their account sentences such as ( 1 1 a) and ( 1 1 b) are semantically equivalent and should be accepted or excluded for the same semantic- pragmatic reasons. ( 1 1 ) (a) * There is the student hungry (b) 111e hungry student exists
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(X) [s There is Iare [N P Det [N · student(s) hungry ]]]
3�
Dynamic Generalized Quantifiers and Exisrcntial Sentences in Narural Languages
The problem is that although ( I t a) is ill formed, ( 1 1 b) is not. The analysis under discussion leaves exactly this unexplained. Why should a logically trivial sentence like ( 1 1 b) be grammatical bur the corresponding there-semence ( I t a) be ungrammatical? Barwise & Cooper ( 1 98 1 ) do not provide any answer. Therefore, they fail to give a true explanation for the DR in there -insertion contexts.5 2.2.2 Keenan
(1 987)
Keenan ( 1 98 7) is another author who has used GQT for analysing there insertion sentences. In contrast to Barwise & Cooper ( 1 98 1 ) Keenan ( 1 987) assumes that the there be -part is semantically empty. Furthermore, he rejects the 'bare NP' analysis and maintains that the postcopular phrase is a small clause of the form (scNP XP]. With these premises a sentence like (s) is translated as indicated in ( 1 2). Next, remember Keenan's ( 1 987= 29 1 ) definition of existential ,
0
D STUD
HUNGRY
D(STUD) D(STUD)(HUNGRY) determiners: existential determiners are either basic determiners D satisfying the existence condition ( 1 3 ) ( 1 3 ) For each model M (E, [ ]) and each ¢ � E, 1/J � E: [ D]¢)(1/J) [ D] (¢ n 1/J)(E) -
=
or are built of such basic determiners by a specified number of operations, such as Boolean combination, composition with an adjective phrase, or the exception operator (e.g. no . . . butjolm ). Keenan's account explains that only in the case of existential determiners arc sentences like ( 1 4) and ( 1 s) semantically equivalent.
( 1 4) (a) There is/are Det Student(s) hungry (b) D(STUD)(HUNGRY)
( I s ) (a) Dec hungry student(s) exist(s)
(b) D(STUD n HUNGRY)(THING)
Therefore, sentences like ( 1 4) get an existential interpretation only when the determiner of the postcopular NP is existential.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 2) [s There is/are [sc [NP Det srudent(s)] [x P hungrylll
Reinhard Blumer
39
It is important to note that Keenan's analysis does not give an explanation for the fact that the sentence ( 1 6a) is unacceptable, while ( 1 6b) is fine. ( 1 6) (a) *There is each stu dmt l111ngry (b) Each stu dent is hungry (c) EACH(STUD)(HUNGRY) (= EACH(STUD n HUNGRY)(THING)) All that can be said is that ( 1 6a) and ( 1 6b) have an interpretation in common, namely ( 1 6c), and that this interpretation is not an existential one. It seems to me that this is not saying enough. Of course, one could exclude noncxistcntial determiners from there insertion contexts by requiring that these structures acquire an existential interpretation, perhaps by stipulating a filter condition like ( 1 7):
This condition is possibly the most rigorous characterization of the DR in there -i nsertion contexts. But it remains a stipulation. If this condition is semantically based it is not clear where the feature of the interpretation of existential then· -scntcnccs expressed by this requirement comes from. Lappin ( 1 9XX) argues convincingly that it would have to be added as a noncomposi tional aspect of the meaning of the entire construction. 2.2 . 3 Milsark ( 1 974) (reconstructed within
GQT)
Milsark ( 1 974) has proposed the, by now, classical analysis that provides a semantic interpretation of standard there - insertion and accounts for its DR. Milsark adopts the 'NP-XP' analysis and argues that the there be -part in structures like (6a) represents an existential quantifier that must bind some variable. Whether such a variable is available for binding or not depends on the type of the determiner of the NP in the coda. Milsark sets up two lists of expressions, one of weak (indefinite) determiners, which function as cardinality predicates having no quantificational force, and one of strong (definite) determiners or quantifiers proper. The 'ambiguous' indefinite determiners (e.g. SOllie (some oi sm ) , many (many oi mny ), twelve (twelve £!/, twlve )) appear in both . lists. According to Milsark's analysis, only codas with weak (cardinal) deter miners (which can be considered as cardinality attributes) provide the free variable to be bound by there . Codas with strong ('=" non -cardinal) determiners do not provide such a variable free for binding by there and hence are unacceptable in such a structure. Adopting the standard algebraic picture of Barwise & Cooper ( 1 9X 1 ) , it is possible to formalize Milsark's intuitive idea that weak (cardinal) determiners function as cardinality predicates and to give a more formal treatment of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 1 7) The determiners narural in existential there -sentences arc just the existential ones.
40
Dynamic Generalized Quanrifiers and Exisrenrial Senrences in Narural Languages
DR involved in there - insertion structures. The following is a convenient approximation of what various authors (e.g. Higginbotham 1 9X7; Lappin 1 9XX) propose as definitions for cardinal or adjectival determiners: ( 1 X) D is a cardinal determiner iff for every model M 1/J � E:
-
(E, [ ]) and every � � E,
l
T if I� II l/JI E K [ D](rp )( 1/J ) = F otherwise where I I designates cardinality and K is a set of cardinal numbers determined by D independently of rp and lj.l .
( l<.J) [TWELVE](� )(
1/J
)
=
ifi,P �I l F otherwise T
II
=
12
Evidently, the cardinal core in this case is the set K { 1 2). Using Barwisc & Cooper's ( 1 9X 1 ) definitions it is a simple task to show that the intuitively indefinite (or weak) determiners like SOME, NO, TWELVE are cardinal determiners, while the intuitively definite (or strong) determiners like EVERY, ALL, MOST, THE, THE TWELVE, BOTH are noncardinal ones.6 It is possible to associate with each cardinal determiner D a certain predicate expression, which I will call .n(D), for convenience. The interpretation domain of these predicate expressions is the domain of plural objects (or individual sums) in the sense of Link ( 1 9X 3 ). Technically, let me introduce this domain as *E a complete Boolean algebra generated by E, where E is the set of atoms in *E. Now let K be the core of the cardinal determiner D. The interpretation of .n (D) is as follows: =
=
)
( lo [ .n(D)] (a )
=
{ T if card(_a) F otherwise
E
K
where a E *E and card( a ) = the number of atoms that generate a .7 With this definition at hand the predicate function of thl' cardinal determiner TWELVE, by way of example, can be expressed as
( 1 ) [ .n(TWELVE)](a ) 2
=
if card \a) { FTotherwise
=
12
The philosophical chestnut (22a) will then be construed as (22b), where A represents the individual sum of the apostles (this example is taken from Higginbotham 1 9X7).
( )() 22
a
Thc apostles arc twclvc . (b) l.n(TWELVE))(A)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us call K the (cardinal) core o f the cardinal determiner. A simple example is
Reinhard Blumer
41
A predicate nominal with a cardinal determiner as i n (2 3a) gets irs prcdicativc interpretation as shown in (2 3b): The associated n (D) acts as a modifying adjective and modifies the head noun. (2 3) (a) They are [ NI' twelve students ). (b) A x. [[ n(TWEL VE)](x) 1\ *STUD(x)] Going back to existential sentences, Milsark's analysis in our example ( 5 ) may be reconstructed within the present framework as shown in (2-J.). Evidently the (24) [s There is/are [ NP Det student(s )J [ x P hungry JJ ). P.3x P(x)
n(D) *STUD
HUNGRY
n(D) n *STUD n HUNGRY 3x
[n(D) n *STUD
n
HUNGRY](x)
derived semantic structure is interpretable only for cardinal determiners-and this would explain Milsark's thesis.
(25 ) The determiners natural in existential there -sentences are just the cardinal ones.
Appreciating the attractiveness of Milsark's analysis, however, there are at least two problems in his approach that have to be noted. The fi r st problem is that Milsark's analysis does nor give the right interpretation in some cases. For instance, consider the sentence (26a), which would get the interpretation (26b).8 This interpretation sharply contradicts the intended one, which is something like (26c). (26) (a) There is no student hun,�ry . (b) 3x [CARD(x) � o 1\ *STUD(x) 1\ HUNGRY(x)] (c) -3x[CARD(x) - 1 1\ "STUD(x) 1\ HUNGRY(x)J The second problem concerns role expressions like those in (27) (the example is taken from Doron I yXX): (27) (a) jolm is vice-president C?fthe club . (b) John is the vice-president ofthe club . (Note that rhe only interpretation of (27a) conforms to the prcdicativc reading of (27b)). Milsark's interpretative rule would predict that role expressions are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
n(D) n *STUD
42
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
fine in existential there- constructions as well. The following example falsifies this prediction: (2X) (::t) * There is via·-prcsidcnt ofthe club . (b) ?There is the vice-president vJthe club. (list reading only)
2.2.4 Higginbotham
( 1 987) and Lappin ( I 9 8 8)
(29)
1
T if i¢1 E � [ ABS(D)](¢) = F otherwise where ¢ � £9
Now we are ready to derive the semantic representation of examples like (5 ) as it is indicated in (30).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Higginbotham's ( 1 9X7) and Lappin's ( 1 9XX) account-outlined here in a slightly modified way-avoids the problems just mentioned. Both authors adopt the 'bare N P' view of there- insertion structure advocated by Williams ( I I)X.J.) and fol low Williams also in regarding the there he-part as semantically empty. Given these premises, it is necessary to treat the postcopular NP as a predicational structure which bears the burden of the proposition expressed by the sentence. Note that this view indirectly extends the claim pur forward by Wiliams ( 1 9X3) and Partee ( 1 9X6) that NPs semantically represent different logical types: the basic NP types e 'referential', (e, t) 'predicative', and ((e, t), t) 'quanrificational'. We have ro extend this list now by introducing the type t for postcopular NPs in there -insertion structures. It has been pointed out that a determiner D basically denotes a function from ordered pairs of sets to truth values, where the first element of this p::tir is realized by a restricting set term. If the postcopular NP in there he -sentences is a predicational structure (type t), then it is necessary to associate an unrestricted, or absolute, determiner function with the postcopular determiner, since its usual restriction, the N ', no longer serves that function. Higginbotham ( 1 9X7) and Lappin ( I I)XX) argue that it is possible to associate with each cardinal determiner such an absolute determiner function which I will call ABS(D). ABS(D) applies to a distributive predicate and forms a predication with it. Let K be the core of a cardinal determiner D. The interpretation of ABS(D) is as follows:
Reinhard Blumer
43
(3o) [s There is/are [N P Det [N · student (s) hungry]]] 0
STUD
n
HUNGRY
[ABS(D)] (STUD
n
HUNGRY)
[ABS(D)] (STUD
n
HUNGRY)
ABS(D)
The resulting semantic form can be interpreted only for cardinal determiners, since only in this case the function [ ABS(D)] has been defined. This leads to the following DR:
This is exactly Milsark's thesis (2 5 ), bur Higginbotham ( 19X7) and Lappin ( 1 9 X X) have derived it in another way. It can be argued that Higginbotham's and Lappin's account avoids the main problems of Milsark's. However, it raises others. Let me note two. The first concerns the justification of type t as a basic NP type. It seems to me that there is no general j ustification and that this claim is completely ad hoc . The second problem arises if we consider languages other than E nglish. Huang ( 19X7) provides a very interesting description of existential sentences in Chinese. He convincingly argues that existential sentences in Chinese cannot be properly analysed along the lines of the 'bare NP' theory. Nevertheless, Chinese existentials exhibit a DR. How, then, can this restriction be explained? Huang's analysis implies that a u niversal explanation for the DR in existential sentences, if it involves the structural properties of the coda, must appeal to them only indirectly. It seems to me that Higginbotham's ( 1 9X 7) and Lappin's ( 1 9 X X ) analysis is based on the 'bare NP' analysis too directly. Their account, for instance, cannot be extended to a 'NP-XP' coda (Huang's preference for Chinese). Let us summarize this section. We have assessed several model- theoretic approaches that use the standard algebraic framework and claim to explain the Dl� for existential there be -sentences. However, we have not found anything like a coherent and truly explanatory theory that accounts for this construction. Instead, the proposals that claim to explain the DR are rather artificial for the most part and suffer from various shortcomings. Furthermore, the proposals do not extend to other constructions exhibiting a DR (cf. Blumer 1 990 for a discussion of this point in connection with predicate nominals and donkey sentences). Therefore, a new beginning seems to be absolutely necessary.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(3 1 ) The determiners natural in existential there -sentences are just the cardinal ones.
44
Dynamic Generalized Quantifiers and Existenrial Sentences in Natural Languages 3
D Y N A M I C G Q T (G r o e n e n d ij k
&
S to k h o f 1 9 9 0 )
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I n the previous section we have concluded that the standard algebraic framework is not sufficient to explain the DR for existential there-be sentences. In my opinion, one general reason for this insufficiency is that GQT does not account for the dynamics of this construction. Clearly, existential there be sentences have dynamic effects that should be accounted for. Intuitively, they arc presentational. That means, they introduce (or deny) the presence of a certain object (or group of objects) in the domain of discourse or in a context relevant subdomain. Since the original GQT docs not work any better with anaphoric phenomena and dynamic effects than more traditional accounts, we must make generalized quantifiers more dynamic and powerful. DRT is a semantic theory that champions a dynamic conception of meaning. However, DRT cannot be considered simply as a way to make generalized quantifiers more dynamic and powerful; rather, it seems to follow a more radical line of reconstructing NP semantics. On this road DRT abandons central principles of NP semantics, such as the uniform treatment of all NPs, the compositionality principle of semantic interpretation, and the conserva tivity postulate for determiners (Keenan & Stavi 1 986). Appreciating the efforts of DRT to overcome the limitations and shortcomings of Montague Grammar and GQT, I think that DRT does it partly too radically and abandons central principles too fast. On the other hand, there are some problems that suggest that DRT (in one of its original forms) is not rigorous enough in accounting for dynamic effects. A case in point is the proportion problem of donkey -sentences (e.g. Kadmon 1 987; Heim 1 990; Kamp & Reyle to appear; Chierchia 1 9XX); another is modal subordination (e.g. Roberts 1 9X9) and other accommodation phenomena (cf Groenendijk & Stokhof 1 990; Chierchia 1 9XX). The semantics of existential there be -sentences and the explanation of the DR involved in this construction may provide another argument for being cautious with regard to the use of DRT. Though it seems possible to modify DRT gradually and to overcome its problems step by step, I think a radically new reconstruction ofNP semantics is unavoidable. Such a reconstruction appears methodologically preferable if it is of theoretical importance (integration of different theoretical frameworks, use of general restricting principles) and if it has some empirical import. The dynamic conception provided by Groenendij k & Stokhof ( 1 990) strikes me as particularly promising in this regard. These authors have developed a strictly compositional account that incorporates the insights of DRT into Montague's intensional logic. In this way, they have provided a sound unification of DRT and GQT. Chierchia ( 1 98 8) has used this theory in order to account for the proportion problem of donkey -sentences and for the treatment of when -
Reinhard Blumer
�S
clauses. Groenendij k & Stokhof ( 1 990) themselves give some hints as to how to deal with modal subordination in a fully compositional way. In the next section I hope to demonstrate how this approach throws new light on the treatment of existential there be -sentences. 3 . 1 An introduction to dynamic intensional logic
(32)
() a
A
(b) 3x (c) 3x (d) Ed;
stu dent [STUD(x) [STUD(x) [tSTUD(d;)
1\ 1\
walks in . WALK - IN(x)] WALK - IN(x) tW ALK- IN(d;)]
1\ 1\
He is hunxry . HUNGRY(x) HUNGRY(x)] tHUNGRY(d;)
Groenendijk & Stokhof overcome the obvious conflict between the binding relation in logical syntax and that in the representation of discourse by separating the meaning of a sentence from its truth-conditional import. These authors would translate the discourse (pa) into a formula of dynamic logic as shown in (32d). Here Ed; designates the so-called dynamic existential quantifier (over the discourse marker d;). The up-arrow (t) converts static expressions denoting truth values into dynamic expressions denoting information change potentials. The semicolon designates dynamic conjunction. Before the relevant definitions are given we should have a look at the structure of the dynamic representation (pd). Notice that the syntactic scope of the dynamic quantifier Ed; is restricted to material introduced by the first sentence of the discourse. However, by converting the dynamic formula (pd) back to its truth conditional counterpart, the expression (pc) results (as will be shown immediately). In the latter expression the existential quantifier has scope over
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The basic idea of dynamic semantics is that the meaning of a sentence resides in its information change potential. According to Groenendij k & Stokhof's ( 1 990) theory, the information change potential of a sentence tells us-informally speaking-which propositions are true after its contents have been processed. This view contrasts with the traditional one, which identifies the meaning of a sentence with its truth conditions. An important aspect of Groenendij k & Stokhof's theory is that this static notion of meaning is retained in their formalism, but as a secondary one, which can be derived from the primary notion of information change potential. In order to illustrate the basic idea and some crucial notations, let us now look at the sample discourse (pa). The usual predicate-logical translation of the sentences of this discourse is indicated in (pb). The problem with this account is that the anaphoric relation between the indefinite NP and the pronoun cannot be expressed by the binding relation of standard logic. A more satisfying translation cannot be obtained in a straightforward (compositional) way.
46 Dynamic Generalized Quantifiers and Existential
Sentences in Natural Languages
does not hold with full generality in OIL, but is subject to restrictions similar to those which it is subject to in Montague's IL. However, the possibility of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the whole discourse, and this provides a 'dynamic binding' for the pronoun in the second sentence of our discourse. Now let us give a fairly precise characterization of the basic techniques of dynamic semantics (as developed by Groenendijk & Stokhof 1 990). In trying to incorporate the insights of DRT within Montague's ( 1 974) intensional logic, Groenendij k & Stokhof develop the language OIL (Dynamic Intensional Logic), which is a variant of the system of intensional type theory IL (used in Montague's PTQ). We will give only a sketch of OIL, following the simplifications used by Dekker ( 1 990). The types of OIL arc the same as those ofiL. The syntax ofDIL is extended by adding a set of discourse markers d 1 , d�, . . . to the expressions of type e. Quantified expressions are constructed with the help of these discourse markers: 3di ¢ , Vdi¢ (¢ of type t). A model structure ofDIL is a triple M - (E, S, F). instead of worlds, a OIL model structure contains a set of states S. These states can be identified with assignments of individuals to discourse markers. As usual, E designates the domain of individuals. The interpretation function F for the constants of the language is state-insensitive. Further, the set of (ordinary) variable assignments is denoted by G. The expression g[x/o], where x is a variable and o an object of the corresponding type, should denote an assignment that is j ust like g with the possible exception that it maps x to o. The related notation s [d/o] can be applied to states (where d is a discourse marker and o E EJ. The semantics of OIL is stated by defining the notion [ E ]M.,.g, the interpretation of the expression c with respect to the model M, state s, and variable assignment g. The differences in semantic interpretation only relate to discourse markers: ( 3 3) (a) [ d]M.,.g - s(d), for every discourse marker d (b) [3d ¢ ]'M·'·� - 1 iff there is an o E E,. such that [¢ ]M.,Idi..l.g - 1 [Vd ¢ ]M.>.g - 1 iff for all o E E,. it holds that [¢]M.,I d!ol.g - I (c) [ ' P ]M.>.g - the function h E Ebs such that h(s) - [ P ] M.>.g for all s E S, where b is the type of /3 [ · P] M.>.g - [ P ]M.,.g(s). The interpretation of a discourse marker is the value it has in the state of evaluation. Quantification over the values of discourse markers is mediated by quantification over these states (not over ordinary variable assignments). The intension operator abstracts over states; the extension operator applies an intension to the state of evaluation. The principle of A. -conversion (for extensional type theory) (34) (a) [A.v .p](o) equals [o!v]P if all free variables in 0 are free for v in p.
Reinhard B lumer
47
abstracting over states and to close the argument term intensionally has the following pleasant consequences for ), -conversion in OIL:
c5) equals 1 · b /v J P if all free variables in c5 are free for v in p. Note that no condition on discourse markers in c5 need be obtained. Although their interpretation is state-dependent, rhe fact rhar ·c5 is intensionally closed
(34) (b) j ).v .IJ](
·
and hence stare-independent is sufficient to validate the equivalence (34b). The following example demonstrates how this equivalence can be used for providing a dynamic binding relation-i.e. a binding relation where a discourse quantifier binds a discou rse marker outside the syntactic scope of the quantifier:
We see that rhe ), -term ), p. 3d jQ (d) 1\ ·p] in fact is dynamic in rhe indicated sense: indirecrly (via ), -conversion) rhe embedded existential quantifier binds a discourse marker rhar occurs in the argument expression of the A -term, i.e. a discourse marker outside the syntactic scope of the existential quantifier. We arl' now ready for a systematic introduction of the notational conventions that will facilitate rhe representation of the information change potential of sentences or pieces of discourse. We will use ·a as short for the type (s, a) and t as short for ( ·r, r)-rhe type of rhe information change potential. In the simplest case, rhe information change potential of a sentence is introduced by means of rhe dynamic operator 1: (36) 1� - ), p. j� 1\ 'p], where � is an expression of type r and p is a variable for propositions (type ·r). The dynamic operator 1 maps rhe static semantic value of an expression of type t into irs dynamic counterpart. Formally, 1 can be viewed as a type shifting operator which maps expression of type r into expressions denoting sets of propositions (1: r I-+ t). The converse of the dynamic operator 1 is rhe static operator !. Given rhe information change potential of a sentence, rhe static operator retrieves irs truth-conditional import, simply by applying ir to a tautology. (37) l1 = 1/J ( 'TRUE), where 1/1 is an expression of type t, and TRUE is an arbitrary tautology. A straightforward computation shows thar ! followed by 1 cancels our: (3H) !l� is equivalenr to � As pointed our by Groenendijk & Stokhof( 1 990), rhe operator ! may be looked upon as a kind of close operator, one which closes a piece of text and 'freezes'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(3 S ) j). p. 3d jQ(d) 1\ 'p] ]( "P(d)) equals 3 d jQ(d) 1\ P(d)J (Note that the subsequent occurrence of · and · cancels out.)
4!1
Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
any dynamic effects it may have had. Once closed off, a piece of text remains that way, even if raised again to the higher type by means oft. A further notion that we have already used in representing the sample discourse (22a) is ';'-conjunction. (39) [1/1 1 , 1/1 2] - A p. t/> 1 ( 1 2(p)), where 1/1 1 and t/>2 are expressions of type t, and p is a variable of type ·t. •
(4o) [t¢ 1 ; l¢ 2] is equivalent to t(¢ 1
1\
¢2), where ¢ 1 and ¢2 arc of type t.
Next, we have dynamic existential quantification over discourse markers (i.e. existential quantification over discourse markers occurring in a dynamic formula): (4 1 ) Ed 1/1 - A p. 3d 1/J (p), where 1/J is an expression of type t, p is a variable of type 1:, and d is a discourse marker. Given the associativity of function composition it immediately follows that the dynamic existential quantifier has the following property (cf. Grocnendijk & Stokhof's ( 1 990: 2 s) Fact 7): (42) EdtQ(d); tP(d) is equivalent to Ed[tQ(d); tP(d)J, where P and Q denote ordinary predicates of type (e, t). Furthermore, in order to determine the truth-conditions of dynamic formulas by means of !, we can replace !Ed tf> by 3d ! t/J (this fact immediately follows from the definition (4 1 ) ). Now it is a straightforward task to prove the follow ing reformulation of the equivalence given in ( J s): (43) l[EdtQ(d); tP(d)J is equivalent to 3x[Q(x) 1\ P(x)J This equivalence provides a clear demonstration of the dynamic binding relation established by means of the dynamic quantifier Ed. In the formu la on the left- hand side of this equivalence the syntactic scope of the dynamic quantifier is restricted to the first cot�tmct. However, taking into consideration its truth-conditional import given on the right-hand side, the existential quantifier has scope over both conjuncts. This is the place to look once again at the sample discourse (J 2a) and its dynamic translation (32d). Using the equivalences expressed in (40) and (43), it is obvious that the application of the state operator ! relates the dynamic
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The expression [ tf> 1 ; tfJ 2] represents the dynamic conjunction, or sequence, of two sentences. Semantially, dynamic conjunction is simply a form offunction com position, one that makes provision for Montague's intension operator. A straightforward computation shows the close correspondence between dynamic cot�unction and ordinary conjunction for sentences without truly dynamic effects:
Reinhard Blumer
49
(44) (a) *Every man 1 walks in the park. He1 whistles . (b) Every player1 chooses a pawn. He1 puts it on square one . The pronoun in the second sentence of (44a) cannot be interpreted as an anaphor which is bound by the universal term in the first sentence. However, in the sequence (44b) the anaphoric relation of the indicated kind is possible. DRT only accounts for the first case; in order to deal with the second case, an additional mechanism would be necessary (e.g. transformational operations on logical forms, c£ Roberts 1 989). O IL can handle both cases in a principled way, using two kinds of negation (c£ Groenendij k & Scokhof 1 990). Though Groenendij k & Stokhof's way of handling the examples (44a, b) may appear insufficient for various reasons, it convincingly demonstrates that OIL is more than just the sum of orthodox Montague grammar and DRT (for more discussion see Dekker 1 990). 3 .2 Translation ofa sample discourse into DIL
Let me outline now how a small fragment of English can be assigned a compositional dynamic interpretation through translation into OIL. The fragment chosen is the one crucially involved in the sample discourse (32). First, let us assign to each lexical expression of the fragment its translation into DIL. In what follows u is a variable for individual concepts (type 'e); P and Q are variables for dynamic properties (type 1 'e, t)); STUD, WALK-IN, and HUNGRY are constants of type (e, c); the d; are discourse markers. The list presented in (4S) now will be taken as our sample lexicon. (4 S) (a) (b) (c) (d)
[N student] [ vrwalk in] [ Ap hungry] [ NPhe;]
4 4 4 4
) u. tSTUD( 'u) ,
A. u.
tWALK-IN( " u) ) u. tHUNGRY( 'u) A. P. 'P( 'd;) (d; xivcn w.r. to previous context) ,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
translation (pd) to the predicate-logical expression (pc) which represents the truth-conditional import of the sample discourse. We have presented an idea of some basic mechanisms of OIL needed in the subsequent discussion. Two general facts should now be added in order to get a more complete impression of OIL. The first one concerns the possibility to express DRT within OIL. Groenendij k & Srokhof( 1 990, 1 99 1 ) have provided a meaning-preserving translation of DRT to OIL. In this way they have shown that OIL is at least capable of providing an account for the same phenomena that DRT can handle. The second fact concerns the possibility to deal within OIL with a wider range of phenomena than DRT is able to cope with. As a case in point, consider the sequences (44a) and (44b):
so
Dynamic Generalized Quamifiers and Existential Senrences in Narural Languages
.... ..t Q..t P. Ed;I .Q(d;); "P(d;)] (d; new w.r. to previous context) '- ..t P..t u. "P(u)
(46 ) (a) FA (Functional Application):
for y of type ( 'a, b) and o of type a: FA(y, o) � y( 'o) (b) GDC (Generalized Dynamic Conjunction): (i) for y, o of type t: GDC(y, o) !y; o j (ii) for y , o of type (a, b): GDC(y, o ) h". GDC(y(x,), o (x,)) �
�
(Intensional) Functional Application is a standard from Montague Grammar. Note that the occurrence of the "-operator makes the corresponding
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The translations (a)-(c) arc like those from standard Montague Grammar. The only difference is the occurrence of the operator 1 which makes sure that a formula such as STUD( · u) which is of type t is raised in the appropriate way to A p.[STUD( ·u) 1\ ·p] to represent the information change potential. The translation (d) takes pronouns basically as discourse markers and raises them to functions from dynamic properties into information change potentials (type ( '( 'e, t), t)). For obvious reasons, expressions of this type are called dynamic generalized quantifiers. Intuitively, a pronoun refers to an entity that is given with regard to the conversation domain. This (dynamic) condition is put in brackets as a condition on the discourse marker d;. Regarding the indefinite determiner-translated by (e)-the main difference from standard Montague Grammar is that the dynamic conjunction occurs instead of the static one and that we use the dynamic existential quantifier. An important point is that we need to choose a particular discourse marker in the translation, which is indicated by the index that occurs in the determiner itself The (dynamic) condition that an indefinite NP refers to an entity that is newly added to the conversation domain is put in brackets as a condition on the discourse marker d; . 1 0 The translation (f) takes as the meaning of the auxiliary be the operator 'apply predicate' (neglecting tense and modals). In this regard we follow Williams ( 1 9X 3) and Partee ( 1 9X6) , but depart from Montague ( 1 974). As will be suggested in the next section, we will treat Montague's PTQ translation of English be as a type-shifting functor that we apply to the generalized quantifier meaning of a NP whenever we find the NP in a predicative position. Next, in order to conrruct the translation of a compound expression from its immediate parts, we need semantic operations that are associated (in familiar ways) with syntactic rules. There are two semantic operations in the present fragment, called FA and GDC:
Reinhard Blumer
51
expressiOn '<5 intensionally closed. This is of importance for applying A conversion (34b). Generalized Dynamic Conjunction is a straightforward extension of Hendriks' ( 1 989) Generalized Conjunction. It extends ordinary dynamic conjunction to the conjunction of dynamic predicates, dynamic properties, and so on. By way of illustration let us follow the translation of a; student , which is given in (47):
[Det
[N student JJ J. u. fSTUD( 'u)
The resulting OIL expression is a dynamic generalized quantifier (rype( '( 'e, t), t)). Now it can be shown that the translation of our sample discourse (32a) really results in (32d). Examples (48a, b, c) present the obvious derivation. (+X) (a)
[N student II J. QJ, P. Ed;[ .Q( 'd;) ; ·p( 'd;)]
). u. fSTUD( 'u)
FA - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
[w walks in II J. u. fWALK- IN( "u)
J. P. Ed; [fSTUD(d;) ; ·p(d;)]
FA - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -- - - - - - - - - - - - - - - - - - - - - - - - - - -
Ed;[fSTUD(d;) ; f WALK - IN(d;)]
[ AP hungry II ), P. ·p( 'd;)
), P), u.J. P(u)
J. u. fHUNGRY( "u)
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - FA
), u. fHUNGRY( "u)
-- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - FA
fHUNGRY(d;)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
J. QJ. P. Ed rQ( 'd;) ; ·p( 'd;)] ( 'J. u. fSTUD( ' u)) ---- - -------- - --- ------------------------ - ------- - reduction J. P. Ed;[ fSTUD(d;) ; ·p( 'd;)] via (34b)
52
Dynamic Generalized Quanrifiers and Existential Senrences in Natural Languages
[s a; student walks in
[ s he; is hungry ]
tHUNGRY(d; Ed; [fSTUD(d;) ; fW ALK- IN(d;)] - --------- - - - GDC ------------------------------fHUNGRY(d;)] [Ed;[fSTUD(d;) ; tW ALK - IN(d;)] ; + E X I S T E N T I A L THE R E BE S E N T E N C E S : A D Y N A M I C A PPROA C H
4· I Predicative interpretation ofnoun phrases
For the following it is necessary to make the category- to- type relationship more flexible than in Montague's approach. In order not to introduce too many novelties at the same time, let me first develop the basic idea concerning the kind of flexibility desired for a sublanguage which is essentially static. In agreement with Williams ( 1 9X3) and Partee ( 1 9X6) I propose three logical types for the syntactic category NP: e 'referential', (e, t) 'predicative, and ((e, t), t) 'quantificational'. Partee ( 1 9X6) argues that the last, the type of generalized quantifiers, is the most general: all NPs have meanings of this type, while only some have meaning of type e and/or type (e, t). Furthermore, she proposes general principles for predicting from the generalized quantifier interpretation of a given NP what possible e-type and/or (e, t)- type interpretations it will have. Though the type-splitting of an NP is a semantic notion, not a syntactic one, different syntactic environments may restrict the assignment of the appropriate logical- semantic type. In accordance with Milsark ( 1 97 4) I will assume that both predicate nominals and the NPs in codas of existential there be -sentences arc restricted semantically to the type (e, t) of predicates. In order to construct these predicates from generalized quantifiers, I will adopt Partee's type-shifting operator BE. This operator is essentially Montague's PTQ translation ofEnglish be . Here is the slightly simplified version of Partee ( 1 9X6): (49) BE(9')
=
J. y. 9' (J. x. x
=
y)
The operator BE applies to a generalized quantifier 9' , finds all the singletons therein, and collects their elements into a set. For the generalized quantifier
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In this section we return to the treatment of existential there be - sentences. In order to accou.nt for the specific character and for the dynamics of these sentences, their translation into OIL is a desirable task. The application of dynamic generalized quantifiers helps us to cope with this task.
Reinhard Blumer
S3
meaning of a student , A P. 3 x[STUD(x) 1\ P(x)], this gives the predicate of being a student, as shown in (so): (so) BE(.Jc P. 3x[STUD(x) 1\ P(x)]) = A y. 3 x[STUD(x) 1\ x = y]
( s I ) BE(.9') = A u. .'!!_( .A v. ·v = ·u), where u, v are variables of type
e,
and
.9
is an expression of type ( "( ·e, t), t).
Apart from the use of states instead of possible worlds, this variant is identical with Montague's origin translation of English be . Next, let us apply this operator to the state-sensitive generalized quantifier meaning of a; student , A �- 3d;[STUD(d;) 1\ "E( "d;)], using the (static) discourse quantifier 3d; � i s a variable of type "( ·e, t)): (52) BE(A �- 3d;[STUD(d;) 1\ 1( .d;)]) = }. u. 3d; [STUD(d;) 1\ d; = ·u] In order to get the translation of hen is a; student , we have to apply the property given by (52) to the (intensionally closed) discourse marker "d11(n i' i). We are allowed to perform A. -conversion, and it results in 3d;[STUD(d;) 1\ d; = d 11], or, equivalently STUD(d11).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us apply this predicate to a variable X11 (different from x). This variable can be taken as the e- type translation of the pronoun lze11 • When we perform the corresponding A -conversion, it results in 3x[STUD(x) 1\ x = X11] which reduces to STUD(x.J Evidently we have found the adequate translation of !zen is a student in this way. Now let us apply the predicate given by (so) to the variable x: [A y. 3x[STUD(x) 1\ x = y]](x). In this case, we are not allowed to convert because the free occurrence of the variable x in the argument would become bound in the resulting expression. Unfortunately, this is precisely what we want in the case of existential sentences: 3x[STUD(x) 1\ x = x] (or, after reduction, 3xSTUD(x)) would be nice as a translation of the existential sentence there is a student . Note that the variable x would be taken as the e-type translation of existential there in this case. The close correspondence between the semantics of ordinary pronouns and existential there will prove to be a striking feature of this kind of analysis and it will yield a number of pleasant consequences (discussed in Section 4-4)· To maintain the pronoun- like analysis of existential there , an approach needs to be provided to A -conversion bringing free variables in the scope of quanti fiers binding them. A particularly simple solution, inspired by Montague's treatment of the same problem in connection with intensionality, is to use Groenendijk & Stokhof's ( 1 990) extension and intension operators sensitive to states (= assignments of values to discourse markers). This solution runs as follows. First, we adopt the state-sensitive variant ( 5 1 ) of Partee's (extensional) type shifter BE.
54
Dynamic Generalized Quanrifiers and Existenrial Sentences in Nacural Languages
Now let us apply the property given by ( 52) to the discourse marker "di (where di is the same discourse marker as that introduced by the state-sensitive translation of a; student ): [A. u. 3di[STUD(di) 1\ di = 'u]]( "di)· Since the argument term "di is i ntensionally dosed, we are again allowed to convert, and it results in 3di [STUD(di) 1\ di dJ The latter can be demonstrated to be equivalent to 3x STUD(x)-the fairly correct truth-condition of the existential there; is a; student . This derivation demonstrates how the binding problem in connection with A. -conversion mentioned above can be overcome as a technical problem if state-sensitive tools are used. Importantly, the dose correspondence between the semantic translation of pronouns and existential there is maintained in the analysis. Up to now we have used an essentially static approach of analysing existen tial there- sentences. This approach accounts nicely for the existential feature in the corresponding truth conditions, but it does not account for the presenta tional character of these sentences. Hence it is insufficient for explaining that there-insertion sentences license discourse anaphora. In the next subsection I want to demonstrate how this problem can be solved if the dynamic element of Groenendij k & Stokhof's theory is incorporated.'' =
It is a simple task to adjust the type-shifter BE to the demands ofGroencndijk & Stokhof's ( 1 990) dynamic theory. Starting from Montague's origin expression, we have only to introduce the up-arrow at the right place: (5 3 ) BE�) = A. u..9' ( "A. v. t 'v 'u), where u, v are variables of type ·e, and .9' is a dynamic generalized quantifier (type ( "( e, t), t)). =
•
Let us apply this dynamically adjusted type shifter to the dynamic generalized quantifier meaning of a; student (cf (47) ):
(54) BE(A. P.Edi[tSTUD(di) ; 'P( "di)] = A. u.Edi[tSTUD(dJ ; tdi
=
'u]
In order to get the translation of the sentence (s sa), we have to apply the expres sion (54) to the discourse marker "dn (n -:1 i) as done in (ssb). ( 5 s) (a) (b) (c) (d)
He. is a; student .
[A. u.EdJtSTUD(di) ; tdi 'u]]( "d11), with n -:1 i Edi[tSTUD(di) ;tdi = dn] 3x[STUD(x) 1\ X - dn] (= STUD(dn)) =
After performing A. -conversion, the expression (s sc) results; its truth conditional import is given by (s sd). The latter outcome is identical with the result found earlier by using a purely static semantics. Now let us assume again that existential there refers to a discourse marker
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4.2 The type-shifting operator BE in dynamic semantics
Reinhard Blumer
SS
and that this discourse marker is exactly the same as the one which is newly introduced by the indefinite NP in the coda of our existential sentence. Consequently, the translation of the existential sentence (s6a} should be (s6c). (s6) (a) (b) (c) (d)
There; is a; student . [A u.Ed;[tSTUD(d;) ; td; - 'u]]( 'd;) Ed; [tSTUD(d;; td; - d;] 3x[STUD(x) 1\ x - x (= 3x STUD (x))
( s 7) (a) There; is God . He; is omnipotent . a;
(b) [Ed;tGOD(d;) ; tOMNIPOTENT(d;)] 4·3 Dynamic GQT and existential there be-sentences
The aim of the present subsection is to present the proposed dynamic approach to English existentials in a more systematic form . To this end let us extend the fragment given in Section 3 .2 as follows. First, we add an entry for existential there to the sample lexicon (4 5): (4 5) (g) [NP there;] 4 A P. 1>( "d;) (d; new w.r. to previous context) The dynamic condition in brackets distinguishes there from ordinary pronouns and gives it a touch of indefiniteness. In order to enforce that the discourse marker introduced by there and that the postverbal (indefinite) NP are identical, a second condition is required. This condition can be stated as follows: (sH} In a dynamic formular of type t each discourse marker has to be dynamically bound. Since the discourse marker introduced by there has to be new w.r. to previous context (see the condition in brackets in (4 sg} ), the only possibility to get it bound is by the existential operator contained in the posrverbal indefinite NP. Consequently, the discourse marker introduced by there and that by the postverbal indefinite NP have ro be identical. Notice thar the condition (sX) is motivated for reasons that are completely
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
After performing A -conversion, the expression (s6c) results. It has the truth conditional import given in (s6d). Note that the latter result is the same as that found in the previous subsection by using state-sensitive, but purely static means. The advantage of the present dynamic approach is that it accounts for the fact that there -insertion sentences license the binding of discourse pronouns. For example, the discourse (57a) would be represented as (5 7b), with a dynamic binding relation established by means of the dynamic discourse quantifier Ed;.
56 Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
independent of interpreting existential there -sentences. It ensures that each ordinary pronoun and each definite NP (old discourse markers!) is dynamically bound by an operator that has introduced the corresponding antecedent and brings about the anaphoric relationship between the anaphoric expression and its antecedent. Obviously, this corresponds to a standard case of local coherence. In this sense, the condition ( sX ) can be grasped as a standard condition of local coherence (of course, it is not the only condition of local coherence that ought to be formulated). Now let us look at the 'novelty condition' in brackets in (45g) again. Together with the (essentially pragmatic) condition ( sX ) it excludes definite NPs (old discourse markers!) from contexts of there -insertion. Consequently, it can be said that this condition constitutes the dynamic trigger of the definiteness restriction in contexts of there -insertion. Up to now we have not considered the question of how the lexical clement there should be analysed syntactically . According to current GB theory, there is usually analysed as an expletive element (i.e. not conteniful expression ) that occupies the non - theta subject position, and its distribution is tied to the Extended Projection Principle (see Chomsky 1 9X6, for example). Recently, this view has been criticized by Hornstein ( 1 99 1 ). In his analysis Hornstein assumes that 'pleonastic' expressions such as English there , German es , and Icelandic bad syntactically arc LF operators occupying the non- argument-adjoined position at LF: [I P t; INFL [vP there; [V P . . . ]]]. As discussed by Hornstein ( 1 99 1 ) in detail, this assumption yields a unified approach to the restricted distribution of English and Germanic 'expletives'. Furthermore, it yields a principled approach to a variety of island effects that existential constructions in English and Icelandic have. Regarding the semantic side of the coin, Hornstein ( 1 99 1 ) is far less explicit than regarding the syntactic one. Corresponding to the notion that there is an operator at LF, Hornstein assumes that this operator indeed has semantic import. This assumption contrasts with the standard GB view that treats there as a pleonastic element with no semantic content. What then is there 's inter pretation? Hornstein suggests the following: ' There is an operator with very limited quantificational force: it is a minimal lambda operator. By "minimal" I mean to suggest that in contrast to other types of overt lambda abstractors, e.g. WH relative pronouns, it does not itself have a range' (Hornstein 1 99 1 : 3 3 ). Unfortunately, Hornstein ( 1 99 1 ) docs not try to characterize his minimal lambda operator in more precise terms. Given current standards of model - theoretic semantics, I think, Hornstein's suggestion concerning the semantic concent of there is rather unsatisfactory. However, I feel that the main part of Hornstein's semantic intuitions may be made explicit if the operator given in (45g) is taken as the semantic content of existential there . Semantically, this operator applies to the (type-shifted) content
Reinhard Blumer 57
of the coda of the existential there be -sentence. Now condition ( 5X ) applies and ensures that the discourse marker introduced by this operator and that by the postverbal (indefinite) NP are identical. 12 In order to interpret our simple fragment let us next consider the semantic operations that are used to construct the translations of compound expressions following the syntactic rules. These operations are as above and extensive use is made of GDC for modifying head nouns and for forming 'NP-XP' codas. Last bur not least, the type-shifting operator BE(.9' ) is introduced as an operator that freely applies (to dynamic generalized quantifier expressions .9' ). Let us now throw a closer look at existential there be -sentences as exemplified in ( 59) : a;
Let us first assume a 'bare NP' analysis of this sentence, as shown in the first line of (6o), 1 3 and let us take the bare NP (- the coda) as a dynamic generalized quantifier. Next, let us apply the dynamic type shifter BE to it. The resulting (6o)
l s [ NP there;]
rs
[N ·
student
hungry
]]]
A P. Edi [fSTUD(di) ; fHUNGRY(di) ; "P( "di)] ------------------------ - - - --- ------- BE A u. Edi [fSTUD(di) ; fHUNGRY(di) ; fdi - "u] -- ---------- - ----------- - ----------------------------- FA
Truth-conditional import: 3x[STUD(x) 1\ HUNGRY(x)] dynamic property (third line of (6o)) functions as the argument of the transla tion of there . After A -conversion, the expression in the fourth line results. This expression licenses the binding of discourse pronouns. Its truth -conditional import is as expected for the existential sentence under discussion. Now let us see how the present theory can be applied to the 'NP-XP' analysis. The first line of (6 1 ) shows the 'NP-XP' analysis for our sample sentence ( 59). In the second line the semantic translation of its main constituents are given. The translation of [ NP a; student ] as a dynamic generalized quantifier has been type- shifted via application o£BE. The fourth line results from applying GDC to the elements of the coda, and the next one represents the result of applying the translation of there to the rest of the sentence. It is noteworthy that this result is essentially the same as that in case of the 'bare NP' analysis shown in (6o). This outcome demonstrates that the present approach, while involving the structural
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( 59) There; is student hungry .
58
Dynamic Generalized Quantifiers and Exisrenrial Sentences in Narural Languages
(6 t ) [ s [ NP there ]
IS
[xP hungry] ]
student l P. Ed;[fSTUD(d;) ; 'P(d;)]
BE ----------------------
). u . fHUNGRY( "u)
;. u. Ed;[fSTUD(d;) ; fd; u] GDC ------------------ -- ------------------- - ). u. [Ed;[fSTUD(d;) ; fd; 'u] ; fHUNGRY( 'u)J -·
=
[Ed;[fSTUD(d;) ; fd;
=
d;] ; fHUNGRY(d;)]
properties of the coda, applies to them only indirectly. In stnvmg for a linguistically universal characterization of (in)definiteness this result may be i nteresting. 4·4 Discussion and conclusions
Finally, let me list some advantages of the proposed theory and let me refer to some open questions. Furthermore, a word of caution should be given concerning rash generalizations about this study. Let me start with the advantages. (i) The conceptual advantage: beside the most elementary type e we only need two basic NP- types-that of dynamic generalized quantifiers {type ( '( 'e, t), t)) and that of dynamic predicates (type ( 'e, t). Furthermore, the shift between generalized quantifiers and predicates is provided in a principled way: by the type-shifting functor BE. This functor is indeed 'natural', both intuit ively and by various criteria as demonstrated by Partee ( 1 9X6) for the static case. It remains left to the reader to contrast this view with that of Higginbotham ( 1 9X7) and Lappin ( 1 9XX) who need an extra NP- type t and an arbitrary shifting device for approaching the postcopular NP in existential there be -sentences. (ii) The present approach does not depend on the structural properties of the coda too directly. It holds both for 'bare NP' codas (English) and for 'NP XP' codas (Chinese). This fact is important for a universal characterization of the DR. Correspondingly, the present framework overcomes a· further disad vantage of Higginbotham's ( 1 9X7) and Lappin's ( 1 9X X) theory, which only holds for 'bare NP' codas and cannot be extended to 'NP-XP' codas in a straightfor ward way. (iii) The present analysis preserves the advantages of Milsark's ( 1 974) theory
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Truth-conditional import: 3x[STUD(x) 1\ HUNGRY(x)J
Reinhard Blumer
59
without being burdened with irs disadvantages. As discussed in Section 2..3, a first disadvantage concerns negative NPs like no student . According to the present approach, the application of the type shifter BE gives the correct result: (62.) (a) no; student 4 A P -Ed;(lSTUD(d;) ; "P(d;)] (- closed negation in the sense of Groenendij k & Stokhof 1 990). (b) BE(no; student ') � A U. -[Ed;[tSTUD(d;) ; ld; � "u]] � A u. -[BE(a; student ')](u) (' designates the OIL- translation of the corresponding expression of English) (c) there; is no; student hungry 4 -Ed;[tSTUD(d;) ; tHUNGRY(d;)]
(2.X) (a) * There is vice-president ofthe club . gets an existential interpretation and would count as acceptable. This is certainly incorrect. In the present analysis (2Xa) comes out as unacceptable because the predicate VICE - PRESIDENT-OF-THE-CLUB does not contain an (implicit) existential quantifier that can bind the discourse marker arising from there (only this would give the whole sentence an existential inter pretation). Instead, the new discourse marker introduced by there remains unbound (independent of the discourse surrounding it). Such structures do not give an interpretable truth-conditional i mport in (partial) dynamic semantics and therefore should count as unacceptable. (iv) The present approach can be extended to plural NPs in a straightforward way if groups or plural objects are treated as entities in the sense of Link ( 1 9X 3). A simple example must be sufficient to demonstrate this point: 1 4 (6 3) (a) twelve students 'J. P. Ed;(lCARD(d;) - 1 2. ; l*STUD(d;) ; " P{ 'd;)] (b) BE( twelve; students') = A u. Ed;(lCARD(d;) - 1 2. ; l*STUD(d) ; ld; = "u] (c) there; are twelve; students hungry 4 Ed;(lCARD(d;) - 1 2 ; l*STUD(d;) ; tHUNGRY(d;)] (v) According to the present theory, DRs derive from two different sources. The first source has to do with the set- theoretic peculiarities of the type shifting functor BE. Although BE is a total function, it sometimes leads to a pragmatically degenerate result. For example, be every student corresponds to a degenerate case in which be the only student would be more appropriate. Therefore, pragmatic reasons can be taken to exclude
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
A further problem concerns the treatment of role expressions. According to Milsark's ( 1 974) theory, rhe sentence (2Xa)
6o Dynamic Generalized Quamifiers and Existential Sentences in Natural Languages
(64) (a) *john is every student . (degenerate world) (b) * There is every student . (tautology) The second source of the DR has to do with the violation of the dynamic condition imposed on the discourse markers involved. It has already been outlined that the special condition imposed on there excludes definite NPs like the president and my friend from there - insertion contexts. On the other hand, these NPs are fine as predicate nominals (<;ontrast (6sa) with (6 sb): the b-cases have list readings only). (6::> ) (a) JoIm IS. ·
the president.
IS
.
myjr1end.
The semantic peculiarities of these NPs as predicate nominals can be accommodated within the present theory as well (cf Blutner 1 990). Con sequenrly, the present theory seems to be a promising approach to explaining the slighrly different DRs for predicate nominals and existential there sentences. Summing up these findings and observations, I conclude that Groenendij k & Stokhof's dynamic variant of GQT provides a uniform semantic framework for explaining some peculiarities of existential sentences. In this way it has been found that dynamic Montague grammar has applications other than those which promoted it and advantages other than those Groenendij k & Stokhof claim for it. However, there arc many open questions and also some potential objections to the proposed theory. Remember first that the present theory restricts itself to the treatment of existential there be -sentences. It would have been nice if the solution was straightforwardly generalizable to other there -insertion contexts and explained-for example-contrasts like those in (66) and (67) (cf Jenkins 1 975)· (66) (a) (b) (67) (a) (b)
There ran a man into the room. *There ran a man around the track. There began a riot. *There ended a riot.
Following Davidson { 1 967), Higginbotham ( 1 9X7) and others, it appears that we have to assume that sentences are all interpreted as making use of an event variable. At present it is rather unclear to me whether the Davidsonian idea can be combined with the present theory in an interesting way that would lead to the desirable generalizations and consequences. Further research is necessary in order to clarify this point
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I myfinen. d. (b) * ] '/1ere . { the president
Reinhard Blumer 6 1
Received: 1 6-2-9 1 Revised versinn received: 3 1-7-92
REINHARD 13LUTNER Ml'G A rlwii.��Zruppcji"ir Slmklurcllc ( ;,,111/lll•llik Pre11zla ua Prome11adc 1 4 9- 1 5 1
0- 1 1 o v lkrli11
Germa11y
Acknowledgements
I rhank M. Bierwisch, F. Hamm, H. Kamp, C. Piiion, I. Zimmermann, and an anonymous reviewer of rhe journal cU" Semalllics for valuable criricism and suggesrions. Thanks also ro M.Schenrke for correcring my English. The responsibiliry for all indequacies and misrakes remains wirh rhe aurhor.
N O TES Orher environmenrs exhibiring a defi niteness resrricrion are, for example, predicare nomiuals, dvnkey-senrences, parririve NP-consrrucrions, clefts, and pseudo-clefts (cf. Higginbotham 1 9�7; Blumer 1 990). 2 For a criticism of Sa fir ( 1 9X2. 1 9X7) see Heim ( 1 9X7) and Higginbotham ( 1 9X7). This is a simple consequence of the universal of conservariviry, D(A)(B) =
+
D(A)(A n B), which srares rhar D(A)(THING) = D(A)(A). For posirive (negarive) srrong 11.- dererminers rhe proposirion D(A)(A) is logically rrue (false); rherefore rhe same holds for D(A)(THING). For a critical discussion of the descriprive correctness of the filrer condition ( 1 o) see Keenan ( 1 9X7) and Lappin ( 1 9XX). For a more derailed discussion of rhis
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Another pomt where further research would be desirable concerns the integration of a more marure syntax with the proposed semantic approach. This point has to do with the question of whether we may expect model- theoretic semantics to be able to contribute to solutions for what seem to be language particular, hence syntactic, facts. Hornstein's ( 1 99 1 ) comparative srudy of English and Icelandic may be important here. I think the crucial parts of Horn stein's syntactic analysis can be interpreted within the proposed dynamic framework. This will be a challenge for the future. Thirdly, it should be mentioned that we have adopted from Groenendijk & Srokhof ( 1 990) the idea that dynamic logic is necessary for the interpretation of pronouns both in intrasentential contexts and in donkey -sentences. However, their theory runs into problems if it comes to evaluating (i) the very close correspondence between narural language determiners and adverbs of quantifi cation, (ii) the proportion problem and the phenomenon of asymmetric quanti fication in the context of donkey -sentences, (iii) the integration of dynamic binding with the £-type strategy. Various proposals have been made to improve the original theory (c(, for example, Chierchia 1 9XX; Dekker 1 990): It is not so clear whether these improvements are compatible with the present approach to existential sentences in narural language.
62 Dynamic Generalized Quantifiers and Existential Sentences in Natural Languages
1
o
1
1
12
�
(ii)
if card�a) - o [ n(NO)] ( a) - { TF otherwise
9 Note the close correspondence between the absolute determiner expression ABS(D) and the predicate expression .n(O): (i) [ABS(D)] (P} = [.n(D)](axP(x)); valid for each cardinal determiner D a n d each distributive predicate P (a 1s Link's ( 1 9X 3) sum operator). Higginbotham's ( 1 9H7} and Lappins ( 1 9H 3) accounts deviate from the recon struction presenred here. Both authors identify sets with plural objects. In this way, they mix up the group (individual) perspective and the set perspective, which leads to some inconsistencies. The presenr
13
reconstruction avoids this mixture and relates set perspective and group perspec tive by means of relationships like (i). Note that for a sound treatment of the terms 'given' and 'new' the use of partial information states (- partial assignment functions) instead of total ones is required (see Heim 1 982: 302 f£). Some authors, e.g. Zeevat ( 1 990), seem to accept the state-sensitive account as a means ro deal with rhe binding problem of A -conversion, but they are rather sceptical with regard to the dynamic aspect of Groenendij k & Srokhof's theory. I cannot see whether the static mechanism proposed by Zeevar ( I 990) t�)r dealing with pronoun binding has some particular advanrage in the conrext of there -insertion constructions. Notice thar in an earlier paper (Blutner 1990) another condition had been used instead of the condition (;X). This condi tion was stated in terms of 'novelry': d; (the discourse marker inrroduced by there ) has to be given with respect to the coda. This formulation is problematic for various reasons. first of all, it gives the wrong impression that there really func tions like a pronoun-a pronoun of which the anrecedent must be found in the coda. Taking there as a kind of pronoun, how ever, leads to a violation of the Reinhart rules (the direction of pronominalization is odd, see Reinhart I 976). Furthermore, the earlier formulation seems to violate, ar least in spirit, the novelry condition thar one would like to impose on the posrverbal indefinite NP. In chis proposal, the indefmite NP inrroduces a discourse marker that has already been introduced by rhe 'pronoun' there . I am indebted to an anonymous reviewer of the Journal 4 Semantics for bringing these inconsisten cies of the earlier formulation to my attenrion. This is a simplified representation that leaves out crucial parts of Hornstein's analysis. However, it seems ro be suffici ent for the present purpose.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
problem and related ones, see Keenan ( 1 9X7). Recently, Johnsen ( 1 9X7) has proposed an improved version of the Barwise & Cooper ( 1 9X 1 } account. The main argumenr presenred above against this account, however, seems likewise available from Johnsen's analysis. 6 Determiners like MANY, A FEW, MORE . . . THAN, which are weak intuitively, are classified as noncardinal by condition ( 1 H). Lappin ( 1 <)XX} extends the cardinaliry condition, so that these determiners are classified as cardinal ones with a relational core. 7 It is tempting to represent the general 'ambiguiry' exhibited by indefinites (c£ Milsark 1 974) byassociatingwirh eachcar dinal determiner both the original deter miner function D (rype ((e, t), ((e, t), t}}) and the 'derived' predicate function .n(D) (rype (e, t}). For a criticism of this solution and for alternative proposals see Lohner (1 984)· X Use the common interpretation of the determiner no as given in (i): ) - { T if l¢ n 1/JI o (i) [ N O](Y',t, )(''' 'Y F otherwise The transition to the associated cardi naliry predicate leads to:
Reinhard Blumer 63 1 -l It has been argued that indefinite deter miners like some , few , many are 'ambig uous' between an unstressed variant and a stressed form, the former being a vague cardinality predicate and the latter a proper quantifier. In accordance with Lohner ( 1 9X-l) I will not take this ambiv-
alence as a case of true lexical ambiguity; rather, I will consider it as a kind of unspecified interpretation. As such it should be treated within the present dynamic fonnalism; note that the transla tion presented by (62) accounts for the unstressed variam only.
R E FE R E N C ES
c
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Barwise, J. & Cooper, R. ( 1 9X 1 ), 'Generalized and indefinite NPs', Ph.D. thesis, UMass, quantifiers and natural language', Linxuis Amherst. tics and Philosophy , 4, 1 59-2 1 9. Heim, I. ( 1 9X7), 'Where does the definiteness Blumer, R. ( 1 990), 'Type shifts in dynamic restriction apply? Evidence from the logic and the definiteness effect', in L. Kal definiteness of variables', in R. Reuland & man & L. P6los (eds), PapersFom the Second A. ter Meulen (eds), 2 1 -42. Symposium on Loxic and Lanxuaxe , Aka Heim, I. ( 1 990), 'E-type pronouns and donkey demiai Kiad6, Budapest, 49-65. anaphora', Linxuistics and Philosophy. I J , Chierchia, G. ( 1 9X!\), 'Dynamic generalized 1 37-77quantifiers and donkey anaphora', in M. Hendriks, H. ( 1 989). Flexible Montaxue Gram Krifka (ed.), Genericity in Natural Lanxuaxe. mar , ITLI/Department of Philosophy, Proceedinxs
64
Dynamic Generalized Quantifiers and Existential Sentences i n Natural Languages
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Keenan, E. L. ( 1 911 7), 'A semantic definition Reinhart, T. ( 1 976), 'The syntactic domain of of"indefinite NPn ', in R. Reuland & A. ter anaphora', Ph.D. thesis, MIT, Cambridge, Meulen (eds), 2.X6-3 1 7. Mass. Keenan, E. L. & Stavi, J. ( 1 9116), 'A semantic Reuland, E.J. & ter Meulen, A. G. B. (eds) { 1 98 7), The Representation of(in)dejiniteness , characterization of natural language deter MIT Press, Cambridge, Mass. miners', Linguistics and Philosophy , 9, 2.5 3J 26. Roberts, C. ( 1 9X9), 'Modal subordination and Lappin, S. ( 1 9II X), 'The semantics of"manyn as pronominal anaphora in discourse', Lin a weak determiner', Linguistics , 26, 977-9X. guistics and Philosophy , 12, oX 3-7 2. 1 . Link, G. ( 1 9X 3 ), 'The logical analysis of Safir, K.J. ( 1 9X2.), 'Symactic chains and the definiteness effect', Ph.D. thesis, MIT, plurals and mass terms: a lattice-theoretic Cambridge, Mass. approach', in R. Bauerle, C. Schwarze & A. von Stechow (eds), Meanin)i, Use and Inter Safir, K. J. ( 1 9117), 'What explains the defi niteness effect?' in R. Reuland & A. ter pretation 4 Lanj!uaxe , de Gruyter, Berlin, Meulen (eds), 7 1 -97. 3 0 2- 2. 3. Lobner, S. ( 1 9X4), 'Indefiniteness, couming, Seuren, P. A. M. ( 1 9X6), Discourse Semanlics , Blackwell, Oxford. and the background/foreground distinc tion', paper presemed at the Fifth Gron Stowell, T. ( 1 911 1 ), 'Origins of phrase struc ture', Ph.D. thesis, MIT, Cambridge, Mass. i ngen Round Table, Groningen,June 1 9X.J.. Milsark, G. L. ( 1 97+). 'Existemial semences in Williams, E. ( 1 97 5 ), 'Small clauses in Et}glish', in J. D. Kimball (ed.), Synlax and English', Ph.D. thesis, MIT, Cambridge, Semantics , vol. 4, Academic Press, New Mass. York, l...J-9-73 · Montague, R. ( 1 97-J.), 'The proper treatmeiH of quantification in English', in It Thoma Williams, E. ( 1 911 3 ), 'Semantic vs. symactic categories', Linj!uislics and Philosophy , 6, son (ed.), Formal Philosophy: Selected Papers 4 42.3-46. Richard Montaxue , Yale University Press, Williams, E. ( 1 9X4), 'There-insertion', Lil1New Haven, l..J-7-70. )iuistic lnquiry , 1 5, 1 3 1 -5 3 · Partee, B. ( 1 986), 'Noun-phrase imer pretation and type shifting principles', in J. Zeevat, H. ( 1 990), 'Static semamics' in J. van Bemhem (ed.). Parlial and Dynamic Seman Groenendij k, D. de Jongh & M. Srokhof tics , vol. I , DYANA Deliverable R2.. 1 A. (eds), Studies in Discourse Representation Edinburgh, 1 3 S-.J.O. Theory and the Theory c/Generalizcd (Juanri jlcrs (� Grass x), Foris, Dordrecht, I I S-+3 ·
}
© N.l.S. Foundation (1 993)
to: 6S-I I I
A Localist Model for Event Semantics
J AN VAN V O ORST
Universite du Quebec a Montreal
Abstract
1 I NTRODUCTI ON This study presents a comprehensive localist conceptualization o f event semantics using the notion of spatial closeness. The analysis uses Hjelmslev's localist primitives of inclusion and cohesion. Conceptually, the model propo sed contrasts strongly with temporally based approaches to the study of verb classification and aspect that make use of notions such as STATE, (SVB)EVENT, PROCESS, CHANGE OF STATE, etc. The localist approach to aspect and the study of verb classification integrates aspectual semantics and semantics that can be considered to be of a more thematic nature. The study concentrates on transitive verbs and their argument structures. All primitives in the model are motivated by phenomena pertaining to the grammar. These phenomena are centred around the focus and interpretation of adverbials belonging to a number of different semantic classes. The primitives uncovered this way are unified into an overarching theory of event semantics. The model is centred around the interaction patterns of entities partaking in events but in a way very different from the patterns subsumed under a thematic relations approach. The entity denoted by the subject either establishes a relation of inclusion with that denoted by the direct object or it does not. If no
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This study presents a localist model illuminating the semamics of transitive constmctions. The model groups transitive argumem structures imo a number of classes similar to those found in many aspectual studies which focus primarily on temporal aspects of evems. The building blocks of the model are based on linguistic evidence in the form of tests involving adverbials of differellt types. A crucial role is played by manner adverbials. These adverbials show that the entity given by the subject has comrol over the emities given by the direct object of accomplishments and activities, bur not over those denoted by the prepositional object and by the direct object of achievemems and states. The notions proposed are inclusion and cohesion which specify the spatial relationship between two emities in terms of the event brought into existence by one of them. The localist model is a comprehensive conceptualization of events expressed through transitive constructions and it explains why individuation has such an importallt role to play in aspect.
66
A Localist Model for Event Semantics
2 T HE S E M A N T I C S O F M A N N E R A D V E RB I A L S
This section is devoted to the discussion of the so-called VP- or manner reading of adverbials such as carefully , clumsily , meticulously , precisely , sloppily , attentively, disorderly, slowly , etc. Adverbials belonging to this class specify the way the entity denoted by the subject interacts with the event it has initiated and with the entities figuring in it. They can have several readings depending on their syntactic position. Jackendoff ( I 972) gives examples illustrating two of the possible interpretations. ( I a) and ( I b) can have the interpretation in (2a); ( I b) and ( I c) can have the interpretation in (2b).
( I ) a. John cleverly/clumsily dropped his cup of coffee. b. Cleverly/clumsily (,)John dropped his cup of coffee c. John dropped his cup of coffee cleverly/clumsily
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
relation of inclusion exists, the sentence expresses a state. I f a relation of inclusion is established, cohesion between two entities may or may not result. I f there i s no cohesion, we have an achievement. I f there is cohesion, i t can either be complete or partial. If it is complete, we have an accomplishment: if it is partial, we have an activity. The prepositional object does not figure in this relation of inclusion and that of the resulting cohesion. This is the reason why it can never be the centre of the change of state that characterizes accomplish ment or perfective aspect. The analysis is markedly nontemporal and phenomena of durativity and punctuality and a variety of tense-related pheno mena, such as the interpretation of the progressive form and the present perfect, have to find their explanation elsewhere. However, it is not incompatible with certain temporally defined aspectual notions such as imperfectivity, perfectiv ity, and stativity which can all be read off the model. The difference between a purely linguistic and a more interpretative, situationally oriented, semantic analysis is very important for the present discussion. Throughout this study, using a variety oflinguistic tests, I will show how various proposals describing the semantics of argument structures contain information for which the linguistic relevance cannot be proved. As such my analysis brings some order in a domain of semantic research that has very few restrictions on the number and type of primitives as well as on the information it can use for its analyses. My proposals remain rather descriptive and I do not attempt to formalize them. This choice is based on the assumption that the determination of the linguistic contents of semantic primitives is an important enterprise and needs to precede any choice of formal representation. I f this is not done, one runs the risk of including in a formal representation primitives that are not clearly linguistic.
J a n van Voorst
67
(2) a. It was clever/clumsy ofjohn to drop his cup of coffee b. The manner in which John dropped his cup of coffee was clever/clumsy
2 . 1 Activities and accomplishments
Verbs like load , paint , spray and some others alternate between two argument structures. I will refer to this phenomenon as the spray/paint -alte rnation.2 I will demonstrate in Section 3 that these verbs can be activities or accomplishments in aspectual classifications. The spray Ipaint -alternation is very interesting because, at first sight, the two argument structures look similar in terms of the semantics focused on. However, this similarity is not reflected in the linguistic data. The truck in the load -sentences in (3) is affected in both instances as the lzay drops on top of it. The hay is also affected in both cases as it is manipulated. (3) a. He loaded the truck with the hay b. He loaded the hay on the truck Both the truck and the hay are necessarily affected in the situations which the sentences in (3) describe. However, the use of manner adverbials shows that there is a semantic difference between both sentences with respect to affected ness. (4) a. He carefully loaded the truck with the boxes b. He carefully loaded the boxes on the truck Under the manner reading the adverbial in (4a) can specify something about the agent, the truck and the manner of loading. The agent is being careful with the truck rather than the boxes. In (4b) the focus of carefully includes the boxes and the agent. The agent is being careful with the boxes rather than the truck.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The above two readings are combined into one semantic formula in Ernst ( I 98 s) . Ernst states that the rough interpretation is that 'the Agent can bej udged ADJECTIVE because of a ', where a can be S for the sentence reading or VP for the manner reading. Ernst ( I 985) is an in-depth analysis of this ambiguity and of the adverbials that do or do not allow it. The author's semantic interpretation rule for the manner readings is that 'there is something about the referent of that verb-the action, state, process, etc.-that can be described by that adverbial' (p. 9 I ). The question we are interested in is not addressed by this rule. This question is more specific and concerns the possible interpretations of the adverbial under the VP- or manner reading. 1 In Section 2. 1 I will analyse constructions with verbs like load and spray . This is followed by a discussion of constructions with verbs like miss and notice in Section 2.2. Finally in Section 2.3 I will show how my findings contrast with the outcomes of a number of earlier studies.
68
A Localist Model for Event Semantics
When the adverbial occurs following the direct object, a similar interpretacion arises. Again the agent is being careful with the truck in (sa) and the boxes in (sb).
( s ) a. He loaded the truck carefully with the boxes b. He loaded the boxes carefully on the truck
Sentences like those in (6) further illustrate these readings. The modifying phrase in (6a) applies to the prepositional phrase of the matrix sentence. The sentence is odd because it is hard to find a link between the fragility of the boxes' contents and the fact that the enrity denoted by the subject is being careful with the truck. (6b) is perfectly natural.
Putting the adverbial in a posicion following the direct object does not change . these interpretations. (7)
??He loaded the truck carefully with the boxes because their were fragile. b. He loaded the boxes carefully on to the truck because their were fragile. (X) a. ??He loaded the truck with the boxes carefully because their were fragile b. He loaded the boxes on to the truck carefully because their were fragile a.
contents contents contents contents
(9a) and (9b) show the same thing. Again it is odd to relate the reason for why someone is acring carefully to the prepositional phrase. This is the case in (9a). A reading that is possible, but maybe a bit far-fetched, could be that the acrion was being performed carefully because the broken springs could directly affect the boxes. Note that under this reading, the reason for why someone was acring carefu lly is related to the enrity denoted by the direct object. Purring the adverbial in other posirions creates the same effect. I leave it to the r�ader to check this. (9)
??He carefully loaded the boxes on to the truck because its springs were gone b. He carefully loaded the truck with the boxes because its springs were gone a.
The sentences with the adverbial heavy in ( 1 o) illustrate the same phenomenon. The modifying phrase in ( 1 oa) applies to the preposirional phrase of the matrix
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(6) a. ??He carefully loaded the truck with the boxes because their contents were fragile b. He carefully loaded the boxes on to the truck because their contents were fragile
J an van Voorst
69
sentence. The sentence is odd because it is hard to find a link between the weight of the boxes and the fact that the entity denoted by the subject is being careful with the truck. A possible, but somewhat far-fetched, interpretation is that the weight of the boxes jeopardizes the functioning of the truck because it has a bad suspension. But now the modifying phrase applies to the direct object of the matrix sentence indirectly. The agent is being careful with the truck because he does not want to damage it with the heavy boxes. The sentence does not mean that he is being careful with the boxes just because they are heavy. ( 1 0b) is perfectly natural. (I o) a. ??He carefully loaded the truck with the boxes because they were heavy b. He carefully loaded the boxes onto the truck because they were heavy
( I I ) a. He loaded the truck with the boxes carefully b. He loaded the boxes on the truck carefully The presence of a quantifier phrase makes possible yet another reading. Sentence ( 1 2) could mean that the agent took care to get the complete set of boxes on to the truck. He did not want to miss one and leave it behind somewhere. This is the S- or sentence reading represented by Jackendoff's description in ( I za). The sentence could also mean that he was being careful each time he loaded a box, which is the manner reading as set our in ( I zb). ( I 2) He carefully loaded all the boxes on the truck a. It was careful ofjohn to load all the boxes b. The manner in which John loaded the boxes was careful Interestingly, when the quantifier occurs in the prepositional object, one of the interpretations does not arise. ( I J) implies that John took care to get all the boxes on to the truck (i.e. that John was concerned about this). The inter pretation of the adverbial, that John was acting carefully with each box independendy, is absent.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The data in (6)-( 1 0) and their interpretations indicate that adverbials of the careful -type do not apply to the prepositional object. Giving a reason for the careful loading of an entity that applies to the entity denoted by the prepositional object is odd. There seems to be a natural tendency to establish a relation between this reason and the entity denoted by the direct object. The acceptability of this relation depends on various nonlinguistic factors. The weight of boxes ( I oa) is easier to relate to the well-being of a truck than, for instance, the price of boxes. When the adverbial follows the prepositional object phrase as in ( I I ) the interpretation arises again that the process of loading the truck ( I I a) or the boxes ( 1 1 b) is done in a careful way. An interpretation that the agent is being careful with the entity denoted by the prepositional object does not occur.
70
A Localise Model for Event Semantics
{I 3 ) John carefully loaded the truck with all the boxes
( 1 4) a. b. (I s ) a. b. ( 1 6) a. b.
He quickly loaded the train with the containers He quickly loaded the containers on to the train He loaded the train with the containers quickly He loaded the containers on to the train quickly He loaded the train quickly with the containers He loaded the containers quickly on to the train
Sentences like those in ( I 7) demonstrate another reading for this adverbial. It could also specify the time it took the agent before he got involved in the event. This is again the S -reading, which is irrelevant for the present discussion. (1 7) a. Quickly he loaded the truck with the boxes b. Quickly he loaded the boxes on the truck There are many adverbials that can apply to a prepositional phrase expressing an object partaking in an action. This is so with the adverbial accidentally in ( I 8). ( I 8a) and (I 8b) may mean that the truck was wrongly chosen, but also that the coal was the wrong material loaded. Accidentally can also apply to the whole event, leading to the interpretation that the event should not have taken place and was performed accidentally. However, this reading is hard to get in this particular case, as loading is a time-consuming and involved task, which makes it hard to believe that someone performs it without wanting to perform it in the first place.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What stands out in the above data is that careful can never include the preposi tional object within its focus. Even when the adverbial directly follows this obj ect, the desired interpretation does not arise. The data with the quantifier all also confirm this finding but in a different way. Careful can apply to the prepositional phrase in these cases but then it has a different meaning. It does not apply to the very action but to the concerns of the entity denoted by the subject about an aspect of his action. Here it is possible to speak of an S-reading. The adverbial quickly brings out the intended differences between direct obj ect and prepositional objects even more strongly. The agent in all of the sentences in ( 1 4)-{1 6) is acting quickly with the entity expressed by the direct object. The a-sentences tell us something about the speed with which the train fills up and the b -sentences tell us about the speed with which the array of containers is eliminated. It is not expressed if the entities denoted by the prepositional object are affected in a quick way or not. Actually, in ( I 4a), the agent could move these entities relatively slowly and still be quick in loading the train. In ( qb) it could be that the train is being loaded slowly which is the case when the train is excessively long.
J a n van Voorst 7 1
( I 8 ) a . He accidentally loaded the truck with coal
b. He accidentally loaded coal on to the truck
Accidentally has similar interpretative possibilities in other positions although one of the interpretations may be preferred depending on its syntactic position. The sentences in ( I 9)-(2 I ) illustrate this.
( I 9) a. He loaded the truck accidentally with coal
He loaded the coal accidentally on to the truck He loaded the truck with the boxes accidentally He loaded the boxes on to the truck accidentally Accidentally he loaded the truck with the boxes Accidentally he loaded the boxes on to the truck
Wisely behaves like accidentally in this respect. The data are given in (22) and (2 3). Again the adverbial can apply to the prepositional object as well as to the direct object. It can mean that it was wise either to choose the car or the particular colour. (22) a. b. (23) a. b.
They wisely painted the car with a neon colour They wisely painted a neon colour on the car Wisely they painted the car with a neon colour Wisely they painted a neon colour on the car
With verbs like load , paint , etc., manner adverbials can have focus over the entity denoted by the subject and by the direct object but not over that denoted by the direct object. The sentence does not imply that it can be careful with the entity denoted by the prepositional object. Adverbials like accidentally are different and they can apply to the latter entity. This contrast becomes espe cially strong when adverbials of the respective types follow the direct object. With manner adverbials the intended reading is not brought out even in this position. What this means for a manner adverbial like careful is that the entity denoted by the subject can only be careful with the entity denoted by the direct object. Note that this is a wholly linguistic-semantic issue as it is very easy to imagine situations in which the entity denoted by the subject has some power over what is happening to the entity denoted by the prepositional object. The linguistic phenomena surrounding manner adverbials show that this is not part of the semantics expressed by the sentence. 2.2 States and achievements
When taking into consideration more data it becomes clear that in a large number of transitive constructions, manner adverbials are unacceptable altogether. The aspectual verb classes I looked at so far are those concerning
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
b. (2o) a. b. (2 I ) a. b.
72 A Localist Model for Event Semantics
activities and accomplishments. The spray/paint -data are activities in case their direct object is nonindividuated (i.e. a mass term or an indefinite plural) and accomplishments in case it is individuated. The test with the adverbial almost (Dowry 1 979, Morgan 1 969) shows that the sentences in (24) are activities and those in (2 s) accomplishments.3 Activities and accomplishments allow the reading that the event was not performed at all. This I will call the begin-read ing. The so-called end-reading is only possible with accomplishments. This is the interpretation that the event was not carried on to its completion although it was on its way to being completed.
Manner adverbials cannot have a VP-reading with verbs from Vendler's ( 1 967) other two aspectual classes, such as the stative verbs like miss and have in the sentences in (26) and (27). (26) a. b. c. (27) a. b. b.
These lapdogs were missing all their fur *These lapdogs were missing all their fur clumsily *These lapdogs were clumsily missing all their fur He had that glass figurine in his hands *He carefully had that glass figurine in his hands *He had that glass figurine carefully in his hands
Achievements, like the ones in (2X)-(3o), do not allow the VP-reading of manner adverbials either. This is the reading that the process expressed by the verb is performed or is done in the way specified by the adverbial. Note that, under the present approach, the achievements are the only class that will be analysed radically differently from Vendler's original proposals. The traditional relation between achievement and punctuality will be severed. Perception verbs like hear as well as a large number of psychological verbs will be analysed as achievements. The motivation behind this move will be given in Section 4 · (2X) a. b. c. (29) a. b. c.
He noticed me in the corridor *He meticulously noticed me in the corridor *He noticed me meticulously in the corridor He has won the lottery *He has carefullywon the lottery *He has won the lottery carefully
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(24) a. I had almost stuffed pillows with the feathers but I realized in rime that I needed them for something else. b. I had almost stuffed feathers in the pillow but he told me in time he was allergic to them. (25) a. I had almost stuffed the pillow with the feathers when Peter arrived b. I had almost stuffed the feathers in the pillow when Peter arrived
J an van Voorsr
73
(3o) a. He had heard the news about the riots b. *He had attentively heard the news about the riots c. *He had heard the news about the riots attentively Certain psychological verbs also disallow .the VP-reading of manner adverbials. Examples of the verbs that figure most prominently in today's literature, such as Belletti & Rizzi ( 1 988), are cases like fear and frighten .
A sentence like (pc) is acceptable with the S-reading that it did not take much time before the feeling of being frightened started to occur. It is very easy to confuse the S- and the VP-reading in certain circumstances and I will give a few more examples to make the reader aware of this. The adverbial in the sentences in (3 3) modifies the time period preceding the event's beginning. It does not specify how the event proceeds once it has started, which is the VP-reading. (3 3) a. b. c. d.
He quickly saw me in the poorly lie corridor He quickly won the race He quickly heard the message He quickly feared all his colleagues
(3 3a) implies chat it did not take very long before he actually saw me. (3 3b) means chat the race was won after a relatively short time period had passed by. The winning is instantaneous and there is no process of winning that can be fasc.4 (3 3c) implies chat it took litcle time before he had heard everything. It does not imply that the process of hearing itself proceeded in a fast manner. Finally (3 3d) does not express that the fearing was proceeding quickly, but that it did not take long before the whole set of colleagues were feared. Restrictions on manner adverbials vary. Under their VP-reading they can focus on the direct object with activities and accomplishments but not on the prepositional object. Achievements, states, and certain psychological verbs do not allow the VP-reading of this rype of adverbial at all. 2.3 Other proposals
The findings of the preceding sections can be contrasted with chose of ocher analyses of the spray Ipaint -alternation. Fillmore's ( 1 96X) description of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(3 1 ) a. He feared this very detailed presentation on depreciation in accounting b. *He meticulously feared this very detailed presentation on depreciation in accounting c. *He slowly feared this very detailed presentation on depreciation in accounnng (32) a. That treatment of depreciation frightened him b. *That treatment of depreciation meticulously frightened him c. That treannent of depreciation quickly frightened him.
74 A Localisr Model for Evenr Semantics
sentences in ( 34) is that the truck bears the role of 'location' and the hay the role of 'instrument'. The sentences are related to each other through a derivation. {34) a. He loaded the hay on the truck b. He loaded the truck with the hay More recent lexical studies such as Levin & Rappaport ( 1 986) give the lexical entries in (3 s ) for the verb load . Underlining indicates that the entities are realized as the direct object. (3 5)
A:agent [material, location] load B:agenc [{material), location]
(36) a. He loaded the hay on the truck b. He loaded the truck with hay What is important for the present discussion is that the holistic/partitive analysis implies that there is a systematic relation between syntax and semantics. An argument may only bear the holistic reading when realized as a direct object NP; otherwise the partitive reading applies. In Section 3 I will show that this is a direct consequence of aspect, and that a link can be made between aspect and my findings regarding manner adverbials.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'A' in (3 5) is the entry for (34a), specifying that the verb has an agent subject: a direct object denoting a certain material; and, finally, an argument indicating a location. 'B' tells us that the location is the direct object and that the argument expressing the material is optional. These analyses give the prepositional object and the direct object the same semantic status. Because of the absence of a well-defined theory of thematic roles, this rype of an analysis is unable to account for my findings that this status is very different other than by the sim ple addition of a new notion or by redefining an old one. This makes it impos sible to speak of a coherent and falsifiable semantic theory where thematic roles are concerned. Opposition to a case or thematic role description of the spray Ipaint -pairs has largely focused on the holistic/partitive issue. Case or thematic roles are said not to give an accurate description of the sentences rypical for these pairs. They are different in that (36a) has a partitive reading and (36b) a holistic one. In (36a) only part of the truck need be loaded, but in (36b) the whole truck.
J a n van Voorst
75
3 A S PE C T A N D THE R E L A T I O N B E T W E E N S U BJ E C T A N D D I RE C T O BJ E C T N P A well-known phenomenon (Mufwene 1 984b and Verkuyl 1 972, 1 989 and many others) is the relation between aspect and the individuation of the direct object as a mass noun, indefinite plural, definite noun, etc. I will explain how the findings of the preceding section and individuation are intertwined. I will conclude this section by discussing a number of other proposals on aspect and transitivity.
When the direct object is not individuated (i.e. when it is a mass noun or an indefinite plural), constructions from the spray Ipaint -pairs are activities. When it is individuated, they are accomplishments. The aspectual test with the adverbial almost is one way to show this, as we have seen in (24) and (2 s). repeated here as (37) and (3 8). All four sentences can mean that the events were not performed at all. In addition (38a) and (38b) also allow the end-reading, which is the interpretation that, although the event was engaged in, it was not completed. (37) a. I had almost stuffed pillows with the feathers but I realized in time that I needed them for something else. b. I had almost stuffed feathers in the pillow but he told me in time he was allergic to them. (38) a. I had almost stuffed the pillow with the feathers when Peter arrived b. I had almost stuffed the feathers in the pillow when Peter arrived The change of state in the direct object in transitive accomplishment construc tions identifies the end of the event. When all the pillows are stuffed in (38a) or when all the feathers are stuffed in (38b), the events are over. Prepositional phrases do not have this semantic possibility. The entity denoted by the prepositional object cannot be the centre of a change of state that identifies the end of an event.5 The fact that they are individuated does not make a difference. Not all the latex need be gone in (39a) for the event to be ended. (39b) does not imply that the event will be over when the whole wall is painted. Parts of the wall may be unpainted after this has been done. (39) a. He painted the wall with the latex b. He painted the latex on to the wall our of boredom (39a) differs from (39b) in that in (39a) the wall is the identification point of the end of the event but, in (39b) this is the latex. When the wall is painted,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3 . I Aspect and the direct object
76 A Localist Model for Event Semantics
the event in (39a) is over; when all the latex is gone in (39b), the event is over. The individuation of the direct object influences the aspect of activities and can tum them into accomplishments. The latter typically have an individuated direct object and, if they don't, they express a series of accomplishments. (4oa) is an ordinary, (4ob) is an iteration of a number of accomplishments. (4o) a. He folded the flyer b. He folded flyers
(4 1 ) The carpenter almost made a chair (42a) only means that the state of having the right answer has failed to occur; (42b) that the state of touching the wall has failed to occur. The sentences cannot mean that a final state of knowing and touching did not occur although the process leading up to these states did. ·
(42) a. They almost had the right answer b. The Chippendale armoire almost touched the ceiling Achievements behave like states in this respect. (43a) expresses that an event of reaching a certain point in space has failed to occur; (43b) that the event of seeing birthday presents has not occurred. The direct objects in these sentences are individuated. (43) a. They had almost reached the end of the subway line b. He almost saw his birthday presents Psychological verbs like fear and frighten behave the same way. Although both direct objects are individuated, (44a) implies that the fearing and (44b) that the frightening have failed to occur. (44) a. Garth almost feared these imposing body builders b. It almost frightened her to see how willing people were to go to war The individuation of the direct object influences the aspectual class that a construction belongs to when it is an activity or an accomplishment. An activity typically has a nonindividuated direct object, an accomplishment an individua-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
With a large number of other verbs the nature of the direct object does not influence the aspectual interpretation. These are verbs from the stative and achievements classes. The direct object of the psychological verbs that were introduced earlier does not influence the aspect of the construction involved either. The ambiguity that we encountered in (3R) with the adverbial almost does not surface in these constructions. In (4 1 ) the adverbial can either indicate that the event has not yet started or that it has been going on but was not carried on to its conclusion, which is the creation of a chair.
J a n van Voorst 77
ted one. The individuation of the prepositional object does not correlate with the aspect of a sentence. This means that the entity denoted by the prepositional object cannot indicate the change of state that concludes an accomplishment. The direct object of states and achievements cannot indicate such a change of state either. For this reason they do not become accomplishments, as do activi ties, when their direct object is individuated. In what follows, I will add another dimension to these findings and work towards a deeper explanation for the data. 3 .2 Power relations
(4 5) a. b. (46) a. b. (47) a. b.
John pushed the table John pushed against the table John loaded the truck with hay John loaded hay on the truck John smeared the wall with paint John smeared paint on the wall
Fillmore observes that the a -sentences imply a greater degree of affectedness of the entity denoted by the direct object than that denoted by the prepositional object in the b -sentences. A similar observation is made in Haiman ( I 9X 3) and Moravscik ( I 97X). Note that aspect plays a role in this difference. In (46a) the NP the truck and (47a) the NP the wall arc the direct objects of a construction expressing accomplishment. In (46b) and (47b) they are not. The direct object in accomplishments is the centre of the change of state identifying the end of the event. In (46a) the truck has been loaded completely and in (4 7a) the wall has been smeared completely. The b-sentences do not imply this form of complete affectedness. In (46b) it is not known if the truck will be loaded completely and in (47b) it is not known if the whole wall will be smeared. (45a) is not an accomplishment and the semantic difference with (4 5b) cannot be explained the same way. Here it may suffice to refer to the notion of realm of power that I will discuss shortly. This notion also covers (46) and (4 7), as we will see later. Grunau adds that the direct object of experiencer verbs, as in (4X), cannot be characterized using the notion of degree of affectedness. Entities denoted by the direct object are not affected at all with these verbs.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
To describe the semantics of the alternation between direct and prepositional objects Fillmore ( I 977), Haiman ( I 9H 3), and Moravscik ( I 97H) use the notion of affectedness. Grunau ( I 9X 5) characterizes the direct object semantically as being part of a subjectively relevant relationship. In this way the author sets these arguments off from prepositional object arguments. Grunau bases himself, among others, on data from Fillmore ( I 977). These are given in (45) (47)
7M A Localist Model for Event Semantics
(4X) a. I saw him in the shopping mall today
b. She heard his song on the radio just a minute ago
(49) He broke his glasses this afternoon The spray Ipaint verbs are either activities or accomplishments in Vendler's verb classification. Experiencer verbs are part of the class of achievement verbs, and causative verbs are part of the class of accomplishment verbs. The subjectively relevant relationship between subject and object Grunau talks about seems to occur only with achievement (i.e. experiencer) verbs, activity verbs, and the accomplishment verbs that are based on activity verbs through a change in the individuation of their direct object. A d isadvantage of the notion of subjectively relevant relationship is that it also i ncludes prepositional objects, because it is not clear where the cut-off point is between the presence of subjective relevance and its absence. Both sentences in (so) can be interpreted as having subjectively relevant objects, as well as subjectively relevant prepositional objects. (so) a. The gardener planted the garden with sycamores b. The gardener planted sycamores in the garden My own findings make it possible to characterize the relationship between the subject and the direct object more effectively. The data surrounding manner adverbials indicate that the sentence meaning expresses that the relation between the entity denoted by the subject and that denoted by the direct object is such that the former can manipulate the latter. This is not the case with prepositional objects.6 The data also indicate the absence of this type of relation with states and achievements, the latter of which includes the experiencer verbs in (48) . I will call it a power relation for now. Later in this study I will redefine the notion and integrate it into a broader system of event semantics. Grunau's subjective relationship now simply becomes a power relation. The entity
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'The inclusion of a preposition,' according to Grunau, 'implies a level of objectivity and detachment in the speaker. Without it we are talking about a subjectively-relevant relationship between subject and object.' This definition includes achievements like the experiencer verbs in (4R) , bur it excludes accomplishments like the causative construction in (49) , a verb-class not included in Grunau ( 1 9R s). Causative constructions can denote a situation in which there is no subjective relationship at all between the causer and the causee. This is the case when the entity denoted by the subject noun phrase simply touches an object without noticing it, resulting in its falling on the ground and the caused state coming about. This entity only functions as the cause of the event, but it has no subjective relationship to what actually happens.
J an van Voorsr 79
denoted by the direct object is within the realm of control or power of the subject. Causative verbs also exhibit the relation as manner adverbials can focus on their direct object, as demonstrated in ( s 1 ).
( s 1 ) The archaeologist carefully broke the glued plates to put them back together again with a nontoxic glue
subject-direct object subject-direct object power subjective relevance relationship STATES ACHIEVEMENTS ACCOMPLISHMENTS ACTIVITIES
no no yes yes
no yes no yes
3 - 3 Power and aspect
There is a correlation between aspect and the realm of power of the entity denoted by the subject. With transitive activities and accomplishments this entity has power over the entity denoted by the direct object. In addition, it is the change of state in the latter entity that influences aspectual readings in transitive constructions. The relevant data are repeated in (5 3); (53a) is an accomplishment and (53 b), with the indefinite plural, is not. (5 3) a. I had almost stuffed the pillow with the feathers when Peter came home b. I almost stuffed pillows with the feathers but I realized in time that we needed them for something else
D
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Vendler's verb classes fall into two groups when using Grunau's notion of subjective relevance and two different groups when using my own notion of power. As shown in Section 2, states do not allow manner adverbials, which implies that there is no relation of power between their subject and direct object. These results are depicted in (52). A major methodological difference between both notions is that my own notion is arrived at with linguistic tests rather than through a purely interpretative analysis. In addition, my notion can be made part of an overarching theory that captures and unifies a variety of phenomena relating to the semantics of argument structures. This will become clear in Section 4· Grunau does not show that his notion has any further explanatory power. Consequently, it only seems to fulfil a purely descriptive purpose.
8o A Localist Model for Event Semantics
Linguistically speaking, the entity denoted by the subject has no power over what is happening to the entity denoted by the prepositional object in (5 Ja) and (5 3 b), but it has control over the entity denoted by the direct object. This finds its reflection in the data in (54). Despite their difference in aspect, both sentences imply that the agent was being careless with respect to the entity or the entities denoted by the direct object. (54) a. They carelessly stuffed the pillow with the feathers b. They carelessly stuffed pillows with the feathers
( 5 5 ) He *meticulously/*slowly feared this very detailed presentation on depreciation in accounting (56) That treatment of depreciation •meticulously/quickly frightened him The entity denoted by the subject can invoke a feeling in someone else or in him - or herself, but this is all it can do. It has no power over the actual emergence of this feeling and the way it develops. This means that power relationships are absent in the events expressed by the constructions with these psychological verbs. 3 · 4 Merging aspects and power
The power relation could be considered a version of the well-known agent patient relation. However, it is important to note that the notions of agent and patient do not coincide with my 'realm of power' as I included accomplishment constructions. These are normally analysed as causative constructions which do not necessarily have an agent as their subject. Accomplishments with a subject that is not agentive abound and are of the kind given in (57).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The generalization can be quite simple. Aspectual readings of transitive constructions can never be influenced by entities that are outside the realm of power of the entity denoted by the subject. This excludes prepositional phrases from influencing aspect. It also explains why states and achievements are never accomplishment� when their direct object is individuated. The nature of the direct object in terms of individuation is essential for a nonaccomplishment to turn into an accomplishment or vice versa. But with states and achievements the direct object's idividuation has no such influence because something more fundamental in the semantics of the sentence is missing. This is rhe relation of power which makes possible an event in which the entity denoted by the subject is affecting that denoted by the direct object in a way that it is able to change its state. Using the notion of realm of power, the behaviour of constructions containing certain psychological verbs can be better u nderstood, too. As we have seen earlier, the constructions are incompatible with manner adverbials under the VP-reading, as shown in ( 5 5 ) and (56).
J an van Voorst 8 1
(57) a. The wind broke the fence b. The nail ripped my shirt In (57) there is a power relation between the entity denoted by the subject and that denoted by the direct object. This relation is brought out by manner adverbials again. The relevant examples are given in (s8). The adverbial brings out the process of the entity denoted by the subject affecting that denoted by the direct object by specifying its force. I n terms of my notion this means that the former entity has the latter within its realm of power. (s8) a. The wind forcefully broke the fence b. The nail slowly ripped my shirt
(59) a. He loaded the hay on the truck b. He loaded the truck with hay (6o) A:agent [material, location] load B:agent [(material), location] Finally, and most importantly, my notions have not been assigned in a relatively arbitrary, interpretative way as is usually the case when it concerns thematic role notions like agent, patient, experiencer, theme, etc. I have tried to give linguistic evidence from two very distinct domains on which to base my primitives.7 The relevance of the power relation becomes greater when it is linked up with the traditional aspectual classification of verbs. To bring together the notion of power and aspect, I will first discuss several aspectual primitives and show how they correlate with each other and with the power relation. First, a distinction exists between states and events. This disrinction is brought out by adverbials like with no effort , easily , with difficulty , etc. Again an S- and a VP-reading can be distinguished. The VP-reading is the intended one for our purposes. Under the S-teading the adverbials simply apply to the event as a whole. Then they may express that it took no effort to get someone to perform an event, which can be an interpretation of (6 1 a). Both sentences also allow the VP-reading, which is that the process of driving took no effort and that the developing of plans was not difficult.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The power relation also differs from agent-patient pairs in that it only applies to the subject and the direct object noun p hrases. Prepositional objects are excluded from figuring in this relationship. Spray/paint -pairs are often analysed with the same thematic roles as shown in Section 2. Levin & Rappaport's ( 1 986) proposals are repeated in (59). Here no semantic distinction is made between the hay in either sentence and it could be said that both are seen as patients if the role of material is interpreted as a patient-type of role.
82 A Localise Model for Event Semantics
(6 1 ) a. He drove the Plymouth to Toronto during a stormy night without any problems at all b. Despite all the conceptual problems involved he developed these plans effortlessly This type of adverbial can occur in accomplishments, activities, and achieve ments. Interestingly, states do not allow it. The semantic distinction brought out by this adverbial is between something taking place or happening and something not taking place. States do not take place or do not happen. (62)-(6 5) contain the data which illustrates these facts.
The use of adverbials of the above type shows that states do not take place and that the other three aspectual classes denote something that does take place. This is the opposition between states and events which is similar to Dowry's { 1 979) definition as states not involving and the other three classes involving change. Stativity is very different from 'staticity', which is a temporal notion. In the literature events are often called stative when it seems that there is something static about it. For instance, Kamp ( 1 98 1 ) calls stative those events that are expressed through a progressive. form, a present perfect, and a habitual. Yet, the nonstative accomplishments, activities, and achievements continue to imply change in Dowry's sense and continue to imply the dynamics of something taking place when expressed through a progressive, a present perfect, or a habitual form. Bach ( 1 986) distinguishes dynamic and static states (i.e. sit , stand vs. be drunk , own , love , etc.). However, Bach's dynamic states imply the dynamism of something taking place and classifY as activities under the present approach. Other aspecrual primitives are those covering the beginning and end of events. That there is a beginning and sometimes an end to linguistic effects can be brought out in the adverbial of the type in X minutes . Accomplishments allow an ambiguity with adverbials of this type. Reference can be made either to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(62) ACCOMPLISHMENTS a. The mechanic repaired my car without effort b. Mrs. March explained these difficult theorems effortlessly {6 3) ACHIEVEMENTS a. They heard the signals without any trouble b. Peter noticed me in the crowd without any effort {64) ACTIVITIES a. Agnes eats potatoes with a lot of trouble because of her bad teeth b. The horse pulled the wagon without any effort {6 5) STATES a. *Without any effort, Mary matters a-lot to the committee b. *He had that book in his hand with no effort
J an van Voorst 83
the end or the beginning of the event. This is demonstrated by such sentences as in (66). (66) a. He will drive his car to the dealership in an hour b. They will spruce up his backyard in a week The other classes only allow the begin-reading, which is here the interpretation that the event was not engaged in at all. This is the case in (67)-(69), which exclude the end-reading.
The above findings can be combined with my findings regarding manner adverbials as in (70).8 There are several differences between this model and earlier observations regarding adverbials and verb classification. Ryle ( 1 949) states that manner adverbials, which he calls heed adverbs, cannot be used to qualify accomplishments (achievements in his terms) and cognitive verbs such as discover, detect , see . The latter class is included in the achievements in (70). His
ACCOMPLISHMENTS ACTIVITIES ACHLEVEMENTS STATES BEGIN-READING TAKING PLACE REALM OF POWER END-HEADING
YES YES YES YES
YES YES YES NO
YES YES NO NO
YES NO NO NO
task verbs, or the activities in (70), allow manner adverbs. According to my observations, transitive accomplishments can occur with manner adverbials. Ryle also observes that perception verbs arc nor rryings, bur things gor by trying or luck. This marches my observations that these verbs do not exhibit a power relation. Lakoff ( I <J7o) criticizes a proposal by Chomsky ( I <;SS) to link up a feature of manner with possibilities of passivization. When a verb allows manner advcrbials, passive is possible and when ir does nor rhe rule cannot apply. Lakoff shows that this link is incorrect as verbs like hear , sec , think do nor
E
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(67) ACTIVITIES a. He almost towed that car b. The kids almost drank chocolate milk (6X) ACHIEVEMENTS a. I think I almost spotted the clown b. I almost angered him (69) STATES a. You almost had it right b. That sky scraper almost dominates the sky line
X4 A Localise Model for Event Semantics
3 · 5 Other studies a nd proposals
The proposals for semantic primitives gi.vcn so far arc based on linguistic information rather than on the interpretation of what kind of situation the constructions can possibly describe in reality. I used advcrbials and some of the knowledge developed in aspccrual studies. The result of this methodology is the formulation of a power relation that only includes the subject and the direct object noun phrase. Situationally speaking, this outcome is odd as entities denoted by prepositional objects are manipulated by the agent, roo. This is especially clear in some of the �pray lpaillf cases. Another example is the usc of mass nouns and its influence on aspect. When one eats sugar, it is very clear that one only cars a limited amoum of sugar, the exact quamity of which is easy to determine with measuring instruments. Still, the aspccrual interpretation is that
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
allow manner adverbials although they do allow passivization. The semantics underlying this discussion can be explained using (70). The only aspecrual class of which the verbs do not allow passivization is the stative one. Verbs from the other three classes can be passivized. The right generalization for passivization now must be based on the primitive 'taking place' which indicates that something is an event or not. Only constructions expressing events, in the sense of the present approach, of course, passivize. A lot of activities and all accomplishments change aspecrual status depending on the individuation of their direct object. This phenomenon cannot be derived from (7o) and seems to be indcpendem from it. However, in Section 4 I will integrate it with the other primitives. The primitive important here is that of measurability. This is a primitive proposed in Tenny ( 1 <)X<J), bur it is also implied by the many analyses proposing that an accomplishment involves a process leading up to a culmination point. Mass nouns and indefinite plurals simply do not allow us to determine what the culmination point is. When we say 'he drank beer' we do not know how much beer needs to be drunk for the event to be considered finalized. Measurability does not play a role either with the achievements in (70). When we see someone there is no process leading up to the actual 'complete" seeing of rhar person. We either sec him or we do not. The same goes for achievement verbs that have a psychological meaning and when 'we dislike something' there is no process that leads up to a final state of total dislike. It seems very normal to assume that first we could dislike something a little, then a little bit more, and later completely. Interestingly, these interpretations arc not reflected in the constructions' aspecrual interpretations, as I will show in Section 6. Verbs like }ear and Ji"ighten arc achievements and do not allow the accomplishment reading that would be natural if the above assumptions were reflected in the linguistic system.
J an van Voorsr 8 5
o f an activity which is a n event that i s not delimited or that is unmeasurable ' linguistically speaking. The feeling of oddness when looking at these findings arises when they are compared to the situations described by the constructions. The distance berween my findings and the situational imerpretations only goes to illustrate how differently linguistically motivated semantic representations can be from those that are solely based on intuitive interpretations of the data at hand. The latter base themselves on what a construction may refer to, or on its extension. These analyses ignore the existence of a level of purely linguistic meaning as has figured so prominently in the work of linguists like Hjelmslev. Hjelmslev's distinction berween contents and sense in his Prolegomenes ( 1 943 ) is well known but often ignored in modern semantic analyses of argument structures.9 Coments is that aspect of semantics that can be found closest to the grammatical system; sense is the semantics that is more remote from the grammatical system. La Cat��toric dcs cas ( 1 93 5 ) is a semantic analysis of case systems that very closely follows this distinction. In this study Hjelmslev warns against the use of extra - linguistic judgements and argues that one needs to use the syntagmatic and paradigmatic relations existing in language. A classification of ideas that does not take into consideration how they arc expressed by the linguistic system is to be rejected. The consequence of formulating semantic primitives without taking into accoum linguistic facts leads to a very loose system with ill -defined primitives. Most of them then end up being based on the personal preference of an individual linguist. A good example of such a situation is the primitive of affectedness which has been used in the spray/paint discussion. In his discussion concerning the spray Ipaint alternation, Fillmore ( 1 977) observes that the sentences in (7 1 ) differ in that the wall in the a- sentence is more completely affected than in the b-scntence. In addition to specifying case or thematic roles, Fillmore proposes a Saliency Hierarchy that helps determine the perspective taken in a construction. Perspective basically describes the mapping relation berwcen case or thematic roles and grammatical functions. ·
a . I smeared the wall with mud b. I smeared mud on the wall (*in a half hour)
Jeffries & Willis ( 1 984) argue against a case or thematic role description altogether. These roles are said not to give an accurate description of the sentences in (7 1 ). They are different in that (7 1 a) has a holistic reading and (7 1 b) a partitive one. The authors argue that the holistic/partitive distinction is not a general one, but depends on the verb meaning, the definiteness, and countabiliry of the object, and pragmatic knowledge. The very different roles aspect and situational or pragmatic interpretations have to play is not recog nized. Boons ( 1 974) is acutely aware of the distinction berween the rwo and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(7 1 )
X6 A Localisr Model for Event Semantics
distinguishes completion from holistics for similar cases in French, thereby respecting Hjelmslev's contents and sense distinction. He separates completion and what the state of completion looks like in reality. Completion is directly related to aspect. When the i ndividuation of the direct object in (7 1 a) is changed, the original sense of completion disappears. When it is changed in (7 1 b), the sense of completion appears. This is demonstrated in (72). 1 0 (72) a. I smeared walls with mud b. I smeared the mud on the wall (in a half hour}
(7 3) Hopper & Thompson: Transitivity
A B C D E F G H I J
PARTICIPANTS KINESIS ASPECT PUNCTUALITY VOLITIONALITY AFFIRMATION MODE AGENCY AFFECTEDNESS OF 0 INDIVIDUATION OF 0
HIGH
LOW
2 or more parnetpants action relic punctual volitional affirmative real is A high in potency 0 totally affected 0 highly individuated
1 participant non-acnon atelic non-punctual non-volitional negative irrealis A low in potency 0 not affectd 0 non-individuated
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The data in (7 1 ) and (72) an.· illustrative of the aspecrual distinction between accomplishments and nonaccomplishmenrs. This is a linguistic distinction based on completion. The totality of the affectedness is an outflow of this and it docs not need to figure prominently in a semantic theory to account for these data. The notion of affectedness can also be found in Hopper & Thompson ( t l)Xo). This study proposes the list of seemingly unrelated notions in (7 3 ) to character ize transitivity. Using my analysis, a large number of these notions can be related in an insightful way, thereby creating some understanding of why they can be found together in the first place. Transitive accomplishments have two participants, they imply an action (if we stretch the idea of action somewhat, causation can be included under this notion}, are relic, have an agent high in potency (if we call the cause an agent), and they have an object that is totally affected and highly individuated. All these properties are a direct consequence of a construction being an accomplishment. It is not necessary to list the primitives independently as they all hang together with completion and the power relation.
J an van Voorst 87
(n) f--o z
�:r:: Vl
:l 0..
HOPPER & THOM PSON BEGIN - READING PARTICIPANTS (2) TAKING PLACE REALM OF POWER AGENCY high in potency ACTION ASPECT-TELIC END- READING AFFECTEDNESS OF 0 totally affected INDIVIDUATION OF 0 highly individuated
� 0 u u <
Vl
Vl
:'::! f--o
> �
u <
f--o z
�
� >
:'::! :r:: u <
Vl
� f--o < f--o
Vl
YES YES YES YES YES YES YES YES NO
YES NO NO
YES YES NO YES NO NO YES NO NO
NO NO NO
YES NO YES/ YES/ NO NO
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Hopper & Thompson use their notions to place constructions on a scale of transitiviry. In my system various primitives arc used to characterize all four aspectual classes very specifically. A construction belongs to a specifi c aspecrual class and consequently has certain observable semantic properties. When encountering a phenomenon the explanation is not that the construction is high on the transitiviry scale but reference can be made simply to one or two spccitlc semantic primitives. Hopper & Thompson's proposals arc hard to refute as they are expressed in relative terms. My own system leads to easily refutable proposals. Interestingly (74) shows that if Hopper & Thompson's notions arc interpreted in a broad way by including causation as a form of action, a strong overlap becomes evident between my notions and theirs. They basically lead to the same subgroupings of aspectual classes. Note that with achievements and states it does not matter if rlw direct object is individuated or not. Individuation does not change the aspectual classitlcarion of verbs belonging to these classes.
XH
A Localist Model for Event Semantics +
A L O C A L IST APPROACH TO ASPECTU A L C LASSI F I C A T I O N
The system developed so far is not coherent yet. (7o), which i s repeated below, contains notions that are at first sight wildly divergent. In addition I added the notion of measurability of events to explain the particular behaviour of activities. It is the purpose of this section to create a more coherent system.
BEGIN-READING TAKING PLACE REALM OF POWER END-READING
YES YES YES YES
YES YES YES NO
YES YES NO NO
YES NO NO NO
It is not only important to use linguistic rather than interpretative evidence to establish the nature of semantic primitives, but the creation of a consistent system is important as well. Consistency will make the theory stronger and it also sets the limits for the addition of new primitives and the way these primitives may look. Localist approaches to the study of the semantics of case systems have very strongly focused on the creation of a coherent overall system. Hjdmslev ( 1 93 5) and Wtillner ( 1 X27) are two excellent examples of such an approach. Hjclmslev creates cohesion in his theory by using three oppositions in his meta - systems of case semantics. The basic idea of the system is that morphological cases and prepositions express relations between objects. The three oppositions that characterize these relations are: ( 1 ) direction (approach, distantiation); (2) cohesion and absence of cohesion; (3) subjectivity and objectivity. Semantic systems for the analysis of cases in a particular language are built along these lines creating a three -dimensional system. The present analysis is far from an overall semantic analysis of all noun phrases in English. It only focuses on subject, direct object, and prepositional object noun phrases. Still, the outcomes of my analysis deserve to be put together in a more encompassing and coherent system of analysis. The oppositions 'approach vs. distantiation' can be used to characterize the relation between subject, direct object, and prepositional object noun phrases. The entity denoted by the subject noun in events (i.e. achievements, accomplishments, and activities) can be seen as the starting- point of that event. It establishes a spatial relation with other entities in its environment such as those denoted by a direct object or a prepositional object. The entity denoted by
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ACCOMPLISHMENTS ACTIVITIES ACHIEVEMENTS STATES
J an van Voorsr 89
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the subject noun 'approaches' or orients itself to entities or an entity in irs surroundings. This is a purely localise conceptualization of argument structure semantics and it is quire different from thematic interpretations. What is important in agent-patient and cause-effect relations is that they involve an approach of the patiem or the effect by the agenr or the cause respectively. With seeing-, hearing-, and other events the 'experiencer' can be seen as an entity that establishes a relationship with another entity in irs environment. The seeing and hearing are the means through which this relation is established. In a way the seer or hearer 'approaches' rhe object perceived and includes it in its realm. 1 1 States differ from the other three aspecrual classes as they d o nor imply an approach. The entity denoted by the subject of states does not establish a relation with another entity. The relation is that between an entity and one of its properties. The notions of cohesion and absence of cohesion help make more precise distinctions. 1--ljelmslev defines cohesion as a relatively close connection between two objecrsY A preposition like into expresses cohesion, along is said to be unmarked for cohesion, and at expresses absence of cohesion. Only into implies the contact necessary to speak of cohesion between two entities. The cohesion or absence of cohesion that is relevant for an event must be a consequence of the dynamics of this event. Cohesion existing before an event starts does not count as one of its defining properties. Food can be in my mouth before I start eating bur this is not the cohesion important to the localise defini tion of eat . What is rclevanr is the cohesion brought abou t by the emity denoted by the subject through the event of eating. Wirh a seeing-event the relation establishes through this event between an entity seeing and an emity seen is crucial. This relation, as we will see, implies absence of cohesion. If the object happens to be in my hand, the cohesion between it and my hand is not relevant. When the three nonstative aspectual classes arc comrasted, there is a dif ference in the type of cohesion between the entities denoted by the subject and the direct object. With accomplishments, the entity denoted by the direct object undergoes a change of state. It can be destroyed, created, consumed, nc. In spatial terms, the entity denoted by the subject noun approaches the other entity and, upon contact, this conract extends through the latter entity from irs one extreme to its other. This is another way of talking about completion, a notion implying that an entity is affected in totality. With activities a different situation occurs. Now the entity denoted by the subject approaches the direct object and, upon contact, this contact extends through the entity bur not from one extreme to the other. The conract is partial. This is the partitive reading that can be related to mass nouns in cases like 'eating sugar'. Not all the sugar need bL' gone for the senrence to be true. Achicvemetits arc diffnenr in another way. Here the entity denoted by the direct object is approached in abstract sense and
90
A Localisr Model for Evenr Semanrics
'
-
'
(75) Event semantics Begin-Reading -- - Taking Place - -- Realm of Power - - - End-Reading
(Subject NP) -
t
State
- - -
1\
�t t t t
Approach - - - - - � (Direct Object NP)
Achievement
Activity
Accomplishment
Accomplishment in this model is directly related to the direct object. This is in accordance with the many observations in the literature that acomplishment or perfective aspect depends on the individuation of the direct object. The localist model establishes a natural connection between the spatial notion of measurability and the interpretation of events. In Carlson's ( 1 9Xo) analysis indefinite plurals are analysed as either (proper) names of kinds or as referring to the various occurrences or stages of obj ects. The exact number of these occurrences is left unspecified. Carlson also suggests that mass terms can be interpreted as names for kinds of things and Link ( 1 9X 3) gives a set of portions of matter for the extension of mass terms. Under each of these proposals, the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
made part of an event, but the entity denoted by the subject does not proceed further. It stays outside the former entity. In terms of cohesion, this means that accomplishments imply full cohesion, activities partial cohesion, and achieve ments no cohesion. Prepositional objects are like the direct object of achieve ments in that no cohesion occurs between this object and the approaching entity. The temporal notion of aspect now has been reinterpreted as a localist one covering the spatial relationship between the entities playing a role in an event. (70) can be reformulated as follows. The subject either expresses 'approach' or it docs not. This means that it either creates a relationship with another entity or it does not do this. This is rhe difference between events (i.e. accomplishments, activities, and achievements) and states and is reflected in the possibility of advcrbials of the type with no problems . In (70) this is expressed by the primitive 'TAKING PLACE'. The primitive relating to the starting point of the 'approach' is covered under the begin reading . The approaching entity then meets up with another entity with which it establishes a relation of cohesion or not. It is this dynamic aspect that is special for ewnts. The creation of cohesion is the process modified by manner adverbials. If no cohesion is created, we have an achievement; if it is, we have an accomplishment or an activity. The power relation introduced in Section 3 is a direct consequence of the existence of cohesion. Cohesion may involve the object partially or completely. In the first case, we have an activity; in the second, an accomplish ment. in the latter case, we have the end-reading. The new version of(7o) is (7 5).
J an van Voorsr
91
(76) a. ACCOM PLISHMENT He almost/quickly stuffed the envelope b. ACTIVITIES John almost/quickly petted the cat c. ACHIEVEMENTS The comic almost/quickly upscr me wirh his racisr j okes d. STATES He almost/quickly had a headache! The diagram given in (75) has to be modified as Hjdmslcv's primitive of approach does not seem to find linguistic evidence. This can be done as follows. An approach is notjust a movement, abstract or not, ir also includes an outcome which is a spatial relation between two entities. When rhe movcnH.'nt part is omitted and this spatial relation maintained, it is possible to redefine events. Transitive events now express that the ent i ty denoted by the subject noun
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
exact number o f things or portions is left unspecified. When the number of entities referred to by a direct object is specified we can say that the entity it denotes is individuated, or measurable in Tenny's terms. The fact that spatial aspects of objects (i.e. their ddimitability in space) play a role in aspect is nothing surprising within the context of the locative model. Toral cohesion depends on the measurability of events. In all other cases nothing can be measured. We either do not know the exact extent of the cohesion or there is no cohesion at all. The former situation occurs with activities, the latter with achievements and states. (75) Is not yet a good enough reflection of the semantics of argument structures. So far I have not given evidence for the approach-part of the event preceding the establishmem of cohesion. To prove its existence, we can simply usc the test with the adverbial almost . The adverbial either expresses the non occurrence of the beginning or of the end of an event. Obviously, if it cannot do this for the approach, it makes no sense to speak of such an approach. Manner adverbials expressing speed can also be used to test the existence of this approach. If it exists, it can be slow or fast. Interestingly, almost in (76a)-(76c) can never relate to the hypothetical approach-part of (75) and express thar although the approach was on its way to reaching its goal, it was interrupted and contact between the entity approaching and the entity approached did not occur. This is the contact between the direct object and the arrow of the approach line in (75). In addition adverbials like quickly can nor specify the speed of rhc approach parr and do nor give linguistic evidence for irs exisrcnceY (76a) can only mean rhar rhe event was engaged in after a relatively short time span had gone by or that the very act of stuffing was performed quickly. (76b) exhibits a similar ambiguity. The other two examples can only mean that the evem in (76c) or the state in (76d) starred soon.
92
A Localise Model for Evenr Semanrics
(77) Event semantics
Inclusion
Begin-Reading Syntax
t
o
Subject NP
Cohesion
o
:
--�---D -i-__ rec t O b�ec t N P
State ------+ Achievements ----;.-;� Activities ------r---L---+-
I I I I I
Accomplishments ____.l._...:-------+i almost in X minutes slow 1 y care fully
1
: 1
*""-- slowly ----+1 '
1
� :+- carefully � with no effort ��
4--..!.--- for X minutes ---+-1
End-Reading Prep. Object NP
almost in X minutes
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
phrase establishes a spatial relationship with another entiry. The nature of this relationship is specified by the verb. We can say that the former entiry includes the latter inside its realm. If no inclusion occurs ar all, we have a stare. The inclusion may be jusr rhat and nor lead ro some form of closer conracr. However, if ir does lead up ro a form of closer contact, ir will result in contact only i nvolving parr of the object or conracr involving this object in its totality. In the former case we encounter partial and in the latter total cohesion. 1 4 Almost indicates that the cohesion failed ro be complete under one of its readings. The dynamic process of the creation and mainrenance of cohesion is what manner adverbials like slowly or carefully modify under their VP- reading like in (76a) and (76b) bur nor in (76c) and (76d). This is also the process, sometimes called the 'evenr', that figures prominently in many aspecmal studies as a s ubpart of the semanric structure of accomplishmenrs or activities, such as, fix instance. Pustejovsky ( I l)X X). The diagram only changes in rhat there is no approach before the direct object is reached bur only an inclusion of this object. The model that emerges is falsifiable as it is wholly based on easily identifiable linguistic evidence. Thematic role and conceptual approaches are much harder to falsify as they do not give linguistic evidence of this nature to motivate the primitives proposed. The aspectual classes are defined on a sem.antic continuum of the creation of increasing spatial closeness which is expressed in the diagram for each aspecrual class by way of the lines. I have added the traditional aspectual tests with the adverbials for X 111i11utcs and in X minutes to n1y diagram (77). These seem to create the same distinctions as almost . 1 5 Notions like begin - and end - reading in the model have to be interpreted not as denoting the begin or end of a rime line bu r as the beginning and the end of an orienred spatial relationship between entities.
J an van Voorsr
93
Activities can be further distinguished by fine-tuning rhe notion of cohesion. This makes it possible to capture rhe meaning of verbs like pus!t , pull , !tit , ere., which all imply rhar the cohesion does nor extend into rhe entity given by rhe direct object. This can be expressed as in (7X} with the term of minimal cohesion. The traditional aspecrual notions of perfectivity and imperfecriviry are represented as forms of spatial inclusion in (7X). Perfective aspect is total (78) Event semantics Inclusion Begin-Reading o
o End-Reading
Cohesion
r
I I 1 minimal
:¥ I
Subject NP
partial
I I
total :
/ '\;.
�
Direct Object NP
Prep. Object NP
cohesion; i mpcrt�·crive aspect is lack of toral cohesion. M i n i mal cohesion exc l u d es a n accomplislnnem reading when rhe d i recr obj ect is ind i vid ua red. The begin-readi ng
of verbs
l i ke p1111 . ere. is possi ble in (7lJ) bur no r rhe end
rea d ing. An acri v i ry i nvolving partial cohesion wou ld turn i mo a n accomplish m e nt
in
rhis case.
(7lJ) a. The horse al most pulled the carr
b.
They al most pushed me
The model can be reformulated in a d i fTerenr more purely localisr way. The notion of inclusion i s a d ynamic one and i m pl ies rhe bringing i nr o exis tence of a n evenr b u r ir also has a localisr conrent of being 'connected i n space' one way or another. Cohesion is a result of inclusion and is spatial closeness through comacr wi th the varyi ng degrees of inrensity being m i n i m a L partiaL or total. W hen we separate the localist from the dynamic content, we arrive at a rype of semamics that reflects the finding of Nikanne ( I l)lJO) regarding Fi n nish prepositions and cases. Basing himsel f on J ackendoff. N i kanne proposes the fea ture-system as depicted in (xo). There i s a nice overlap berwl·en th i s and my own proposals when we add rhe notions of pu re attachmem and u n i ty to rhis system. Evem semamics i s
the comb i n a tion
of rhe above localisr t�·arures a n d evem
dynamics. In even t semanrics we are tal ki ng
about th e bringing i mo existeJJCl' cnrain rime period. of a purdy loca ti ve rel a tion, a relation of pure comacr, of p u re a ttachmem or of u n i ty. Al though ir wi l l be essenrial t�Jr rhe ana lysis ofimransitive verbs to distinguish the dynamic and sta tic (spatial) aspects ofevenrs. I will nor treat these as separate i n this study
a n d rhe keeping
in
ex isten ce d u r i n g a .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
---�-----
94
A Localist Model for Event Semantics
(80) Location
�
Pure Location
Contact
� Attachment
Pure Contact
�Unity
Pure Attachment
achievement
I
activity partial
cohesion
cohesion
I
accomplishment total cohesion
Finally, it should be noted that although event semantics is conceptualized localistically, it also applies to events that occur in nonspatial domains such as that of possession, that of psychological events, etc. This is a commonplace extension of the localist idea to nonlocal domains that can be found in Hjelmslev ( 19 3 5) but that also figures very prominently in the semantic generali zations made in studies like Gruber ( 1 976) and Jackendoff ( 1 972, 1 990).
5 A S P E C T U A L A P P R O A C H E S TO E V E N T SE M A N T I C S I n this section I compare the present localist conceptualiz ation of verb classification with some aspectual analyses that conceptualize events using temporal notions. Dowty ( 1 979) distinguishes Vendler's ( 1 967) four verb classes using logical definitions and the primitives BECOME, DO and CAUSE. These are combined with a formula representing states. In (8 1 ) I have given the cases relevant for the present discussion. a and j3 stand for arbitrary individual terms, l111 and Pm stand for arbitrary n-place stative predicates, and � is an arbitrary formula. (8 1 ) a. Simple Statives: John knows the answer Jtll(a J , · · ., all) b. Achievements: John discovered the solution BECOME[n11(a � . . ., al l) ' c. Activities: John is walking DO(a1 , [n11(a 1 , , a11)]). d. 1 . (Non - Intentional) Agentive Accomplishments: John broke the window [ [DO(a � ' [n11(a1 , , all)])] CAUSE [BECOME [Pm(l3 1 , •
•
•
•
•
•
•
•
•
, 13m)])]]
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
activity minimal
J an van Voorsr
95
2. Intentional Agentive Accomplishments: John murdered Bill DO(a 1 , [rt11(a 1 , , a,J) CAUSE �]) where � may be any non-stative sentence •
•
•
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The transitive stative in (X 1 a) expresses the occurrence of a predicative relation between two entities. I have defined states negatively. They do not imply inclusion, nor do they imply cohesion. This means that they do not take place and that the relation between the entities involved is a nondynamic one. As such my states are not very different from the ones in (8 1 a). The achievements in (8 1 b) involve a change of state. My achievements imply the inclusion of rhe entity denoted by the direct object by rhar denoted by the subject. The change of state Dowry talks about is the coming into existence of rhe inclusion. This coming into existence is also part of activities and accomplishments. After this change of state the achievement event may last for a longer or shorter duration and be durative or punctual. The latter two temporal notions are nor defining parts of my model and achievements are nor typically punctual contrary to what is argued in many other approaches. Durative perception verbs like see and hear and psychological verbs are also achievements. Dowry's model does nor capture the dynamic relation of inclusion nor does it permit durarive achievements. The activity in (X 1 c) is defined by the primitive DO and a state. The first primitive expresses the involvement of the agent in the event and implies, al though does nor express explicitly, rhar the agent brings the event into existence by establishing a relation between himself and the entity denoted by the direct object. I take it to be the case here that Dowry's analysis of the intransitive walk extends to transitive verbs without further modifications. The process of cohesion is nor expressed by the embedded state in (X 1 c). Accomplishments are defined as an action causing a change of stare or as an action causing an action. My model does not include any reference to intentionality. Note that it is also not clear to Dowry that this notion is a linguistically relevant one or nor. Accomplishments are simply inclusions with total cohesion resulting from it. The change of state, that figures so prominently in aspectual studies, is a subcase of cohesion and has no special status in the model. The distinction between subjects denoting events or indi viduals, as often expressed by a CAUSE versus DO opposition, is also not expressed in the model. The model implies that this opposition has nothing ro do with how events are structured. It will have to be captured in a different way, possibly by distinguishing ways in which individuals can be understood referentially, as outlined by Coppieters ( 1 982). However, treatment of this important issue goes beyond rhe scope of this study. Punctuality and durariviry are nor building blocks of rhe localisr model. Punctuality is the defining factor of achievements in Vendler's and Dowry's
96
A Localise Model for Event Semantics
(82) a. b. (8 3) a. b.
He will type the letter A in one second He will type the letter to his brother in five minutes He will find his wallet in ten minutes I n ten minutes or so, as you get closer to the marsh, you will notice the ground getting softer
The present localise analysis does not take into account temporal aspects of events. I nstead it uses cohesion or inclusion to describe aspects of events that may be but do not have to be punctual. This makes it easier to distinguish accomplishments from achievements. (84) shows that the sentences in (82) are similar in terms of cohesion. A possible reading for these sentences is that the he is being careful not to hit the keys too hard in order not to damage the paper or to make too heavy imprints. The achievements in (8 3) do not allow manner adverbials under their VP-reading that the finding or noticing were being performed in a careful manner or that they were proceeding slowly. This is shown in (8 s). Slowly in (8 sb) allows an S-reading which is that it took some time before the man was noticed. (84) a. He carefully typed the letter A to make it as dark as the letters surrounding it b. He carefully typed the letter to his brother as he was concerned about typos (8 s) a. He *carefully/*slowly found his wallet b. He *carefully/slowly noticed the man in the corner Although the in X minutes test shows that it is very hard to separate punctual accomplishments and achievements, I do not think that this gives sufficient grounds to abandon the distinction. The tests with the manner adverbials show that accomplishments imply cohesion but that achievements do not. A punctual accomplishment is simply a very short period of cohesion culminat-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
temporal analyses of events and it has found its way into a large number of more formally oriented studies on event semantics such as, for instance, Bach ( I 986) and Kamp ( I 98 I ). However, working within the context of a temporal analysis, Verkuyl ( I 989) relegates the notion of punctuality· to the area of pragmatics. 'Typing one symbol' on (82a) is punctual but 'typing a letter' in (82b) is not and can take some time. The end-reading of the adverbial in X minutes is odd in a punctual sentence but not in a durative one. Verkuyl uses data like (82) and (8 3) to abandon the opposition between achievements and accomplishments. Achievements do not allow the end-reading either with the adverbial in X minutes . (8 3) simply expresses the occurrence of the punctual achievements offinding and noticing after ten minutes have gone by just like (82a) expresses the punctual event of the typing of the A after one second has gone by.
Jan van Voorst
97
ing in a change of stare. A punctual achievemenr is a very shorr period of inclusion without a change of stare as its culmination. The change of stare often talked about in the context of achievements is the coming into existence of the inclusion relation. Even accomplishments that look punctual at first sight allow manner adverbials. (X6a) and (X6b) look like punctual accomplishments bur (X7) shows that these events can be performed slowly and carefully indicating rhar they rake more rime than an indivisible moment.
(X7) shows that it is easy to force a durarive reading on seemingly punctual accomplishments. The change in puncrualiry does not influence the semantic properries of rhe argument structure in terms of the localist model. This indicates that cohesion is a more fundanwntal property in the definition of events than punctuality. The problem of punctuality leads to another issue in rhe srudy of verb classification. Verkuyl ( 1 9X9) concludes on the basis of the pragmatic nature of punctuality that it is impossible to maintain the four-way classification and that it needs to be replaced by a tripartite division. This division distinguishes between events (i.e. the original accomplishments and achievements), processes (i.e. the activities), and states. In addition to belonging to a certain class, verbs arc characterized with a primitive called additivity. This primitive indicates if a verb implies a process that lasts more than one rime instant. States always carry the minus value for the primitive, processes and events the plus value. Verbs like push , pull , carry , etc. arc a problem for this analysis, as it is recognized by V crkuyl ( l l)X9) in his note 5· They are marked as plus additive like cat , drink, break , etc. Still, if their direct object is individuated (or expresses a specified quantity in Verkuyl's terms) the construction does not become terminative as shown in (XX). The sentences do nor allow rhc end - reading. (XX) a. The horse almost pulled the wagon b. They almost pushed the car When analysing events localistically, rheSl' cases do not constitute a problem. They imply minimal cohesion, indicating that the direct object cannot be the cenrre of a change in their aspecrual srarus. Duration also plays a role in Krifka's ( 1 9X9) formal analysis of evenr semantics. The activity and accomplishmcm readings of dri11k are created by combining rhe formula for rhe prcdicats for mass nouns, definite and indefinite nouns with the formula for rhe verb. The author stares that the way an object
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(X6) a. He broke the cookie b. He snapped the breadsrick (X7) a. He slowly/carefully broke the cookie so that the rwo pieces would be of equal size b. He slowly/carefully snapped rhe breadsrick in rwo equal pieces
9H
A Localist Model for Event Semantics
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
refers is carried over on to the verb. Some predicates are analysed as 'kumulativ' and consist of subparts that are similar in nature to the larger unit they make up. Other predicates are analysed as 'gequantelt', indicating that the subparts are dissimilar from the greater whole. However, rhe carrying over of referential features does not take place with verbs like see . This verb's aspectual status does not depend on its direct object. In order to account for this, the author distinguishes between 'Sukzessiv-Patiens' ('successive patients') with verbs like drink and 'Simultan-Patiens' ('simultaneous patients') with verbs like see . The object of the former verbs is included in the event step-by-step but the object of the latter is included in it immediately. Constructions with see are not necessarily 'gequantelt'. They are more cumulative, as one can see a subpart as well as the bigger whole at the same time and subparts of seeing are included in the larger event of seeing. The distinction between the two types of verbs is similar to what Verkuyl expresses with the notion of additivity and relates to the well -known distinction between events having a process that may lead up to a change of state and those that do not include such a process. Within the context of the present model no refuge has to be taken to notions like 'Sukzessiv-Patiens' and 'Simulran-Patiens', or additivity. The aspectual phenomena Verkuyl and Krifka try to capture this way follow quite naturally from the localist analysis. Perception events do not imply cohesion, making the bringing about of a change of state, which is subsumed under the accomplish ment reading, impossible. Verbs like drink imply partial cohesion and, therefore, allow this reading. Note that no reference is made to the temporal properties implied by 'Sukzessiv-Patiens' and additivity, thereby avoiding the pragmatic problems surrounding punctuality and durativity. In addition, this lack of any reference to the temporal structure of events makes it possible to create a natural connection between events and objects, as both are spatial in nature. As mentioned earlier, Krifka states that the way an object refers is carried over on to the verb. However, this carrying over is from spatial entities, which are cumulative or not, to temporal entities, which are step-by-step processes or not. There exists a more natural connection between cumulativity and event space. When something is cumulative, cohesion takes place in a nondelimited event space. It is impossible to determine a point in space at which total cohesion is a fact. When something is 'gequantelt' we do know the limits put on the event space and it is possible to determine when total cohesion is a fact. 1 7 The localist analysis of events makes it possible to separate the semantics related to verb classification and the semantics related to tense phenomena. Binnick ( 1 99 1 ) goes so far as to suggest that three dimensions have to be distinguished in the study of aspect. These are, first of all, the Aristotelian categorization or Vendlerian classification such as the ones focused on in this study; secondly, Akrionsarten which is defined as applying to the phases of
J an van Voorst
99
(X9) a. STATE: I have almost had a heart attack, too b. ACHIEVEMENTS: He had almost scared me with the intensity of his anger c. ACTIVITIES: I had almost eaten shaving cream mistaking it for whipping cream d. ACCOMPLISHMENTS: I had almost eaten the chocolate bar, when the nurse stopped me horrified Secondly M u fwene ( 1 9X4a) shows that the progressive does not discriminate between aspecrual classes and can apply to verbs from all four of them. It is well known that accomplishments lose their end -reading when in the progressive form in English or in the imparfait in French. This phenomenon is at the base of the imperfective paradox. For the other classes the changes are less clear. Adverbials can help us again to puc a finger on what is going on here. The state (90), although in the progressive, still does not allow adverbials of the type with no tjfort or of the careji1lly - type. The achievement (9 1 ) does allow the former adverbial but not the latter. The activity (92) allows both adverbs, as well as the accomplishment (9 3).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
siruations or subsiruations studied in Freed ( 1 979) which are expressed by aspecrualizer verbs like begin , continue and resume; and, thirdly, aspect proper which concerns the relationship between a siruation and a temporal frame against which it is set (Binnick 1 99 1 : 2 1 3 ). In the extensive literature on these topics, it happens often that these three are not distinguished. Reyle's ( 1 9X7) formal analysis of sequences-of- tense in discourse uses notions from classifica tional semantics to characterize tense phenomena. Moens & Steedman ( 1 9XX) mix semantics from Binnick's first and third dimension in their proposal to account for various temporal and aspecrual interpretations. Binnick himself shows how the English present perfect or the French passe compose is often said to express accomplishment or perfectivity. 1 8 The English progressive or the French imparfait is often characterized as stative or as a non-accomplishment. The terminological confusion is widespread and when using the spatial analysis of events, it can be partially undone. Binnick's first dimension then simply becomes nontemporal, setting it off from the other two temporal dimensions. That separation of the first from the other dimensions must be on the right track becomes clear from the following two examples. First of all, whatever it is that the present perfect expresses, it cannot be accomplishment in the sense of total cohesion. If this were the case then all sentences in (89) would imply a process leading up to a final change of state. Yet, the inter pretations of the sentences have not changed in this respect and the end reading with the adverbial almost is the same with or without a present perfect. It is excluded with states, achievements, and activities but not with accomplishments.
1 00
A Localist Model for Event Semantics
(9o)
STATES a. *Without effort, he was having a heart attack b. *He was very careful having a fit over that (9 1 ) ACHIEVEMENTS a. They were hearing these signals without any trouble b. *They were carefully hearing these signals (92.) ACTIVITIES a. Agnes was eating potatoes with a lot of trouble because of her bad teeth b. Agnes was eating nucs carefully because oF her bad teeth (93) ACCOMPLISHMENTS a. The mechanic was repairing my car without effort b. The mechanic was carefully repairing my car •
(94) a. STATES I'm sure that Mary will be having a fir over these changes in a few minutes b. ACHIEVEMENTS They will be hearing our signals in an hour or so c. ACTIVITIES Agnes will be eating potato chips in an hour when the visitors arrive d. ACCOMPLISHMENTS The mechanic will be repairing my car in an hour The progressive affects all verb classes in the same way and suppresses the begin-reading. But as we have seen earlier in this study, constructions from every aspectual class imply the begin - reading, so it cannot be maintained that the semantic change the progressive brings about is a localist change in aspectual class. As such the progressive docs not coerce a construction into another localist class. The semantics the progressive imposes goes beyond the semantics covered by the present modcl. 1 9 The cwo notions most strongly related to verb classification in Reyle's analysis are Zustand, 'state', and Ereignis, 'achievement'. These cwo seem to be crucial for the discourse analysis of tense sequences. In Reyle's study, they are temporal and help determine how events can be related to each other in discourse. But as shown earlier, constructions entering the present perfect (i.e. an Ereignis) or the progressive (i.e. a Zusrand) do not lose their localise characteristics or they undergo changes that seem hard to define as localist ones. The use of the term achievement to cover punctual icy is rather widespread and
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What does change is the begin -reading related to the adverbials in X minutes . The presence of this adverbial makes i t no longer necessary to place the beginning of the event or the state at the point of rime it indicates. They may have been going on before this point in rime.
Jan van Voorst 1 0 1
6
A N A PP L ICATION O F THE L O C A L I S T M O DE L
This section is an application of the localise theory to constructions containing certain psychological verbs. The roles of experiencer and theme are often used to describe these verbs and then it seems that constructions like those in (9 S) are each other's mirror image. This type of analysis can be found in studies such as Belletti & Rizzi ( 1 9XX), Grimshaw ( 1 990), and in many other places. The former article proposes a derivation of the argument structure ofJr��hten from one in which the experiencer and the theme both occur post-verbally. (9S) a. Joh t\.,p..-r;,.,"",·r detests detail work,�o,.,". b. Detail workrl"·m,· frightens J ohn,·xpcri,·11ccr Pesetsky ( 1 990) proposes another description of these facts. His proposal looks like that in (96). Note that the primitive CAUSE in (96b) groups the verb with
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
dates back to Vendler's original proposals and is also present in Dowry's analysis discussed at the outstart of this section. However, temporal phenomena are external to the present model of event semantics and any one of the four classes contains durative as well as punctual verbs. Therefore when the semantics used by Reyle to characterize events as interpreted this way it is unrelated to the analysis of verb class classification proposed in this study. One could wonder because of this if it might not be sufficient simply to use the notions of duration and punctuality and omit any reference to classificatory semantics in the analysis of discourse phenomena. Reyle seems to be aware of this possibility, as he discusses Vendler's system only very briefly and does not attempt to integrate it into his own proposals. MufWene's careful analysis of the English progressive is an independent argument to separate verb classification and tense phenomena. MufWene shows that the existence of a correlation berween these classifications and the progressive is ill founded and that possibilities of this form are strongly related to temporal properties inherent to an event or a state. The correlation is not an absolute one and depends on a number of pragmatic factors. The more transitory an event or state is (i.e. or the shorter its natural duration), the more likely it is that it has to be used with the progressive form. The localise model of event semantics conceptualizes events in a way very different from aspectual models. It focuses on the dynamic process of the creation of spatial closeness. Notions like change of state, causation, action, event in the sense of subevent, etc. do not figure prominently in this rype of model. Temporal notions such as punctuality and durativity are also external to the model and will have to be accounted for at a level different from that implied by the present analysis.
1 02
A Localisr Model for Event Semantics
accomplishments from the break -type which are often characterized as causatives in thematic role or conceptual descriptions. (96)
a.
J ohnexperienccr detests
detail wor�argcr/subjecr matter
b. Derail workcause frig htens Johnmriencer
Grimshaw, basing herself on Pustejovsky ( 1 988), adds an aspecrual tier to her model and describes psychological verbs as in (97). The lines i ndicate the relation beween the roles and the aspectual tier. (97)
Frighten
b
Fear
Experiencer
><
Theme
(Thematic Tier)
CAUSE STATE (Aspecrual Tier) Experiencer Theme (Thematic Tier) I I STATE STATE (Aspecrual Tier)
Furthermore, Grimshaw states that psychological verbs like Fighten may also have an agentive subject. These cases have a separate lexical entry. Psychological verbs of thefrighten -type allow manner adverbials bur only when the adverbial focuses on the actions performed by the entity denoted by the subject as in (98). Interestingly, the process- reading is impossible. The sentences in (98) cannot mean that the psychological event of frightening itself was done carefully as John has no control over the feelings of the bear. (98) a. Peter tried to carefully frighten the grizzly bear to chase it out of his backyard b. Peter tried to frighten the grizzly bear carefully to chase it out of his backyard The bear is either frightened or it is not and there is no process leading up to its being totally frightened that is controlled by Peter . The only role Peter plays in this is that of the starting-point of a relationship of inclusion between himself and the bear. The inclusion does not result in cohesion. Linguistically, there is no process leading up to the moment at which inclusion takes place. The inclusion is there right from the beginning of the event. Peter 's actions do not preceed this inclusion but they are its source. An event is the inclusion and, if it occurs, the cohesion. The existence of an inclusion relation with psychological verbs is attested by the possibility of adverbials referring to the 'taking place' of the psychological event. (99a) expresses that the ongoings during the detesting event were intense. In (99b) it is expressed that the frightening came about easily. (99) a. John intensely detested those calculations20 b. John was easily frightened by those sales people yesterday
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
a
J a n van Voorst 1 03
The verbs do not imply cohesion, so the reading under which the psychological process is modified by adverbials like careji1lly is absent, as is demonstrated in (97). For some readers, a sentence like (97b) may be grammatical with an S- reading. This reading I discussed in Section 2.2. (97) a. *john clumsily/slowly detests derail work b. Derail work slowly frightened John The above data show that verbs like detest and Jr��hten belong to the aspectual class of achievement verbs. Thematic analyses distinguish these verbs. The present analysis (see (9X)) implies rhar they are similar and rhar there is no one to-one mapping between Pesersky's or Grimshaw's feature CAUSE2 1 and
THEMATIC fear frighten POWER ASPECT fear frighten
Bellerri/Rizzi Pesersky Grimshaw Bellerri/Rizzi Pesersky Grimshaw
( I ) experiencer- rheme (2) experiencer- target subject (3) experiencer (cause)- rheme (state) ( I ) theme-experiencer (2) cause-experiencer (3) experiencer (srate)- theme (state
inclusion + no cohesion inclusion + no cohesion
aspect, because achievements arc nor causative. In addition to these psycho logical verbs there are some action verbs rhar allow a psychological reading. Voorst ( I 99 2) shows that the constructions change aspect and go from being an acriviry to an achievement. (99) gives an example of an activity verb and its psychological readings.
(99) a. He struck me hard on the head
b. He struck me as really odd c. These lengthy quarrels struck me as very confusing
Thematically speaking this verb changes as in ( J Ooa); according to our model it changes as in ( 1 0ob). ( 1 00) a. strike (action) agent-patient strike (experiencer)-(psych) theme b. strike (action) inclusion + partial cohesion (activity) strike (psych) inclusion + no cohesion (achievement)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(98) Psychological Verbs
1 04
A Localise
Model for Event Semantics
( 1 o 1 ) a. He struck me slowly and carefully b. He *carefully/*slowly struck me as really odd c. These lengthy quarrels *carefully/slowly struck me as very confusing
The analysis I have given for psychological verbs is based on a well-defined model of event semantics. I have not added notions to incorporate these verbs but my description has remained as general and simple as possible and follows the general pattern of the semantics of transitive argument structures.22
7 CONCLUS ION The test with manner adverbials shows that, with activities and accomplish ments, the entity denoted by the subject has power over what is happening to the entity denoted by the direct object. In contrast, the sentence meaning does not express that this entity has power over what is hilppening to the entity denoted by the direct object in achievements and states or to that denoted by prepositional objects. My findings with respect t� the focus of manner adverbials make ir possible ro analyse aspect in a nontemporal localist way. Perfective aspect only occurs when the entity denoted by the subject has power over that denoted by the direct object. The fact that direct objects rather than prepositional objects influence the aspect of a construction can be explained this way. It can also be explained why the direct object of achievement verbs and states cannot influence the aspectual interpretation of the construction
concerned.
The above findings combined with the outcomes of other aspectual tests make it possible to formulate an empirically motivated localist model for the evenr as expressed in transitive constructions. The most important feature of this model is that it is based on linguistic data and that the primitives are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In ( 1ooa) the thematic strucmre changes in an unpredictable way. This is a cumbersome procedure for the grammar as the phenomenon of acriviries changing to psychological verbs is very productive, as is pointed out in Ruwer ( 1 972b), for French. Ruwet's remarks are also valid for English. Most verbs will end up with two argument structures having wildly divergent thematic roles. In the present model the semantic change is regular and expressable rather simply as in ( 1 oob). The only thing that changes is rhe cohesion of the entity including and the entity included, which is demonstrated in ( 1 o 1 ). The inclusion remains the same, so only a subpart of rhe semantics of the event changes. This subpart is responsible for the change in aspectual class from activity to achievement. The psychological uses of these verbs do nor allow rhe VP-reading of manner adverbs.
Jan van Voorst
l OS
Received 1 7- 1 2-9 1 Revised version received 1 4-4-92
JAN VAN VOORST Departement de lin)iuistique Universite du Quebec a Montreal Montreal H3C 3P8 Acknowledgements
Secrions 2. 1 an d 2.2 are based
NOTES 1
There are other issues relating to the exact inrerpretarion of these adverbs rhat we will nor delve inro. First, the adverb's focus can be limited to the subject NP; the process; or the object NP. (a) can mean
that the he is being careful 'with himself in order not to hurt himself; that the very process of loading is being done carefully; or rhat the b"xcs are being treated care fully.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
motivated by a number of linguistic tests involving adverbials. The primitives are not derived from mere interpretations of the situations described by the constructions. The event is analysed as the establishment of a relation of inclusion starting out from the entity denoted by the subject noun phrase. Different degrees of cohesion may result from this inclusion. These two primi tives together create a continuum that can be used to characterize the four tra ditional aspecmal verb classes spatially without reference to their temporal structure. This provides a deeper explanation for why the individuation of the direct object is narrowly related to aspect and why changes in individuation do not always influence the aspect of a construction. The localist model makes it possible to tackle a number of issues in the study of the semantics of transitive argument structures. It can be better understood what is the exact nature of the primitives defining events and, consequently, which primitives usually proposed to analyse events do not belong to the linguistic realm. Linguistic evidence shows that prepositional objects are very different from direct objects making inadequate thematic proposals that give them the same role; that the primitive of affectedness is ill conceived; and that the semantics of certain psychological verbs is not as irregular as may seem at first sight.
1 o6
A Localise Model for Event Semantics (a) He carefully loaded the boxes on to the truck Secondly, there is the problem that the sentence-initial position does not exclude a manner reading as in (b). This is not predicted by Jackendoffs rule.
2
(a) Gosh, you almost danced! (said of someone attempting to dance) I will not focus on the evaluative reading of the adverbial in rhis study. As is well known in the literamre, verbs can occur in more than one aspectual class. Various factors influence this and one could speak here of rype-coercion in the sense of Moens & Steedman ( 19XX). The best example is the coercion of an acriviry into an accomplishment by changing the individuation of the direct object. Another example is that a lot of
(a) They won the lottery (b) He won the chess game wi tlwu r too many problems Preposi tional phrases can express the tlattlrl' of rhe change of stare, as in (a)-(c) but their object is not the emiry that undergoes this change of state. In (a) Hou din i and in (b) the bux are the emities that undergo the change of stare. These sentences were poimed out to me by an anonymous reviewer. (a) Houdini escaped from the sack in 90 seconds (b)The bug crawled across the paper in 30 seconds () This study only treats transitive construc tions. It remains to be seen how our find ings can be generalized to semence pairs as in (a)-(b) and (c)-(d). (a) He clumsily pushed the cart (b) He clumsily pushed at him (c) He carefully chewed his meat (d) He carefully chewed on his meat It will have to be shown if these pairs can also be distinguished on the base of the notion of power or rhe primitives of inclusion and cohesion char will replace rhis notion in a later section. If the use of these primitives proves to be the right way to go. then we can expect to uncover that emities given by prepositional objects in imransirive constructions are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4
(b) Carefully/Slowly he loaded the boxes on to the truck Third, the adverbs do not always allow an S-reading when occurring in sentence initial position. The semantics pertaining to manner adverbials and to the holism-issue are a consequence of the more encompassing semantics of the event. These semantic issues are in no way special for the verbs occurring in the spray Ipaint alternation. The only thing special for these verbs is that they can occur in rwo different argument structures unlike many other verbs that only occur in one of the rwo structures. Why certain verbs and nor others allow alternating argument struc mres is discussed extensively in Pinker ( 1 9X9) in the context oflanguage acquisi tion. A lmost can also have an evaluative read ing in contrast to the event-oriented readings discussed here. For (24) the relevant interpretation under this reading is that the event does not quite qualify as one of stuffing and that a number of random activities were performed. This latter possibiliry is clearer in cases like (a).
achievement verbs seem to be able to take on an accomplishmellt reading. The accomplishment reading implies that the entiry given by the subject noun phrase exerts some control over the evem. (a) is most rypically an achievemem. One either wins a lottery or one does not. There is no process preceding rhis winning. (b) is more like an accomplish mem with a process leading up to a final change of stare. The semence can imply that the chess master comrols the process of winning through his well-planned moves.
Jan van Voorst 107
(a) What Hill did to rhe truck was load it with books (b) ?What Hill did to rhe truck was load the books on ro i r (c) What Bill did to the books was load rhem on to rhe rruck (d) •what Bill did ro rhe books was load the truck with them The author tries ro explain the above grammaricality differt·nces by staring rhar rhe direct object is more directly affected by Bill's actions implying rhat the pre positional object is less directly affected. This is not confirmed by our findings that prepositional objects can nor be under the scope of manner adverbials. One could consider my localist model a further extension and refinemellt of the action ner. X In most studies on aspect the emphasis is on the change of state rhar comes about as a result of the event. This is rhe end reading in (70). Under our approach the begin-reading plays an equally important role in the definition of aspect. The explanatory value of such an approach has been defended in van Voorst ( • <JXX). This approach also explains why rhere is an additional dimension to the phenometHlll tlf the imperfective paradox. When accomplishmenrs are used in the progres siw form not only is rhe change of srare
tlnalizing the evenr not implied. bur also rhe time ar which the evenr begins. I will discuss this phenomenon in Section 5 · It is oftt·n overlooked tn the literature. alrhough Vet's ( t l)Xo) treannenr of the french impartait captures it. l) The distinction seems to have found its way back inro modern North American studies of argumenr structures in Pinker ( t l)X<J). To solve Baker's paradox in lan guage acquisition Pinker imroduces the notion of grammatically relevant seman tics without being aware of Hjelmslev's distinction and the ideas many other grammarians have had on rhis topic in an earlier era. Baker's paradox states thar argument structure alternations cannot be learned correctly. if rhe child has knowledge only of irs structural aspects. The absence cll" negarive evidence resulrs in rhe child being made to predict if a certain verb occurs or does nor occur in a certain alternation. One would expect many ungrammatical utterances because of rhis, which is nor rhe case in reality. Using grammatically relevant semantic clues the child can find its way out. These semantic clues could be comparable ro the semantic primitives rhar are arrived ar using Hjelmslcv's philosophy by analys ing language dara semantically in close relation to various grammatically rele vanr subclasses. 1 0 Rappaport, Levin &: Laughren ( t <JXX) are unaware of rhe influt·nce of individuation on rht· inrerpreration of rhese setltenct·s alrogether and ignon: data like (72.b) wirh the entity that is to be located individua ted. 1 1 A differenr inrerpretation of perception verbs like rcccil'c can be found in Pinker ( l l)X<J). Using rhe force-dynamic analysis of Tal my ( l l)XX), Pinker analyses the subjects of these wrbs as an enabling cause. This notion does nor rie in directly wirh our own localisr analysis as primi tives like cause do nor figure in it. In addi tion. aspectually speaking. the use of this primitive in rhese cases makes lirrlc senst·.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
not within the realm of power of the subject either. The difference berween the sentences in each pair then would be that in (a) the entity given by the subject can influence what is happening ro the cart but nor in (b); in (c) this entiry can influence what is happening ro the bone, bur not in (d). Again. we can use the scope of adverbs from the car�fid -class to show if these assumptions are correct. 7 Jackendoff ( 1
I
o!l
A Localist
Model for Event Semantics
'
handled at a level of temporal analysis. I will discuss the separation of the temporal and the nontemporal aspects of events in Section 5 of this paper. I (, At a far more gei!eral, metaphorical level. there is a strong resemblance between my own lexical semantic analysis and Lakoffs ( I 990) analysis of the metaphor of reason. Lakoff proposes that event structure can be understood metaphorically in terms of space, motion, and force. In the analysis I am proposing, space is the spatial relation between emities figuring in an event or stare; motion is implied when. starring out from the eiHity given by the subject. an inclusion or cohesion relationship is being established. Force is what I S necessary to make rhe inclusion or cohesion come about. I 7 If a formal semamic analysis of the present localisr model is possible, ir will probably be able to bring out these facts quite elegantly making unnecessary predicates like Krifka's [ [SUKJI and [[SIM]). The punctuality or durarivity of events will have to play a role at a dif fereiH level in such an analysis. I X Following Guillaume ( I 92.9), Marrin (I 97 I ) proposes that the auxiliary <JI!Oir expresses accomplishmem (i.e. ('accom pli) and rhar verbs can be perfective and imperft·crive. Under the present approach perfectivity is non-temporal. The sen�an rics expressed through the auxiliary is temporal. ThL· distinction made in Fried rich ( I 97+) between ancieiH Greek aspect as centred around durativity and Slavic aspect as centred around completion can be re-analysed along similar lines. Anciem Greek grammaticalizes temporal features relating to evems and stares: Slavic grammaticalizes locative features relating to the internal structure of events. I 9 It was poimed out to me by an anony mous reviewer that tl.e present model has little ro say about phenomena like the progressive. I think that what is ar issue here is not this question but a more
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Perception and reception verbs are achievements under the present analysis. They only have the begin-reading with adverbs like in X minutl's and almost . There is no change of state that finalizes the event as in the case with accomplish ments. The only 'change of state' that takes plan· is the coming into existence of the event. An 'enabling cause' then is better defined as an event initiator or the emity that establishes a relation of inclu sion under the present localist analysis. I 2. Hjdmslev ( 11:)3 s: 12.9): 'Par coherence nous comprenom le fair general d'erre lit par u ne connexion relativement intimt: a un autre objet. I 3 The treatment of these adverbs is based on suggestions of Denis Bouchard. Valerie Lanctot and Yannick Morin, members of the pn�ecr Theorie des para metres, variations et argument nuls at the Universite du Quebec a Montreal. 1 4 Voorst ( I 9XX) characterizes the subject noun phrase of nonstative transitive constructions as the event creator. The present study makes this notion more specific and actually tells what this event creator 'does' (i.e. it brings into existence a relation of inclusion and, in a subset of cases, it also establishes cohesion). The grammatical relevance of the notion of event creator is discussed in depth in the above study. The notion of object of termination in that study coincides with the present notion of complete cohesion. The former notion only covers the direct object of accomplishment constructions. The present model allows us to integrate direct objects of achievemems and activities as well. I 5 An anonymous reader pointed out that there is some inconsistency in tht· model ! am proposing as it uses cerrain notions or tests that focus on the temporal aspects of events. These notions or tests will have to be purged from the localist analysis. It means that phenomena like those brought to the foreground by adverbials like j<'>r or iu X miuutl's may have to be
Jan van Voorst
In (a), it is not an event of stinking that is taking place in a terrible way but the smell is terrible. Degree adverbs like int!'llsciy can be grouped with our class of adverbs like with '"' problems or cj};>rtlessly . for rhar rt·as
(b) *He intensely had a headache (c) *They i ntensely lacked a lot of nwtiva non 2. 1
2. 2
Tht· discussion regarding the prinunvt• CAUSE is made difficult by the fact that the notion is not well defined and is often used as a labelling device in lexical or conceptual approaches. A proper in depth semantic analysis of tht· contenrs and the lexical-semantic phenomena covered by this primitive is lacking. This phenomenon seems to exrend rn most verbs implying cohesion. If we assume rhat a psychological reading as well as the 'cogniriw' reading in (a) general! y involves lack of cohesion, rhen we can infer that most verbs can swirch back and ftlrth berween cohesion and inclusion. (a) Your paper really hclped/sri mularcd mt• (b) They really helped/stimulated l l l t' wirh rheir rt·marks This means thar most verbs can occur in constructiom that imply the initiator's conrrol over what is happening to rhe entiry denoted by tht· direct object but also in ones thar i mpl y a tota l lack of this control. In rhis conrext. it can be said that whether a verb is agentive or nor dt depend that much on the verb itself bur rarher on tht· nature of rhe evenr structure in which ir happens to occur.
R E F E RE N C E S Emmon ( II) X(>), The algebra o fevents',
pretation et connaissance du monde a propos du verbe PLANTER', in C. Rohrer l3elletti, Adriana & Luigi Rizzi ( 1 9!!!!), 'Psych & N. Ruwet (eds), Actes du Coll.>que Franco A llemand de grammain· transjormationelle , verbs and e-theory', Natural Lan,J!uaxc and Niemeyer, Tiibingen. Lit�J!uistic Tl11wy, 6, 2.9 1-3 )2.. l3innick, Robert I. ( 1 99 1 ), Time and tlte V£•rb: A Carlson, Gregory N. ( 1 9Xo). R£ji·rellce to Kinds Guide to Tense and Aspect , Oxford Uni in English , Garland Publishing, Inc., New York!London. versiry Press (Canada), Oxford. l3oons,Jean-Paul ( 1 974), 'Acceptabilite, inter- Chomsky, Noam ( 19) s). The L<�J!ical Structure Bach.
Lin�uistics and Pltilos<>plty , 9,
) - 1 6.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
.w
fundamental one. It cannot be denied that there are temporal phenomena relating to events as well as spatial ones. What my analysis indicares is rhat the rwo musr be distinguished. The more fundamental question that has not been addressed in this study is how rhese rwo levels of analysis imeract. A problem relating to the progressive would be if rhere is a correspondence ofsume sort berween this form and the localist classification. If we have to believe MufWene ( 1 9X.p) this cannor be the case where it concerns the possibiliry of a verb to occur in the progressive. Another very importanr problem i.� w see if the i m pe rfec tive paradox involves a change in the sense of the localist model or if it involves a semantic change of a differem order. Intensely is a degree adverb. Degree adverbs are also possible wirh srares. The differenet· importanr to the presenr discussion lies in what the adverb modi fies. In rhis case rhis is rhe process of the dctcstit�� event but wirh states like stillk in� , in (a), the degree of the state is speci fied. (a) These dogs stink terribly
1 09
1 10 A Localist Model for Event Semantics
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of Linguistic Theory ( 1 975 reprint), Uni Hjelmslev, Louis ( 1 943). Prolegomenes a une versity of Chicago Press, Chicago. tlzeorie du langage (French edn of 1 9X4), Les Coppieters, Rene (1 982), 'Descriptions and Editions de Minuit, Paris. attitudes: "The problem of reference to Hopper, Paul J. & Thompson, Sandra A. individuals" ', Studies in Languaxe , 6, ( 1 9lio), Transitivity i n grammar and discourse', Language ' s6. 25 1-99· 1-22. Dowry, David ( 1 979), Word Meaning and Jackendoff, Ray S. (1 972), Semantics Inta Montague Grammar: The Semantics of Verbs pretation in Generative Grammar, MIT and Times in Generative Semantics and in Press, Cambridge, Mass. Jackendoff. Ray S. ( 1 990), Semantic Structures , Montague's PTQ , Reidel, Dordrechr. MIT Press, Cambridge. Ernst, Thomas Boyden ( 1 9g5), Towards 011 Integrated Theory of Adverbial Position in Jeffries, L. & Willis, P. ( 1 984), 'A rerum to the spray/paint issue', journal
Jan van Yoorst 1 1 1
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Mufwene. Salikoko S. { 1 9/L�a), Stativity and the Ryle, Gilbert ( 1949), The Concept oJMind (edn Pnl)!ressive , Indiana Linguistics Club, of 1 97 3), Penguin, Harmondsworrh. Bloomington. Tal my, Leonard ( 1 9X8), 'Force dynamics in MufWene, Salikoko S. ( 1 9IL�b), 'The count/ language and cognition', Cognitive Science , mass distinction in the English lexicon', in u, 49- 100. D. Testen et a/. (eds), Papers on Lexical Talmy, Leonard (1 976), 'Semantic causative Semantics, Chicago Linguistics Society. types', in Masayoshi Shibatani, Syntax and Chicago, 200-22. Semantics , 6, Academic Press, New York. Nikanne, Urpo ( 1 990), Zones and Tiers: A Tenny, Carol Lee (1 989). The Roles ofInternal Study of Thematic Stmcture , Studie Fennica A�l,'uments: Measuring Out Er,ents , MIT 3 s Linguistica, Helsinki, Suomalaisen Centre for Cognitive Science, Cambridge, Kirjallisuuden Seura. Mass. Pesetsky, David ( 1 990), 'Experiencer predi Yendler, Zeno ( 1 967). Linguistics in Philo cates and universal alignmem principles', sophy, Cornell U niversity Press, Ithaca. MS, MIT, Cambridge, Mass. Yendler, Zeno (1 976), 'Causal relations', Pinker, Steven ( 1 9X9), Learnability and C�l,' journal ofPhilosophy, 64, 704- 1 3· nition: The Acquisition ofAr)!ymmt Structure , Yerkuyl, Henk J. ( 1 972), On the Compositional MIT Press, Cambridge, Mass. Nature oJtht• Aspects, Reidel, Dordrecht. Pustejovsky, James ( 1 9XH), 'Event Semantic Yerkuyl, Henk J. ( 1 989). 'Aspecmal classes Structure', unpublished paper, Brandeis and aspectual composition', Linguistics and University. Philosophy , 12, 39-94· Rappaport, Maika, Levin, Beth & Laughren, Yet, Co (1 980), Temps. aspect et adverbes de Mary ( 1 9XX), 'Niveaux de representation temps en fran�ais contemporain. Essai de lexicale', Lexique, 7, 1 3-32. semantique formelle Geneve, Librairie Reyle, Uwe (1 9X7). Zeit und Aspekt bei der Droz. Verabeitung natiirlicher Sprache, Lilog Yoorst, Jan G. van (1 9HH), Evellt Structure , Report, IBM Deutschland, Stuttgart. John Bet�amins Inc., Amsterdam. Ruwet, Nicolas ( 1972a), Theorie sytaxique et Yoorst, Jan G. van (1 992), The aspecrual syntaxe duJran(ais , Editions du Seuil, Paris. semantics of psychological verbs', Unxuis Ruwer, Nicolas ( l l)72b), 'A propos d'une tics and Philosop/,y , 15, 65-92. classe de verbes "psychologiques" ', in Wiillner, Frantz ( 1 X27). Die Bedeutun,l,' dcr Ruwet ( 1 972a). sprachlichm Casus rmd M,Jdi, Eil1 Vcrsuch , Ruwer, Nicolas ( 1 972c), 'Les constructions Miinster, 1 X27. facrives', in Ruwet {I <J72a).