JOURNAL
OF
SEMANTICS
AN INTERNATIONAL j OURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE ...
6 downloads
480 Views
19MB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
JOURNAL
OF
SEMANTICS
AN INTERNATIONAL j OURNAL FOR THE INTERDISCIPLINARY STUDY OF THE SEMANTICS OF NATURAL LANGUAGE MAN AGING E J) I TOR: REVIEW EDITOR:
EDITORIAL BOAIU):
PETER BoscH ( IBM Gamany) IJARTGEURTs(JI3M Gamany) PETER BoscH ( IBM Gamany)
SiMON C.GARROD (Univ. of Glasg ow)
BARTGEURTS (JUM Germany)
PAUL HoPPER (Carnegie Mellon Univ .. Pimburgh) LAURENCE
H.
HoRN (Yale University)
STEPHEN lsARD (Univ. of Edinburgh) HANS KAMP (Univ. of Stuttgart) LEO G. M. NoORDMANN (Univ. ofTilburg) Ron A. VAN DER SANDT (Univ. of Nijmcgcn) P1ETER A.M . SEUREN (Univ. ofNijmcgen)
CONSULTING EDITORS : It BARTSCH (U niv. of A ms te r d am)
SJR JoHN LYONS ( U niv. of Cambridge)
J. VAN BENTHEM ( U n iv. of Amstn dam)
D.
S. IJREE (Univ. of Manchester)
G.
BROWN (Univ. of Cambridge)
H. H.
E.
W. MARSLEN-Wn.SON (MRC. Cambridge) J.
A. J. SANFORD (Univ. of Glasgow)
H.
ScHNELLE (Huhr Uni v. Uochum)
A. VON Sn:cHow (Univ. of Konstanz)
G. FAUCONNIER (U niv. of California, San I)iego )
M. STEEDMAN (Univ. of Pennsylvania)
D. VANDERVEKEN (Univ. of Quebec)
J. HonBS (SRI, Menlo Park)
D. ISRAEL (SRI, Menlo Park) P. N.JoHNSON-LAJRn (MI{C, Cambridge) E. L. KEENAN (Univ. of Calif(mJia.
Z.
VENDLER (Univ.
of California, San Diego)
B. L. WEBBER (Univ. of Pcnnyslvania)
Y. WILKS
Los Ange ll' s)
(
of Chicago)
ll. RICHARDS ( Imperia l Colle ge, London)
H. CLARK ( Stan f(> rd University) 0. DAHL (Univ. of Stockho lm) H.-J. EJKMEYER (Un iv. of Biele feld)
E. LANG U niv.
D. McCAWLEY (Univ.
H. REICHGELT ( Univ. of Nottingham)
ilREKLE (Univ. of Regensburg)
(New Mexico State Univ.,
Las Cruces)
D. Wn.soN (Univ.
of Wupper tal )
Collc·gc, London).
E J) IT OR I A L AD J)R E SS: Journal of Semantics. IBM Germany Scientific Centre. IWBS 7000-7). Postfach XooXXo, D-7000 Stuttgart Xo. Germany. Phone: �y-71 1- My)-SW Telcfax: �y-71 1 My)-jOo.lliTNET: bosch@ds.,Jilog.
(
)
(
)
New Subscribers to the journal of Semantics should apply to the Journals Subscription Department, Oxford University Press. Pinkhill House. Southfield Road. Eynsham. OXX 1 . For further information see the inside back cover.
jj
Volumes 1-6 arc available from foris Publications Holland. PO Box soy. Netherlands.
© Copyright by NIS
3300
Am Dordrecht. The
Foundation
All rights rest·rved; no pan of this publication may
be
rt· pr odun·d . srort·d in a rt:tricval sysrc.·m. or rransmi ned in any
form or by any llll'ans. elt·crronic. mech;mical. plwr ocopyi n g. recording. or orherwiSt· wirhour t•irher rhl· prior wrirren
permission of thl· Publisht.·rs, or a lict•nce pl·rmining resrricrnl copying issul·d in rhe UK by the Copyright Lin· nsing
Agency Lrd. l)O Torr l·n ham Coun Hoad. London W 1 P 9HE. or Congress Srn.·cr. Salem. Mass OH)70.
Tht· Journal ofSemanrics is publislwd yuarrcrly hy Oxford p•:nding at N,·wark. N,·w Jersey. JSSN 011>7-5 1 JJ.
POSTMASTER send addrt·s�
in
rhc USA by
University
corrl·ctions to Tht• Journal of St.·manrics.
1 )O, Nc:wark International Airporr. Nl·wark. New Jcrst·y 071 1..._, U SA .
dw
Copyright Clearance Ce mcr. 2.7
Press. Subscription is $l)O pt•r year. St·ctllld class c/o Virgi n Mailing and Disrriburion. Building
JOURNAL OF SEMANTICS Volume 8 Number 4
CONTENTS
]. RIJKHOFF Nominal Aspect ]. P. McDowELL Quasi -assertion
311
C. ScHWARZE Concept Types and Parts of Speech: the Lexicon of Space to French
333
M. HERWEG
Two Classical Approaches to Aspect
Journal ofSemanlics 8: 291-309
© N.l.S. Foundation (1991)
Nominal Aspect J. RIJKHOFF University ofAmsterdam
Abstract
I NT R O D U C T I O N1 In a general way the notion 'aspect' can be defined as the way in which a property or relation is represented in some dimension. Two kinds of aspect can be distinguished: verbal and nominal aspect. The study of verbal aspect has a long tradition, but nominal aspect has only been introduced recently, at least in the sense in which it is used here (Rijkhoff 1 989b, 1990a, 1 990b).2 After a brief look at the more familiar verbal aspects, each of the nominal aspects is discussed in some detail. Then the relevance of nominal aspect will be considered in connection with (i) certain 'number markers' (which will be analysed as nominal aspect markers below), (ii) noun-incorporation, and (iii) predicate nouns. I
A S PE C T
I.I Verbal aspect
Properties and relations in the temporal dimension, which are designated by verbs (sit, walk, read, etc.), can be characterized in terms of the typically temporal features BEGINNING and ENDING. This gives us four ways of representing such properties and relations, i.e. four verbal aspects (Figure I ) . Although at least within some of these aspects finer subdivisions can be made (such as ±progressive in the imperfective and ±momentaneous in the perfective; see Comrie I 976), and despite the fact that some verbs are inherently
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In this article I argue that besides verbal aspect, which concerns the way a property or relation is represented in the temporal dimension, there is also nominal aspect, which relates to the way a property is represented in the spatial dimension. I will contend that certain elements, which are often believed to be number markers, are in fact nominal aspect markers. Evidence to support this will be taken from several genetically unrelated languages. Additionally it is suggested that nominal aspect plays a role in connection with incorporated and predicate nouns.
292
Nomi nal Aspccr
TIME ending
unmarked ending
marked
beginning
beginning
IMPERFECTIVE
INGRESSIVE
EGRESSIVE
PERFECTIVE
marked
u nmarked
Figure 1
Verbal aspects
1 985), we may say that generally any property or relation designated by a verb can be represented in any of these four ways. Notice that these aspectual distinctions may not only be expressed grammatically, i.e. by means of inflectional morphology, but also by lexical elements or in a periphrastic con struction. For a more extensive discussion of verbal aspect I refer to Verkuyl ( 1 972), Comrie ( 1 976), Lyons ( 1 977: ch. 1 5.6), Hopper (ed.) ( 1 982), Tedeschi & Zaenen (eds) ( 1 98 1 ), de Groot & Tommola (eds) ( 1 984), Chung & Timberlake ( 1 98s).
1 .2
Nominal aspect
Nominal aspect is defined as the way in which a property, as designated by a noun, is represented in the spatial dimension with respect to the features SHAPE a.nd STRUCTURE. Nominal aspect is usually coded covertly, i.e. it is a lexical feature of the noun as it occurs in the lexicon. For example, there is nothing that explicitly marks the English noun car as a count noun or
water
as a
mass noun. The fact that generally nouns (but not verbs) are inherently coded for some aspectual meaning may be related to the idea that spatial orientation is primary in human cognition (Anderson
1980:
ch.
1 2; Lyons 1 977= 7 1 8 £).
1 973; Traugott 1 978; Lakoff & Johnson
This is also reflected in the fact that in some
languages the spatial origin of verbal aspect and tense markers is fairly obvious (Comrie
1 976: 98 £, 1 29-30; 1 98 s: I s).
In its turn
temporal elements may
provide the material from which modal elements may develop (Fleischman
1 989)· At this point it may be useful to emphasize that referents of noun phrases (NPs) are not entities in the external physical world; hence we speak of 'intended referent' rather than 'actual referent' (McCawley
1 968: 1 3 8;
Dik
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
coded for the aspectual features ENDING (telic verbs such as 'recover') or BEGINNING +ENDING (momentaneous verbs such as 'hit'; c£ de Groot
J. Rijkhoff 293 I989a: I I I
f).
This may be demonstrated by the fact that we can discuss
referents that cannot be perceived in the immediate extra-linguistic context (Lake Titicaca, the Armada) or never existed (unicorns). Referents of NPs are mental entities, which are construed on the basis of linguistic material. Hence properties of referents of linguistic expressions are due to properties of the elements on the basis of which they are construed (quantifiers, nouns, adjectives, etc.). In other words, there may be discrepancies between referents of NPs and their real-world equivalents (if they exist).3 In this view a referent may be regarded as a collection of facts about participants and props as they (i.e. the facts) are mentioned in ongoing discourse and which are stored in the minds of the speaker and hearer. Using a computer
participants and props that figure in a particular discourse. Every time a new participant or prop is introduced a new record is created (that is, another referent is a construed), which may be updated and which helps the speech participants to keep track of the referent (cf Heim I 982; Rijkhoff I989a). The records in the temporary referent file must have access to various types of long term knowledge. For instance, to the knowledge about the situations, events, etc. that some referent was involved in earlier (non-linguistic episodic knowledge; Dik I989b: I I-12). Properties in the spatial dimension can be characterized in terms of the spatial features SHAPE and STRUCTURE. If a property as designated by a noun is marked as having STRUCTURE, this means that the referent that was construed on the basis of such a noun is characterized as being divisible. That is, if the space for which such a property obtains were to be divided, that particular property would still obtain. If a property as designated by a noun is marked as having SHAPE, this means that the referent that was construed on the basis of this noun is characterized as having a definite outline. Thus, in the spatial dimension, too, there are in principle four ways to represent a property. In other words, there are also four nominal aspects (Figure 2). Thus, basically the same property may be represented in different ways. One finds, for example, that one language employs concept nouns where another language uses individual nouns in connection with the same object in the real world. Mandarin Chinese
shu
'book/books', for example, can be regarded as a concept noun (see below). This means, among other things, that it cannot be pluralized and that the referent of the NP headed by this noun may involve one or more individuals (Li & Thompson I98 I: I I; see section I .2.1 below). The English noun
book,
on the
other hand, denotes a singular, spatially bounded entity (i.e. an individual). This implies that it must be pluralized when the speaker wants to indicate that he is referring to more than one book (see section 1.2.2).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
database system as a metaphor, a referent can be compared with an indexed record in a special referent file. This temporary file keeps records on all
294
Nominal Aspect
SPACE
structure unmarked
structure marked
shape unmarked
CONCEPTUAL
MASS
shape marked
INDIVIDUAL
COLLECTIVE
Figure 2
Nominal aspects
coded covertly, ] (L] (-2]F
=
=
MUST(P) MAY(P)
where (0] adds the propositional content condition that the propositional content is stative and (L] adds the preparatory condition that the speaker does not know that, but only infers or deduces that, the propositional content is true.
4 T R U T H C O N D I T I O N S F O R Q U A S I - A S S E RT I O N We have seen above that the truth conditions for a simple assertion F(P) are that P E w • where w • is the actual world. w • can be characterized by a set of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3·3
J. P. McDowell
325
propositions P; which are true in w •. Informally this set of propositions can be viewed as a complete description of the actual world at a time t .
(25)
w
•
=
{P;/P; i s true in
w
•
}
=
•
•
·
·
where p '; is the i th proposition in the set which describes and determines the world w 1 and p '; is true in w 1.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Clearly the same truth conditions cannot hold for quasi-assertion. It is of course possible that where P is quasi-asserted, P E w •. That is, it is always possible to check on the truth of P in w • and find that P is in fact true in the actual world. But the propositional content condition of an illocutionary act F(P) requires that the speaker acknowledge and accept the truth conditions imposed on P by F. And when a speaker quasi-asserts P by performing F(P), he is explicitly and overtly signalling that he does not know that P is true in w • but only infers or deduces that P is true. Furthermore, he believes that P is true in w • only with the degree of strength that is justified by the strength of the evidence on which he has inferred P. Now there is a strong intuition in the literature on the English modals that epistemic modal sentences say something about the possibility or necessity that the proposition expressed in such sentences is true in the actual world (c£ Boyd & Thorne r969 and Steedman I 977). But as it turns out, logical necessity (o) is too strong for sentences with must and logical possibility () is too weak for sentences with may . To elaborate, we cannot translate a strong quasi-assertion (Must(P)) as oP because, as we have seen above in the discussion on the epistemic paradox, oP(P is true in all logically possible worlds) is stronger than P (P is true in the actual world), but MUST(P) is weaker than P. Similarly P (P is true in at least one logically possible world) makes far too weak and nonspecific a prediction about the interpretation of MAY(P). The truth conditions are simply wrong. What we need is to retain the quantificational properties of o (correspond ing to V) and (corresponding to 3) with respect to some subset of the set of all logically possible worlds. The trick is to restrict the domain of quantification to a subset of possible worlds which is related to rhe actual world systematically via the inferential or deductive process. We designate this restricted domain w •, the set of epistemically accessible worlds (where accessibility is to be understood in the manner of Lewis I973 ). We construct w•, the set of epistemically accessible worlds which can be deduced from w •, by imposing an ordering of the sort proposed by Kratzer ( 198 1 ) on W the set of all logically possible worlds, as follows. We begin by repeating our assumption that a possible world w is a set of propositions which are true in it. (2 6) W I {p i I • P I Z• ., P Ii• ., P I n}
326 Quasi-assertion
The actual world w* is also a set of propositions true at utterance time:
(27) w * = {p "' 1, p *2,
.
.
•
, p \ . . ., p "'n)
For any speaker s, s knows only some of the propositions tht are true in w*. P* is the set of propositions which are true in w• and known by s, such that p *i) . Then Q "' is the set of propositions which P * � w•. Suppose p • = {p *1 arc true in w* but not known by s, here Q * � w * and Q * = {p *j . . . p *n)· Together P* and Q * completely determine w *, p• 11 Q * = o, P* u Q * - w •. While Q • is a completely determined consistent1 4 set of propositions, so far as s knows, Q * could be any set of propositions X; such that P* 11 X; = o and p• u X; is a consistent set, i.e. X; is compatible 1 5 with P* and nonintersecting. There is a very large if not infinite set of sets of propositions X; which meets these conditions. •
•
•
A world W; = P * 11 X;, is a possible state of affairs for w * so far as s knows. That is, W; may be identical to w • so far as s knows. P*, the set of propositions known by the speaker s to be true in w * , imposes an ordering on W ( W = 11 W; in the manner ofK.ratzer ( I 98 I ). The ordering ranks each W; E W in accordance with its probability of being the actual world in the eyes of the speaker. W* is the set of all and only those members of W which meet the probability test in the eyes of s. We call w• the set of all such Wi• (29) w• = ( W;I W; = p• u X; and W; is highly probable for s) where P* and X; are as defined above.
W* is the set of epistemically accessible worlds. 1 6 Truth conditions for strong quasi-assertions MUST(p) and weak quasi-assertions MAY(P) can then be stated using standard quantifiers operating over w•. (3o) MUST(P) = I iff \fw E W*.., P E w P is true in every possible world that is episternically accessible to s MAY(P) = I iff 3w E w·•. P E w P is true in at least one possible world that is episternically accessible to s An informal interpretation of these truth conditions is this. When a speaker strongly quasi-asserts P, he is saying, 'I am reasonably confident that P is true in the actual world because P is true in every possible state of affairs that could reasonably be the actual world and which is, at the same time, highly probable according to the evidence available to me.' Similarly, when a speaker weakly quasi-asserts P, he is saying, 'I am able to accept that P is true in the actual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(28) X 1 : P* 11 X1 = o, P* u X1 is consistent X 2: P* 11 X 2 - o, P* u X2 is consistent Xn: P* 11 Xn = o, P* u Xn is consistent
J. P. McDowell
3 27
s
C O N CL U S I O N
We return briefly now to some points raised above. We suggest that the felicity conditions on quasi-assertion and the truth conditions on the propositional content imposed by the illocutionary force or quasi-assertion provide some diagnostics for determining what is or is not an epistemic modal sentence and what arc the subtle differences between true epistemics and other utterances which have an epistemic flavor. We will not go into the details here because it would take us far afield from the topic of this paper. But we will sketch the lines of argument and leave the details to other expositions (and c£ McDowell 1987). One point is that sentences with will and can are only epistemic to the extent that their truth conditions involve sets of possible worlds which are related inferentially to the actual world w *. But these sets of possible worlds are not related to w • in the same way as W* the set of epistemically accessible worlds. Recall that every w in W* includes P* the set of propositions that the speaker knows to be true in the actual world w •. This is clearly not the case for FUT, the set of future worlds which is the domain of prediction (will ). FUT will intersect with w• minimally to the extent of timeless truths (the laws of physics, etc.) and facts about the past. All the other propositions which determine a
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
world because P is true in at least one possible state of affairs that could reasonably be the actual world and which is, at the same time, highly probable according to the evidence available to me.' Notice that w • e W* and P is true iff p e w *. It is important at this point to distinguish between truth conditions for quasi-assertions and felicity conditions. For example, under the truth condi tions presented above, both MUST(P) and MAY(P) are true if P e W*P This agrees with intuition. If I 'know' that Clark Kent is Superman (i.e. I am committed to the truth of the proposition expressed by the sentence Clark Kent is Superman ), then it follows that Clark Kent must be Superman (whatever is cannot possibly be otherwise) and that Clark Kent may be Superman (whatever is is possible). Whether I choose to fully assert, strongly quasi-assert or weakly quasi-assert some proposition P, however, depends on a host of social and practical factors which have nothing to do with truth conditions. In ordinary conversation we have to assume that speakers are being cooperative in the Gricean sense (Grice 1975) and that they agree to the truth conditions for the propositional content of their utterances that are appropriate to the illocu tionary act they are performing with their utterances. Indeed speech act theory explicitly includes the satisfaction of propositional content conditions (among these truth conditions) as a condition for satisfying felicity conditions (see section 2, above).
J2X
Quasi-assertion
(3 I) a. The sun should come up right over that tree. b. ??The sun must come up right over that tree. JOYCE P. MCDOWELL Intelligent Text Processing, Inc. tJt o Montana Av. Suite zot Santa Monica, CA 904 03 USA
N O TE S 1 Throughout this paper we follow the exposition of speech act theory to be found in Searle & Vanderveken ( 1 98 5). 2 In most cases in English there is a performative verb which can be used to perform some particular illocutionary act. In many cases there is also a noun which names the act. For example, ifl say Walker is a spy, I am performing an act which I might also perform by saying I
hereby assert that Walker is a spy and which goes by the name of assertion. Assertion is the primitive act which makes the asser tive point, and sentences like I hereby assert that Walker is a spy are redundant because Walker is a spy is sufficient to make the assertive point. That is, any declarative sentence is implicitly or by default an assertion. Other assertive performatives are more felicitous, particularly the more
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
world in FUT will have to be inferred by considering what is a reasonably orderly extension of those members of P* which are not timeless truths or facts about the past. Thus, for every world w E FUT, only some of the propositions in P* are true in w. The domain of possibiliry expressed by ca n , call it K, on the other hand, is a set of possible worlds which is compatible with the speaker's knowledge of the actual world, in other words, compatible with P*. Each world in K is just P* plus some other proposition p , K - (w: P* v (q} where q is a proposition). One of these worlds includes the proposition p expressed by the sentence containing ca n . This domain is epistemic just to the extent that any domain is epistemic which contains P*. Thus, for every world w E K, every proposition P* is true in w. Sentences with should are best thought of as reports of norms Uust like sentences with deontic must are reports of commands). Norms differ minimally from commands in that the issuer of the norm has no authority in the relevant domain to command the expressed proposition. What sometimes gives these sentences an epistemic flavor is that norms include not only moral precepts but knowledge of what is or is not usually the case, what Kratzer (I 98 I) calls a normative ordering source. Once again this involves a generalization on P *. This latter point can be easily seen by comparing the perfectly natural ( 3 I a) with the anomalous (3 I b).
J. P. McDowell
indicate a starive interval because the action described by a progressive precedes speech rime, carries on through speech rime, and continues afterward. Similarly, a perfective such as John has read a book indicates a srarive interval because ir describes a properry of an individual in the sense of Carlson ( 1 977). Properties of individuals hold before, during, and after speech rime. I o There is some cross-linguistic evidence for this from Chinese. In Chinese the episremic operator is a pure adverb (i). (i) Zhangsan yiding shi ge yisheng Zhangsan certainly be MEASURE doctor Zhangsan must be a doctor Starive verbs in Chinese, like shi 'be' and zhidao 'know' can receive present tense interpretacion and can co-occur with the episremic operator (i, ii). (ii) Zhangsan yiding zhidao zenmo zou. Zhangsan certainly know how go. Zhangsan must know how to get (there). Nonstarive verbs cannot gee present tense interpretation without the presence of a modal or ocher verb. That is, nonstarive verbs m Chinese cannot recetve rhe habitual or characteristic interpretation. These verbs are also impossible with the episremic operator (iii). (iii) *Zhangsan yiding zi chezi. Zhangsan certainly wash car. Zhangsan must was the car (regu larly). Zhangsan yiding yao xi chezi. Zhangsan certainly will wash car. Zhangsan will surely wash rhe car. anonymous reviewer suggested these examples. 12 S & V actually scare char rhe entire structure of illocurionary logic which rhey construct depends on a somewhat idealized view of rhe verbs they discuss (p. 1 8 2).
1 1 An
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
rhey differ from primitive assertion, for example, I claim that Walker is a spy, I deny that Walker is a spy, I suggest that Walker is a spy, ere. There is no English performarive verb which can be used to perform quasi assertion and, except for chis label char we have given ir, there is no noun which names rhe ace. 4 We make rhe common assumption char our example sentences represent tokens of urrerances in actual conversation. S c£ Fraser ( 1 97 5) and see below. 6 Srricdy speaking, the nonmodalized ver sion of The Maori must have come from Tahiti is The Maori have comefrom Tahiti. However, chis sentence seems a lirde odd because of rhe present relevance supplied by rhe perfective. Contrast TheJonesfamily comefrom Boston wirh TheJonesfamily have come from Boston . The first sentence reports rhe origins of rhe Jones family who may or may nor be present ar rhe rime of utterance, while rhe second reports rhe departure point of rhe recendy arrived and probably present Jones family. Since there is no past tense form of must rhe perfective must have stands m for borh simple past and perfective in the modalized version. The Jonesfamily must have comefrom Boston can either quasi-assert char rhe Jones family has irs origins in Boston or char rhe Jones family is recendy arrived from Boston. Such a contrast is nor actually possible with respecr to the Maori because, although it 1s possible to report rhe origins of rhe Maori, it is not possible for them to be recendy arrived. 7 See Barnes (1 984) for a language with a rich evidential morphology. 8 This is, of course quite different from saying, 'Whatever is could not possibly have been otherwise.' c£ Karrtunen ( I 97 1 ) for a full discussion o f this. 9 We use the term stative to describe a starive interval in rhe manner of Vlach ( 1 98 I). A stative interval is any interval which includes speech rime. Progressives
329
3 30
13
14
Bernard Comrie pointed out these exam ples to me and the comments of an anonymous reviewer refocussed my attention on them. Kratzer ( 1 98 1 ) defines a consistent set as follows: 'A set of propositions A is consistent, if and only if, there is a world in W (the set of all possible worlds-JPMJ here all propositions of A are true.' We know, of course, that people can hold inconsistent beliefs. But it seems to me that in spite of this people expect know ledge to be consistent That is, given that I must hypothesize what the unknown portion of my world is like ( Q *), I do not expect that both p and �p will be true in it, even though my beliefs about P* may indeed lead to contradictions. Kratzer { 1 98 1 ) defines compatibility of worlds as follows: 'A proposition p is compatible with a set of propositions A if, and only if, A u {p} is a consistent set of
I6
I7
propositions.' We can extend this notion by saying that a set of propositions X is compatible with a set of propositions A iff A u X is a consistent set of proposi tions. In general, a proposition p will be compatible with a set of propositions A so long as �p is not in A and �3q in A such that q - �p . This view of epistemic accessibility is not intended as a model of the dynamics of the knowledge state of an individual, but only as a framework for addressing the meaning of epistemic modal sentences. Nevertheless, it is compatible with such a model, such as that presented in Gaerden fors ( 1 988), which includes probability orderings on propositions within a pos sible world, probability orderings over sets of possible worlds, and rules for dyamic modifications to epistemic states. An anonymous reviewer pointed out these implications.
RE FERE N CE S Antinucci, F. & Parisi, D . ( 1 97 I ), 'On English modal verbs', CLS, T 28-39. Barnes, ]. ( 1 984), 'Evidentials in the Tuyuca verb', Internationaljournal ofAmerican Lin guistics ' so: 25 5-7 I . Boyd,]. & J. Thorne ( I 969), 'The semantics of modal verbs', journal ofLinguistics, 5, 5774·
Coates, ]. ( 1 98 3 ), Tlze Semantics of tlze Modal Auxiliaries, Croom Helm, London and Canberra. Fraser, B. ( 1 97 5), 'Hedged performatives', in P. Cole and ]. Morgan (eds), Syntax and Semantics 3: Speech Acts, Academic Press, New York and London. Gaeardenfors, P. ( I 988), Knowledge in Flux, MIT Press, Cambridge, MA Grice, H. ( 1 97 5 ), 'Logic and conversation', in D. Davidson & G. Harman (eds), Tlze Logic of Grammar, pp. 64-74, Dickenson, Encino, CA. .
Hofmann, T. ( I 966), 'Past tense replacement and the modal system', Mathematical Lin guistics and Automatic Translation , 17, Harvard Computational Laboratory Report to the National Science Founda tion, Cambridge, MA. Hambourger, R. ( 1 987), justified assertion and the relativity of knowledge', Philo sophical Studies, 5 1 , 2: 24 1 -70. Karttunen, L. ( 1 97 I ), 'Possible and must', Syntax and Semantics, 1: 1 -20. Academic Press, New York. Kratzer, A ( 1 98 I ), 'The notional category of modality', in H-J. Eikmeyer & H. Rieser (eds), Words, Worlds, and Contexts, Walter de Gruyter, Berlin and New York. Lewis, D. ( 1 97 3 ), Counterfactuals, Harvard University Press, Cambridge, MA Lyons, ]. ( 1 977), 'Modality', ch. 1 7, Semantics Vol. 2 , Cambridge University Press, Cam bridge. .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
15
Quasi-assertion
J. P. McDowell McDowell, J. (1987), 'Assertion and modal ity', Ph.D. dissertation, University of Southern California. Searle, J. & D. Vanderveken (198 s). Founda tions of Illocutionary Logic , Cambridge University Press, Cambridge.
33 1
Steedman, M. (1977), 'Verbs, rime, and modality', CcgnitiveScience, 1: 216-34. Vlach, F. ( 1 98 1), 'The semantics of the progressive', in Syntax and Semantics 1 4 (Tense and Aspect): 271-92.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Journal ofSemantics 8: 3 3 3-361
© N.LS. Foundation (1991)
Concept Types and Parts of Speech with Special Reference to the Lexicalization of Region Concepts in French C H R I S T O P H S C H W A R ZE
University ofConstance
Abstract
I N T RO D U C T I O N It is a tradition in grammar and linguistics to express overall structures in the lexicon of a language by classifying its words according to their formal as well as to their semantic properties. Formal classifications rest mostly upon inflection and distribution; they yield systems such as the parts of speech (or 'lexical categories'). Semantic classification is less obvious. The methods range from simply postulating common-sense general concepts, such as action, object, quality, etc., to more refined attempts such as the lexical field analysis, the aspectual analysis of verb meanmgs, etc. The problem which this paper addresses is how formal and semantic classes of words relate to each other. I am not going to discuss the general problems which are inherent in the formal and semantic classifications in themselves. For the overall formal structures of the lexicon, I will assume that the traditional system of pans of speech is basically a reasonable classification, at least for the European standard languages. For the overall semantic structure, I do not claim that there is a compara tively simple system of classes: there are far more semantic classes than parts of speech. I think, however, that traditional grammar was right in postulating a closed system of general concepts, according to which the single word meanings are organized. I share the belief of many linguists and Artificial
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It is a tradition to express overall structures in the lexicon of a language by classifying its words according to their formal as well as to their semantic properties. The problem treated here is how formal and semantic word classes relate to each other. The problem will first be discussed on a general level. Then some results of an empirical study on the lexicon of space in French will be presented. It will be shown how the concept of' region' is distributed across the pans of speech of that language. The study confirms and specifies the current assumption that there are typical relationships berween formal word classes and concept types, but it also raises the question of what happens within the domain of atypical lexicalization.
3 34
Concept Types and Parts of Speech: the Lexicon of Space in French
1 THE P R O B L E M O F H O W P A RTS O F S PEE C H A N D C O N CE P T TYPES R E L ATE T O E A C H O T H E R I.I
Three possible answers
To the question of how parts of speech and concept types relate to each other, there are three possible answers:
( ) There is a regular and full correspondence between parts of speech and 1
concept types. (2) Parts of speech and concept types are entirely independent from each other. (3) Between parts of speech and concept types there are regular correspondences, but they are far from exhaustive. My own position falls within the third answer, to which I deliberately gave a rather vague formulation. The aim of this paper is to make this formulation more precise. I will start out by briefly commenting on the two more radical answers. Thesis ( 1 ) underlies the parts of speech system of traditional European grammar. The single parts of speech were defined simultaneously in terms of formal and conceptual properties. For a language like Latin, traditional grammar makes statements like the following:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Intelligence scholars who think that concepts such as physical object, natural kind, colour, motion, location, etc. rank high in the hierarchy of conceptual structure and organize the bulk of the single concepts. (This does not exclude, of course, that an elaborate analysis of a given lexicon may show that more and different general concepts have to be postulated.) It follows from this that I view the question of how formal and semantic classes relate to each other more precisely as the problem of the relationship between parts of speech and general types of concepts. I should add that I do not claim that my observations on this problem are universally valid. For those languages in which word classes are formally less distinct, the problem virtually disappears. This paper has two parts. In the first part I am going to discuss the problem on a rather general level. In the second part I will present some results of an empirical study of the French language. I will pick out one concept type belonging to the conceptual domain of space, namely the concept of 'region'. Concepts of this type are lexicalized in words belonging to various parts of speech. I will try to show how this concept type is distributed across these parts of speech and try to draw conclusions about the general problem.
C. Schwarze
335
(4) The verb is inflected according to tense, mood, number and person. It may govern several kinds of objects and complements. It denotes events or states. (s) The noun is inflected according to case and number. It has three genders. It denotes various kinds of objects. (6) The adjective is inflected according to case, number and gender. It denotes qualities. Traditional grammarians did not worry about possible counter-examples, such as: (4 ) The verb resemble certainly does not denote an event, and one may easily find situations in which it does not denote a state either. (s ) The noun departure does not denote an object of any kind. (6 ) The adjective next does not denote a quality. '
'
Structural linguists generally did not accept the intuitive semantic character izations of traditional grammar.• The notion of morpheme became more important than the notion of word, parts of speech were replaced with distributional form classes, and linguists generally refrained from attributing general class meanings to these form classes. Thus many structural linguists would probaby have argued in favour of thesis (2), if one had asked them to state their point of view on the topic. The problem seemed to be settled. But, seen from the historical distance, structural linguistics possibly just 'dissolved' the problem, instead of solving it. I .2
Associations between concept types andformally defined parts ofspeech
There are a few linguists who have proposed new ideas about the old parts of speech problem. Lyons ( 1 966, 1 977, 1 989) repeatedly argued for the following general claim: there is a systematic association between parts of speech and ontology. But this association, for each part of speech, only holds for a central domain, outside of which word-class2 membership is only weakly motivated or even arbitrary.3 I fully agree with this general conception,4 which may be regarded as a more precise variant of what I formulated above as thesis (3). I will come to the details ofLyons's conception in the second part of this paper. But it may be said immediately that Lyons leaves open various important questions. These questions are: (i) How is the ontology organized? How do the basic concepts connect to more specific ones? (ii) Is there a motivation for given types of concepts to be typically associated with determinate parts of speech? (iii) If clear associations between parts of speech and types of concepts only hold within a central domain, what do we have to assume about the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
'
3 36
Concept Types and Parts of Speech: the Lexicon of Space in French
possibly large number of words that do not belong to the central domain of a given part of speech? Is there some rationale for class membership in the peripheral domains? May a given concept be lexicalized in the central domain of one part of speech and in the peripheral domain of others? And how about word meanings which amalgamate concepts belonging to different types? As an example, take the verb to hunt : it denotes a kind of activity, the objects ofwhich are wild animals. Ifactivities and wild animals belong to distinct types of concepts, how does the complex concept of hunting match with the ontological structure?
1 .2 . 1 Syntactic vs. l ex ical d erivation in Kuryl owicz ( 1 936)
In Kutylowicz ( 1 9 36) the problem of the relationships berween concept types and parts of speech appears as a phenomenon of derivation. The author distinguishes berween a 'syntactic' and a 'lexical function'. The 'syntactic function' of a word is the way in which it appears in the sentence. A word may have more than one 'syntactic function'. The 'lexical function' of a word is the concept associated with it as its lexical meaning. Kurylowicz makes the following claims: (i) The 'syntactic function' of a word may vary, whereas its 'lexical function' may remain constant. If this happens, the word undergoes 'syntactic derivation'. As an example he gives the transition, in Latin, from the finite verb to the present participle: a mat is a verb, whereas amans is an adjective, but their 'lexical function' is exactly the same (Kurylowicz 1 936: 42). (ii) Among the various 'syntactic functions' which a word may have, there is a hierarchy: one of these 'syntactic functions' is primary, the others being secondary and derived from the primary one.5 Secondary 'syntactic functions' can be recognized by their morphological derivedness.6 (iii) The 'primary syntactic function' of a word is determined by its 'lexical function'; in other words, the primary form class of each word is determined by its lexical meaning.7 (iv) The 'lexical function' of a word may vary without a change of its 'syntactic function'. If this happens, we have 'lexical derivation'.8 As
an example, Kurylowicz mentions diminutive formation.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
There are rwo publications, Kurylowicz ( 1 936) and Plank ( 1 9X4), which provide answers to some of these questions. They are probably not as well known as Lyons ( 1 977). I will therefore try to summarize or reconstruct their thoughts (but I will not always respect their terminology).
C. Schwarze 3 3 7
These claims give three important qualifications to the point which had been left vague in our initial thesis (3), namely: (a) There are regular correspondences between parts of speech and concept types, but they are restricted to a domain of primary associations of 'syntactic' and 'lexical functions'. (b) These associations are extremely strong, since, within the primary domain, lexical meaning determines syntactic properties. (c) The secondary domain is connected with the primary one by derivational processes, which cancel the original associations.
-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This conception is surprisingly modern. What I summarize under (a) is roughly the same idea which was to be proposed, 30 years later, by Lyons ( 1 966). Point (b) concerns the question of how the prototypical associations are motivated (even if Kurylowicz does not explain how 'syntactic functions' are determined by 'lexical functions'). Point (c) is a statement on the relationship between prototypical and non-prototypical members of a word class; this relationship is based upon derivational processes. This idea is certainly correct. Derivational processes are an important means of changing syntactic word classes without modifying the conceptual meanings. Bur the idea can be generalized: we also have cases where an identical concept is simply lexicalized in different word classes without any derivational relationship. Think of French, where the concept of'falling' is lexicalized in the verb tomber and in the noun chute , where 'hurt' appears in the verbal lexeme faire mal and in the noun douleur, and where for 'cat' you have the noun chat and the adjective Je1in . These examples may not be very typical. But they might suggest that the distribution of a given concept across the different parts of speech should not simply be viewed as a result of derivational processes. Inversely, they make it plausible that derivational processes are just one means for loosening the primary associations between concepts and parts of speech, the other means being given by syntax (cf e.g. the noun chat and the noun modifier de chat ). Therefore Kurylowicz's first thesis about 'syntactic derivation' can be reformu lated as follows. A given concept may be lexicalized in several words belonging to different parts of speech. That is, a difference of part of speech is not necessarily connected with a modification of conceptual meaning. But among the various parts of speech, only one is primarily associated to the concept. A similar generaJization may be made with regard to Kury:l-owicz's notion of 'lexical derivation'. There are, of course, processes ofword formation which do not change the word class. Diminutive formation is one of them. But there also are the facts of polysemy, which often may be interpreted as the result of synchronically active rules of semantic shift (Schwarze 1 979: 3 1 3 1 6, Schwarze
338
Concept Types and Parts of Speech: the Lexicon of Space in French
1 989). In other words, semantic derivation is not necessarily connected with morphological processes. But this generalization does not invalidate Kurylowicz's idea that primary associations between concept types and parts of speech may be cancelled by semantic processes. This may be illustrated by the polysemy of the French verb longer. In one of its readings it denotes a motion event, namely the motion along an object in the dimension of its length, cf: (7) Je longeais Ia grille du pare. 'I (walked, drove, etc.) along the fence of the park'
(8) La voie ferree Ionge le canal. 'the railroad track (goes) along the canal' In the first reading, the association between parts of speech and conceptual type is intuitively primary. The second reading can reasonably be conceived of as derived by a productive rule of conceptual shift, c£ examples like the following: (9) Germ. Das Wasser ging rnir his zum Knie. 'the water went up to my knee' Engl. The trail crosses a road. It. La strada sale fino a 2000 m. 'the road climbs up to 2000 m' Kurylowicz's conception of 'lexical derivation' can thus be reformulated in the following way. If a word has various related meanings, these may be connected by conceptual derivation. In such a case, only one of the meanings is primarily associated with the part of speech the word belongs to. Conceptual derivation cancels primary associations in the same way as formal derivation does. Two further remarks can be made. The first one concerns the effects of derivations. In the cases we have discussed, the direction always went from the primary to a secondary association of a conceptual type with a word class. But this is not always the case. There are conceptual derivations which change secondary into primary associations. As an example, consider the French noun passage. It is morphologically derived from the verb passer 'to pass'. It denotes an event type, the action of passing. Let us assume that the concept type 'event' is primarily associated with
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In another reading it denotes a static relationship, namely the parallel orienta tion of two objects considered in the dimension of their length, as in:
C. Schwarze 3 39
I
.
2 2 Parts of s peech and s eman ti c categ ori es in Plank ( 1 984) .
Plank ( 1 9R4: s o s , s 1 6) argues that parts of speech are nothing else than descriptive generalizations over the grammatical behaviour ('grammatisches Verhaltensrepertoire') of the words of a given language. He explicitly rejects the idea that parts of speech are based upon semantic categorizations. Nevertheless he does not consider as unsound the associations ofsemantic and formal properties by which traditional grammar claimed to define the parts of speech. 1 0 According to Plank ( 1 984: 509), these claims are most easily confirmed by what one finds in the central domain of the lexicon. This central domain is made up of the words which belong to the basic vocabulary ('Grundwortscharz'), in their kemal constructions ('Kernkonstruktionen') and in prototypical usage ('prototypische Verwend ungen '). He would therefore object neither to Kurylowicz's distinction between primary and secondary associations of 'lexical' to 'syntactic functions' nor to Lyons's formulation that parts of speech reflect ontology in their central domains. But he obviously does not share Kurylowicz's claim that 'syntactic functions' are determined by 'lexical functions'. In fact, instead of simply postulating those prototypical associations of grammatical behaviour and semantic categories, Plank tries to explain and to motivate them.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the verb. Then passage in the present reading shows a secondary association between concept type and word class. In another reading, passage denotes a kind of physical object, namely a small and short covered street, reserved for pedestrians, which connects two larger streets. If we assume that the concept type 'physical object' is prototypically associated with the noun, then this conceptual derivation goes from a secondary to a primary association.9 The second remark concerns the interaction of what Kurylowicz called 'syntactic' and 'lexical derivation'. He had assumed that derivations of both types may apply successively to the same word (Kurylowicz 1 936: 4 5 ) . But his dichotomy of 'syntactic' vs. 'lexical derivation' obscures the fact that a modification on the conceptual level may occur simultaneously with a change of the formally defined word class. If we try to analyse the relationship between the nounfish and the homonymous verb in terms of his conception, we would have to say that the 'syntactic function' and the 'lexical function' change simultaneously. The 'syntactic function' changes from noun to verb, and the 'lexical function' goes from 'a fish' to 'to try to catch fish'. And, what is more important, the conceptual change seems to be necessary. Denominal verbs, in fact, take the concept lexicalized in the noun as an argument of some typically 'verbal' predicate. There seems to be a semantic constraint on purely formal derivations. I will come back to this point in the second part of this paper.
340 Concept Types and Pans of Speech: the Lexicon of Space in French The skeleton of his reasoning is as follows.
A given word
has determinate
'communicative functions', which motivate its repertoire of grammatical behaviour. Words which have similar communicative functions will naturally have similar grammatical behaviour. 1 1 Plank gives several examples in order t o explain how communicative functions (and principles of human perception) motivate grammatical beha viour.
I will cite j ust one of them, the explanation of why the conceptual types
'animate being' and 'physical object' are lexicalized as nouns. Words which denote animate beings or physical objects 'should be pre destinated to be used in referential expressions, the main communicative function of which is to identify the participants of a situation as themes of pre
I 984: s I 7, my translation). Since 'individuation and possibly
quantification of such lexical meanings is perceptually easy and . . . since there are good chances that they are communicatively necessary', 'the expressability of oppositions of number belongs to the privileged properties of grammatical behaviour of this semantic class of lexemes.'1 3 I f we compare Plank's
I 984 article with Kurylowicz ( I 936) and Lyons ( 1 977),
we see a considerabl e agreement: all three authors distinguish between a prototypical and non-prototypical domain in the lexicon. They all assume prototypical associations between concept types and parts of speech. Further more, Plank's article helps us to clarify our position in two important points: (i) Even within the prototypical domain of the lexicon, the association of conceptual types with parts of speech does not have the character of determination, but rather of motivation and preference: (ii) These associations can be motivated on the level of communicative function and of perception. Let us also observe that all of these authors do not say very much about the relationships between the prototypical and the non-prototypical domain of the lexicon. Is anything possible within the latter? We shall pursue this question in the second part of this paper.
1.3
Semantic types and concept types
The traditional thesis of a regular and full correspondence between parts of speech and concept types as formulated above under (r ) has been vigorously ,
renewed by Montague Grammar. This theory associates so-called semantic types to syntactic categories, in such a way that syntax and semantics have a perfectly parallel structure. In this theory, the parts of speech appear as 'basic categories'. Like all categories, each basic category represents a semantic type (Cresswell 1 97 3 :
72). A proper
I 27-
noun represents an individual, an intransitive verb represents a
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
dications'12 (Plank
C. Schwarze 34 I
( 1 0) There is a privileged association between the conceptual type 'transmis sion event' (which includes concepts such as giving, bringing, sending, etc.) and the semantic type (o, 1 , I , I ) (which is syntactically represented by three-place transitive verbs). This privileged association is motivated by the fact that three-place transitive verbs open slots for the persons and objects essentially involved in transmission events. But what can be said about the non-prototypical associations? It would be nice if we could confirm an hypothesis like the following: ( I I ) A given concept c can be lexicalized in the parts of speech p 1 , . . ., Pn• iff p 1 , , Pn represent the same semantic type. •
•
•
Let us see whether this hypothesis is defensible. It seems to be supported by the semantics of nominalized verbs, or, more generally, the conceptual correspondences between verbs and event nouns. Consider, e.g., the verb to call and the corresponding noun (a) call . On the level of semantic types, both words represent a property (in standard notation: (o, 1 )). On the level of conceptual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
property, and so on. One point is of special interest for our problem: there are more parts of speech than semantic types, so that different parts of speech may represent the same semantic type. Thus intransitive verbs and the so-called common count nouns both represent properties (Cresswell 1 973: 1 3 3). This could remind us of the above thesis that a given concept may be lexicalized in different parts of speech. We must keep in mind, however, that the semantic types of Montague Grammar are not identical with the conceptual types we have discussed until now. Semantic types are based upon a general and a far more simple ontology, and they are motivated exclusively by the needs of sentential semantics: the aim of the theory is to explain how sentence meanings are built up from the meanings of words and constituents. If we accept the idea that for the sake of sentential semantics it is reasonable to postulate a set of semantic types, then we must postulate a model of lexical meaning made up of two levels: the level of semantic types and the level of conceptual meaning. 14 We can now come back to the question ofhow conceptual types and parts of speech are related, reformulating it like this: are the contents of both levels of lexical meaning completely independent from each other, or are there constraints of some kind? As far as the central domain of the lexicon is concerned, Plank's claim that there are prototypical associations of concepts and repertoires of grammatical behaviour seems co be a good answer, and it seems easy to add more examples of motivation to the one he gives; cf the following:
342
Concept Types and Parts of Speech: the Lexicon of Space in French
r .4
Tentative conclusions
The following propositions can be concluded from the above discussion: (i) There are concepts which, in a given language, may be lexicalized m different parts of speech. (ii) There is a central domain in the lexicon, in which prototypical associations between parts of speech and conceptual types exist. In the peripheral domain of the lexicon, those prototypical associations are cancelled. (iii) These prototypical associations are motivated on the level ofcognition and of communication. (iv) But there are semantic derivations which may re-establish prototypical associations even within the peripheral domain. (v) In the lexical meaning of a word there is also a level of semantic types, which accounts for the manner in which the word meaning is integrated
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
meaning, they both denote the same kind of activity. This is what the hypothesis predicts. There seems to be an additional advantage: if the hypothesis is correct, it would also explain why lexicalizing an event as a noun does not imply that the event is changed into a thing. In other words, the hypothesis excludes mysterious processes such as 'reification' of a process. But, unfortunately, the hypothesis proves to be incorrect. Let us consider the relationship between colour adjectives and the corresponding colour nouns. Adjectives and nouns represent different semantic types. (Adjectives are ((o, I ), (o, I )), whereas nouns are (o, I ).). Consequently, the hypothesis predicts that, e.g., the adjective green and the noun green (in expressions such as I like that green ) cannot represent the same concept. But this evidently is not true, the concept 'green' being exactly the same in both words. Another false prediction of ( I I) concerns denominal verbs, such as to fish . Intransitive verbs and nouns both represent properties (their semantic type is (o, I )). The conceptual meaning of the noun should consequently appear unaltered in the verb. But, as we know, it is amalgamated with another concept, and there are strong reasons to assume that this is necessarily the case. Denorninal verbs seem to require an amalgamation of the nominal concept with a concept of the type 'action' or 'behaviour'. It is difficult to tell what follows from the refutation of (I I). One possible conclusion is that the level of semantic types does not constrain the level of conceptual types. Another possible conclusion is that the semantic types of Montague grammar prove to be inadequate if we take into account lexical semantics. More research is necessary, both on sentential and on lexical semantics.
C. Schwarze 343
into sentence meaning. It is an open question whether associations between conceptual types and parts of speech are constrained by semantic types.
2 RE G I O N C O N CE P T S I N T H E F RE N C H L E X I C O N
( 1 2) External regions: the in-region 1 5 the outside-region the unspecified near-region the whole near-region the external front-region the behind-region the external side-regions
the on-region the under-region the above-region the below-region the between-regions
('in x', 'through x') ('outside of x') ('near x') ('around x') ('in front of x') ('behind x') ('on the right-hand side of x') ('on the left-hand side of x') ('on x') ('under x') ('above x') ('below x') ('between x') ('among x')
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Let us now see how the conclusions of the above general discussion will appear in the light of actual language description. As I said in the beginning, I will consider the lexicon of French in order to see how a certain type of spatial concept, namely the general concept 'region', is distributed across the parts of speech. I will first introduce some terminology. Regions are sections of space, which are defined with regard to physical objects. There are two varieties of regions, external and internal regions. An external region is a section of space which is distinct from the section of space occupied by the physical object. An internal region is included in the section of space occupied by the physical object. It can therefore always be used for referring to a part of that object. Regions are typically defined, as is well known, by means of the following concepts: unspecified nearbyness, includedness (with the distinction between within and without) horizontal orientation, from which derive frontality, laterality (with the distinction between right-hand and left-hand), and the vertical axis (with the distinction between above and below). In the fragment of French which I will consider, the following region concepts occur:
344
Concept Types and Parts of Speech: the Lexicon of Space in French
( 1 3) Internal regions: the interior region the exterior region the internal front-region the rear region the internal side-regions the upper region the lower region the central region the peripheral regions
('the interior of x') ('the exterior of x') ('the front of x') ('the rear of x') ('the right-hand side of x') ('the left-hand side of x') ('the top, the upper part of x') ('the bottom, the lower part of x') ('the middle of x') ('the peripheral parts of x')
( 1 4) External regions Internal regions m-regwn interior region outside-region exterior region internal front-region external front-region rear reg10n behind-region internal side-regions external side-regions upper regwn on-region, above region under-region, below-region lower region
Criteria included ness includedness frontality frontality laterality vertical axis vertical axis
In the following presentation of empirical facts, will begin with the concepts of external region, and I will examine whether and how they are lexicalized as prepositions, adverbs, verbs, adjectives and nouns. The concepts of internal region will come in during this overview. Throughour this section, for each part of speech ':"e will ask the question whether the association with region concepts is prototypical or not. We will start out with a Lyons-type assumption about the prototypical semantics of the specific word class and see what it predicts with respect to region concepts. We will then look at the French inventory and try to verify whether it meets the prediction. The criteria for this will be twofold: (a) the derivational status (simple vs. derived) of the singular items, and (b) the elaborateness of the inventory. The first criterion is a consequence of Kurylowicz's idea that non prototypical members of a part of speech are derived. The second criterion is based on the assumption that the lexical items which are in the prototypical centre of a word class represent a differentiated and well organized conceptual system, whereas peripheral items may represent just fragments of such a conceptual system.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
External and internal regions are different subtypes of the more general concept of region. But there are partial correspondences between both systems. They are due to the fact that the criteria of definition of the regions are partially the same. These correspondences are summarized in ( 1 4):
C. Schwarze 3 4 5
The combination of the two criteria may yield an interesting result that has not yet been discussed: it may be the case that, within a given word class, there is a group oflexical items which are morphologically derived but which also form an elaborate system. This is a contradictory situation if we postulate a clear-cut dichotomy between a central and a peripheral domain. But if we conceive the word class as a scalar structure, we will be able to admit degrees of typicality. More precisely, we will be able to see a typicality structure even within the peripheral domain of a word class. 2.1
Prepositions
( 1 s) Regions m-regwns
Prep ositions
dans x chez x d travers X hors de x outside-region unspecified near-region pres de x autour de x whole near-region devant x external front-region enface de x derriere x behind-region a cote de x external side-regions a droite de X 1 7 agauche de x on-regwn under-region above-region below-region between-regions
le long de x sur x sous x au-dessus de x au-dessous de x entre X parmi X
'in x' 'in x's home' 'across x' 'outside x' 'near x' 'around x' 'in front of x' 'opposite x' 'behind x' 'next to x' 'on the right-hand side ofx' 'on the left-hand side of x' 'along x' 'on x' 'under x' 'above x' 'below x' 'between X' 'among X'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Prepositional phrases with lexical prepositions (as opposed to grammatical prepositions, which have a mere case marking or conjoining function) denote properties or special kinds of entities. Local prepositions use region concepts in order to create expressions which denote properties or places. 1 6 We will therefore expect concepts of region to be prototypically associated with the word class of prepositions. The following table shows the inventory oflocal prepositions in French:
346
Concept Types and Parts of Speech: the Lexicon of Space in French
2.2
Adverbs
As is well known, the term 'adverb' covers a collection of lexical subcategories which are quite different in form and content. But in the present context there is no need to discuss this problem, since we will only be concerned with the so called pronominal adverbs. Syntactically, pronominal adverbs are constituents which have the same distribution as prepositional phrases, but, unlike prepositional phrases, they contain no governed noun phrase. Semantically, they are predicates which incorporate an anaphoric argument. They are comparable to transitive verbs used without an object: I'll wait outside is to I'll wait outside the house what I know is to I know the story . Consequently, if a given language has identical forms for ordinary prepositions and prepositions with an incorporated argument, there is no reason to postulate a difference of word class: one would not like to say of that language that it has pronominal adverbs. But the prepositions and the corresponding adverbs may be different in form. It is only in this case that one would postulate different word classes and that the question of association of a word class with concept types arises. So, before any further discussion, let us look at the facts of French. Here is the mventory:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The table shows that French prepositions lexicalize a quite differentiated field of concepts of external region. Looking at the items, one might think, however, that the system is morphologically heterogeneous. For some of the concepts there doesn't seem to be a synchronically simple preposition, but only a complex expression with de (hors de, pres de , etc.). But a more thorough formal analysis shows that this impression is inadequate. In fact, French lexical prepositions behave like verbs: their arguments may be direct or indirect. Thus dans , devant , derriere , sur and sous have a direct argument, whereas hors , pres and au tour have an indirect argument, the marker ofwhich is de . There remain some complex prepositions, which are lexicalized phrases: a travers , en face (de), a cote (de), a droit (de)!a gauche (de), le long (de), au-dessus (de) and au-dessous (de). Lexicalized phrases are not morphologically derived, but the central domain of the lexicon typically is made up of (synchronically) simple words. Consequently the complex prepositions above do not belong to the central domain of the lexicon. But this has nothing to do with the question of whether the lexicalization of region concepts as prepositions is prototypical or not; it has to be explained by the marginal position of the respective concepts in the conceptual system. Hence we can conclude that the above prediction is correct: the association of 'external region' to 'preposition' is prototypical.
C. Schwarze
Prepositions
(16) Regions in-region
dans x
chezx d traversx
outside-region unspecified near-region whole near-region external front-region behind-region external side-regions
Adverbs dedans d travers dehors
(tout) pres autour devant enfoce derriere a cOte a droite a gauche dessus dessous
au-dessus au-dessous
The table shows that there are adverbs for most of the region concepts which are lexicalized as prepositions. But the fact that some prepositions have no adverbial counterpart cannot be ignored, especially if we see that the same phenomenon also occurs with non-spatial prepositions; c£ the following list:
( 1 7) preposition avec x sans x pourx contre x envers x chez x parx avant x apres x pendant x jusqu'a x malgre x
'with x' 'without x' 'for x' 'against x' 'towards x' 'at x' place' 'through x' 'before x' 'after x' 'during x' 'until x' 'in spite of x'
pronominal adverb avec 18 'with (it)' sans 'without' pour 19 'in favour (of it)' contre 'against (it)'
avant apres
'before' 'after (it)'
It is not obvious how these data should be interpreted: one might say that there is some kind of word class distinction between prepositions and pronominal adverbs. But an alternative description is possible, namely that there is no word
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
on-regton under-region above-region below-region between-regions
hors de x pres de x autour de x devant x enfoce de x derriere x a cOte de x a droite de X agauche de x le long de x sur x sous x au-dessus de x au-dessous de x entre X parmi X
3 47
34S
Concept Types and Parts of Speech: the Lexicon of Space in French
class of pronominal adverbs, but merely a distinction within the class of prepositions, some of them being able, and others not, to incorporate an anaphor. But there is a possibility of deciding between these alternatives. Even though in the majority of cases the preposition and the adverb are morphologically identical, the adverb differing from the preposition only by incorporating the argument (including the grammatical marker de), there are some cases where there is (synchronically) no systematic formal reltionship between the pre position and the adverb; c£
( I 8) dans 'in' vs. dedans 'inside'. sur 'on' vs. dessus 'on top'.
2.3
Verbs
According to tradition (and also Lyons 1 97T 448 £), verbs most typically refer to actions. There is no way, of course, to consider regions as actions. But shall we conclude that, if a language has verbs which lexicalize concepts or region, these verbs must belong to the peripheral domain? French belongs to those languages which lexicalize path aspects in verbs of motion (Talmy 1 98 5; Schwarze I 985; Wienold & Schwarze 1989). Now a path can be described by specifying the sections of space where it begins, where it ends, and through which it passes. And these sections of space may be identified with respect to some physical object; that is, according to the definition given in section 2, they may be regions. We will thus expect region concepts to be lexicalized in French verbs of motion, and this is actually the case. Of course a path can also be described without reference to a region, e.g. by specifying a direction. So we will not be surprised to see that several French verbs of motion do not lexicalize region concepts. The verbs for coming and going (aller and venir) specify a direction with respect to a point of reference which must be identified by deixis. The verbs for moving up and moving down
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
These cases, which are marked by bold characters in table ( I 6), are numerically a minority, but they are central in the conceptual system of spatial relationships. So the conclusion will be that in the French language, prepositions and pronominal adverbs are different word classes, but with an extremely low degree of differentiation. With respect to the general problem which we are discussing, namely the distribution of concept types across parts of speech, it is legitimate to consider them together as the class of 'prepositions and pronominal adverbs'. Consequently we can reformulate the result we had obtained for preposi tions by saying: the association of concepts of external region to prepositions and pronominal adverbs is prototypical.
C. Schwarze 3 49
( 1 8) Regions m-regwn
Self-motion
Transport
entrer, rentrer, rentrer, introduire, traverser , penetrer enfoncer , inserer, enfouir, importer
outside-region
sortir
sortir, extraire, exporter, extrader, expulser
unspecified near-region
accourir , passer
( r )approclzer, eloigner
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
( monter and descendre ) specify directions along the vertical axis. (They are, of course, closely related to the prepositions and adverbs which specify regions with regard to the vertical axis, but the interpretation of mon ter and descendre does not require one to identify a region.) The verbs for arriving, leaving and passing (arriver, partir and passer) do not even specify a direction. They simply characterize a motion event with regard to the origin, the goal or an intermediate section. The lexical meanings of these verbs only require that a section of space be identified from the context. The verbs themselves do not specify any direction or region. But let us return to those verbs which do lexicalize region concepts. The region concepts that are lexicalized in verbs are also lexicalized in prepositions, with one exception.20 Now regions being places, we will expect, in anology to what we have seen regarding afish and tofish , that the region concepts can only appear in the verb when they are amalgamated with some other concept. The facts meet this expectation. The region concepts are amalgamated with the concept of motion.2 1 This amalgamation is not an amorphous one; the amalgamated concepts are organized in a predicate-argument structure. Since the region is used to characterize the path of a motion, it is a partial factor of the global motion event. Therefore we can say that the region has a subordinate position within the complex verb meaning. The verbs which lexicalize region concepts comprise two syntactically and semantically distinct groups, the verbs of self-motion and the verbs of transport. The verbs of self-motion describe the motion event without postulating a cause. The physical object (or animate being) which moves appears as the subject. The verbs of transport describe the motion event as being caused by some external agent. The agent appears as the subject, whereas the moved object appears as direct object. Verbs of both groups, but especially transport verbs, may amalgamate not only region and motion, but additionally a large variety of non-spatial concepts. In the following table I present only some of them; they are printed in bold characters.
JSO Concept Types and Parts of Speech: the Lexicon of Space in French
Regions whole near-region external front-region behind-region external side-regions on-region under-region above-region below-region between-region
Self-motion
Transport
contourner longer survoler
( 1 9) If a verb meaning has a predicate-argument structure, then the lexical centrality of the verb depends only on the concept type associated to the highest predicate. The incorporated argument is not relevant.
Hence the morphological simplicity of verbs like entrer, sortir and passer actually is an expression of their prototypicality. They are prototypical because they lexicalize the concept of motion. The fact that the motion verbs of French (and of other languages) lexicalize path aspects not only in terms of deictic orientation (like 'come' and 'go'), bur also by means of regions is a very natural consequence from the concept of motion: motions go along paths, and paths can be defined with respect to regions. But in languages which, like French, have an elaborated system oflocal prepositions, there is a tension between the parts of speech system and conceptual structure.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This overview shows an asymmetry between the verbs on the one hand and the prepositions and pronominal adverbs on the other hand. A considerable part of the region concepts lexicalized by the latter are not lexicalized in verbs. The difference between verbs on the one hand and prepositions and pronominal adverbs on the other hand also appears on the morphological level: there are almost no derivational relationships.22 What can we conclude, from these observations, about the prototypicality of region concepts lexicalized in the meanings of verbs? Kurylowicz's derivational criterion seems to suggest that the association is prototypical. Verbs like entrer, sortir and passer are visibly not derived, and there is no (synchronical) derivation between verbs and prepositions or adverbs. But the criterion of elaborateness seems to suggest the contrary: in comparison to what we saw for the prepositions and adverbs, the conceptual system appears only in fragments. How can this apparent contradiction be explained? We have seen that there is no simple lexicalization of region concepts in verbs, and that the region concepts are assigned to incorporated arguments. We can now formulate the following additional criterion for the prototypicality of associations of concept types with word classes:
C. Schwarze 3 S I
2.4
Adjectives
(2o) Internal regions Adjectives interieur internal region extirieur exterior region internal front region anterieur postirieur, arriere rear reg10n internal side-regions lateral droit gauche superieur upper reg10n infirieur lower region central central region piripherique peripheral regions
'internal' 'external' 'front' 'rear' 'side' 'right-hand' 'left-hand' 'upper' 'lower' 'central' 'peripheral'
The table shows a conceprually well-differentiated inventory, which is regularly related to the prepositions and adverbs, according to the correspon dences given in table ( I 7). The criterion of elaborateness, therefore, seems to suggest that the association of region concepts with adjectives is prototypical,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
According to Lyons's theory of the ontological foundation of parts of speech distinctions, adjectives are prototypically associated with simple concepts of quality (Lyons I 977; I 4 f). Since it would not be plausible to consider regions as qualities, Lyons's assumption predicts that the lexicalization of region concepts in the word class of adjectives, if there is any, is not prototypical. When we consider the lexicon of French, we do find adjectives which lexicalize region concepts. But only very few of them lexicalize concepts of external regions; they are adjacent 'adjacent', contigu 'contiguous', proche 'near', and lointain 'remote'. The remaining majority of region adjectives lexicalize internal regions. It must be added, however, that the adjectives do not denote the regions by themselves. The region as such is denoted by a noun; it is only specified by the modifying adjective. The meaning of the noun often is or implies the concept 'part of something'; typical nouns of this kind are partie 'part', cOte 'side', quartier 'neighbourhood', etc. As a consequence of this, and according to the definition I gave above, there must always be a global physical object, of which the region is, loosely speaking, a part. This global object has no typical syntactic realization, and in general it is actually not realized. Examples are Ia levre superieure 'the upper lip', where the global object is the mouth, or les boulevards exterieurs , 'the outer boulevards', where the global object is the street system of a city. The following table gives an overview of the French adjectives of internal regwn.
3 52
Concept Types and Parts of Speech: the Lexicon of Space in French
that is, that the above prediction is wrong. Kurylowicz's criterion of non derivedness yields the same result. None of the adjectives (except peripherique) is morphologically derived. But intuitively this result needs further discussion. Lyons's claim that adjectives typically denote simple qualities of what he calls first-order entities is absolutely plausible. And there are comparative facts which seem to confirm it: Germanic languages do not have adjectival equivalents for the whole French list. Several of the English translations I gave in table (zo) are not genuine adjectives, and in German most translational equivalents are the first element of compounds; cf Innenwand Oberlippe Mittelschiff
Also Plank's thesis, according to which a word is liable to show a prototypical association of concept type and part of speech if it belongs to the basic vocabulary of the language, might yield an argument in favour of the initial prediction: should it be the case that the adjectives of table (2o) do not belong to the same basic vocabulary of French? None of the adjectives (except arriere) belongs to the historical layer of native Romance forms. And although it is not quite clear how the basic vocabulary of a language is to be defined, one might tend to think that native forms are more likely to belong to the basic vocabulary than the learned words which make up most of the above list. The apparent contradiction berween the a priori prediction and the results suggested by the criteria of elaborateness and derivedness can be overcome if the undeniable elaborateness and underivedness of the inventory of region adjectives can be explained independently from considerations of prototypi cality. This is actually the case. Why does French have an elaborated system of region adjectives? In order to answer this question it is useful to observe that there are rwo basic types of adjectives. Besides the adjectives of simple quality, which Lyons had in mind when he formulated his thesis about typical adjectives, there is another type, the relational adjective. Adjectives of this kind also denote simple qualities, but these qualities are not intrinsic: they are characteristic relations. English examples are adjectives which denote an origin or a class membership (e.g. European (settlers), literary (texts ), domestic (animals). Now in French, as in other Romance languages, relational adj ectives are far more numerous than in English or German. The reasons for this are both cultural and grammatical: the standards of modern Western literacy create a demand for words which makes it possible succinctly to characterize entities with respect to their relations. Some languages, e.g. English or German, meet this demand with powerful mechan isms for the generation of compounded nouns. Now the Romance languages,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(2 1) paroi interieure levre superieure nef centrale
C. Schwarze 3 5 3
2.5
Nouns
Nouns mainly appear in noun phrases.25 According to Lyons ( 1 977: 442 £, 446), noun phrases typically refer to 'first-order entities', i.e. 'persons, animals and things' ('more or less discrete physical objects') which are 'relatively constant as to their perceptual properties', 'located, at any point of time, in . . . a three dimensional space', and 'publicly observable'. Now, if noun phrases typically refer to 'persons, animals and things', then there is a prototypical association between the word class 'noun' and the type of concept that Lyons calls 'first order entity'. On the basis of this assumption, we will expect that nouns which lexicalize concepts of regions will not be prototypical nouns: regions are no 'first-order entities', because they are not persons, animals or things. But there is an important difference between external and internal regions. External regions are places. Even if places are typically defined with regard to individual physical entities, they are clearly distinct from these entities. We will consequently expect that nouns which lexicalize concepts of external regions are marginal. Internal regions, on the other hand, are less distinct from individual physical objects, since they are parts of them. In fact, if parts are no first-order entities, they certainly are not second-order or third-order entities.26 This will make us
c
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
having only poorly developed nominal compounding, meet that demand by their abundance of relational adjectives. Therefore we can conclude: the French inventory of relational adjectives is elaborate because these adjectives meet a communicative demand. Let us turn now to the criterion of derivedness. When dealing with a language like French, one must have in mind that derivation is not the only means of expanding a word class: there also is borrowing, especially from the extinct, but culturally coexistant languages Greek and Latin. Thus there are no adjectives which are morphologically derived from rei/ 'eye' or eau 'water', the respective adjectives being oculaire and aquatique (or ophtalmologique and aqueux respectively, with a difference of meaning). Now it is interesting to see that these words are morphologically derived on the level of their languages of origin. This also holds for all the learned region adjectives in our list. So the conclusion is: the French region adjectives form an elaborate system because they are needed, and they seem to be underived because they are borrowingsP So there is no reason not to confirm Lyons's thesis: the lexicalization of region concepts as adjectives is not prototypical on the level of a general theory of parts of speech. But, on the other hand, it is typical for the modern French language, and it can be explained on the level of French word formation.24
3 54
Concept Types and Parts of Speech: the Lexicon of Space in French
expect that nouns which lexicalize internal regions will be prototypical to some extent, especially if the internal region is a salient one. French nouns of region resemble the adjectives we just discussed: most of them lexicalize concepts of internal regions. There are only two nouns which lexicalize a concept of external region: les environs , les alentours , which both mean 'the surroundings'. But the region nouns systematically differ from the region adjectives in their semantic structure. Whereas region adjectives merely identify regions denoted by nouns, region nouns denote (and specify) the internal regions by themselves. The following table shows the inventory of nouns of internal region. Internal regions
Nouns
interior region exterior region internal front region rear region27 internal side-regions upper region lower region central region peripheral regions
/'interieur /'exterieur le devant
'the inside' 'the outside' 'the front'
le cote le haut le bas le milieu le bord Ia peripherie
'the side' 'the top' 'the bottom' 'the middle' 'the edge' 'the outskirts'
The region nouns belong to the larger lexical field 'part of something', which contains the names of the body parts, words like top , bottom , summit , bank (ofa river) , etc. Unlike most other nouns in this lexical field, nearly all the nouns of region are in some way derived, and they stem from quite heterogeneous sources. Two of them (l'interieur, /'exterieur) are nominalizations (without suffixation) from the corresponding region adjectives; two others (le haut, le bas) are norninalized adjectives of dimension; one (devant) is a nominalized preposition; two ( le bord, le cOte) have diachronically been derived by generaliza tion of meaning;28 one (milieu ) stems from a noun phrase (Lat. medius locus) meaning 'middle place'. The high degree of derivedness of the inventory is a point for the conclusion that the association of region concepts with nouns is not prototypical. On the other hand the inventory is conceptually well differentiated, which speaks in favour of the prototypicality of the association. The fact that the criteria of derivedness and of elaborateness yield contradictory results is not surprising if we think of the position of internal regions in Lyons's system of entities: they are located between 'first-order entities', which are typical for nouns, and 'second-order entities', which are not. A more refined ontology would be needed for specifying this intermediate position of internal regions and of parts in general. It would probably postulate
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(2 1 )
C. Schwarze 3 S S
a category of physical entities which resemble persons, animals and things in all respects, except that they are not discrete with respect to some first-order entity, of which they are a part. Salient body parts (the head, the feet, etc.), salient parts of artefacts (a handle, a roof, etc.) and of plants (the roots, the branches, etc.) would be typical members of this category, and internal regions would be derived from these by metaphor or on the basis of general region concepts. A model of this kind could explain why nouns of region, without being prototypical nouns, are not merely to be seen as lying within a domain of ontologically unclear instances.29
Summa ry cfthe empirical part
The overall picture we get is the following. Concepts of external region are prototypically lexicalized as prepositions and pronominal adverbs. The semantic difference between prepositions and pronominal adverbs does not affect the conceptual meaning. Concepts of external region also appear in the meanings of verbs , namely verbs ofmotion. This fact is due to a language-specific property ofFrench. Since the region concepts are only incorporated arguments, their presence is not relevant for the prototypicality of the verbs in which they are lexicalized. Concepts of internal region are lexicalized in adjectives and in nouns. There is a systematic semantic difference between adjectives and nouns, but it does not affect the conceptual meaning. The association of concepts of internal region with adjectives is not prototypical with respect to a general theory of the parts of speech. But it is anchored in the grammatical system of the French language. Nouns which lexicalize concepts of internal regions do not belong into the prototypical domain, but they are very close to it.
3
GENERAL CONCLUSIONS
In order to bring the two parts of the present paper together, let us reconsider the tentative statements that we had formulated at the end of part one.
3.1
One concept type in various parts ofspeech
The first statement was: 'In a given language, a concept may be lexicalized in different parts of speech.' This statement is confirmed by the empirical analysis. Even if not all particular concepts of region appear in the various word classes,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
2.6
356 Concept Types and Parts of Speech: the Lexicon of Space in French
the more general concept type is represented in prepositions and pronominal adverbs, verbs, adjectives and nouns. 3 .2
Central vs. peripheral domain ojthe lexicon
The second statement was: 'There is a central domain in the lexicon, in which prototypical associations berween parts of speech and conceptual types exist.' The anaysis of the distribution of region concepts across prepositions, adverbs, verbs, adjectives and nouns has confirmed this thesis. But the following additional statements turned out to be needed.
3·3
Motivatedness ofassociations
Throughout the empirical part, I did not discuss explicitly the third statement, which read: 'These prototypical associations are motivated on the level of cognition and of communication.' The only section in which I could have tried to do so is the one which treats prepositions. What I could have said is this: if a language has prepositions, then prepositional phrases are highly appropriate for denoting external regions for the following reason. We defined external regions as sections of space which are defined by some local relation with a reference point, typically a physical object. Now local prepositional phrases consist of a predicate, the preposition, which specifies the local relation, and an argument, the governed noun phrase, which specifies the physical object. I think that this attempted motivation is plausible and not circular. But it has the weakness that the assumption of a conceptual (or ontological) category 'region' relies on a specific view of the semantics of local prepositions. I think that no general conclusions can be drawn from this difficulty. But a certain amount of scepticism seems to be in order. Even if we can propose plausible motivations in particular cases, it is not certain that the thesis according to which prototypical associations of word class and concept type are
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(i) If a concept is amalgamated with other concepts, it may occupy different ranks in the word meaning: it may be superordinate or subordinate. Only the superordinate meaning component is relevant for the prototypicality of the association of concept type and part of speech. (ii) The mere dichotomy berween a central and a peripheral domain is not satisfactory. There are intermediate positions, which might be defined by a more detailed ontology. Furthermore, the grammatical structure of a given language may systematically favour associations which are not prototypical on the general level of a theory of parts of speech.
C. Schwarze 357
motivated on the level of cognition and communication can be fully sustained. In all events, finding these motivations is not as easy as it promises to be when discussed in general terms. 3·4
Prototypicality re-established by semantic derivation
to the fourth statement, according to which semantic derivations may re establish prototypical associations even within the peripheral domain, our fragment of French did not contain relevant material. But the claim is safe, since it would be easy to multiply examples such as French passage (c£ 1 .2. 1 , above).
As
Semantic types
The fifth statement was about whether the associations of conceptual types with parts of speech is constrained on the level of semantic types. I have not discussed this problem throughout the empirical part. But there are results which, even though they do not resolve the problem, give some relevant evidence. The fact that there is a systematic semantic difference between prepositions and adverbs, as well as between adjectives and nouns, is certainly due to differences in semantic type. But since these differences do not concern the conceptual aspects of word meanings, these facts prove that a difference in semantic type does not necessarily induce a conceptual difference. On the other hand, the possibilities oflexicalizing concepts in verbs seem to be clearly constrained. But, as I suggested above, these facts need further investigation.
4
P O S T S C R I PT
It was only after I finished the present text that I had the opportunity to read William Croft's article, in this journal , on the conceptual and communicative theory of parts of speech (Croft 1 990). As far as the association between parts of speech and concepts is concerned, Croft follows and further elaborates the typicality approach, which, as has been shown above, goes back as far as 1 9 36. His most original idea obviously has been to combine conceptual structure and categories of speech act theory in a theory of the parts of speech. This combination may partly explain why there are so many untypical elements within the major lexical categories. According to Croft, nouns are used for categorizing and adjectives for modifying. Now, a given concept, say 'sick', is likely to be lexicalized as an adjective. As such, its function is to modify
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
J.5
3 5 8 Concept Types and Parts of Speech: the Lexicon of Space in French
Acknowledgements Previous versions of this paper were presented in March I 990 at the annual meeting of the German Linguistic Society (DGfS) in Saarbriicken and at the Berkeley Linguistic Colloquium. I wish to thank Peter Pause, Frans Plank and Gotz Wienold for their comments on this paper in its various earlier states, and I am grateful to Bruce Mayo who checked the English text. CHRISTOPH SCHWARZE
Universitiit Konstanz Philosophische Fakultiit Fachgruppe Sprachwissenschafi Universitiitsstr. 1 o 7750 Konstanz Germany
NOTES I For a good characterization o f the tradi
2.
tional theory and a summary of the main structuralist objections, see Lyons ( I 977: 42.3 ff.). I use the term 'word class' as a stylistic variant of'part of speech'. 'The thesis that will be maintained here is that the semantic, or ontological part ofthe traditional definitions of the parts-of speech define for each part-of-speech, not the whole class, but a distinguished sub class ofthe total class. Each such semantic ally defined subclass is focal within the larger class . . .' (Lyons I 977= 440).
4 I prefer to use the term 'types of concepts' rather than Lyons's term 'ontology'. I would like the terminology to hint at the fact that the system underlying the parts of speech is not the structure of the world in itself, but the result of human categori zation. And I am· not sure whether this conceptual system is as well organized as the ontology inherent to a formal lan guage. 'Du fait qu'un mot designant une qualite (c'est-a-dire un adjecti� peut fonctionner soit comme epithete, soit comme attribur (predicat), soit comme support autonome
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the category expressed by the noun to which it is syntactically related. But the concept 'sick' may also be useful for categorizing. Therefore some languages, e.g. French or German, have untypical nouns which are derived from adjectives: un malade , ein Kranker 'a sick person'. But speech act categories probably do not explain all untypical associations of word class and concept type. Complementary explanations may be found on the level of the organization of discourse. There are certain nouns, such as fact, matter, event , person , object , place , which, in spite of being excellent means for categorization, are untypical nouns. Their raison d'etre seems to be that they appear in noun phrases which are used as anaphoric expressions. Other untypical nouns, such as departure , connection , result, sincerity, foolishness , seem to be needed in order to make it easier to treat events and properties as arguments of other predicates. They also are useful, on the level of control of the flow of information, for presenting events and properties as topics.
C. Schwarze 3 59 de determination (adjecrif anaphorique),
had formulated interesting reflections on
il ne s'ensuit pas que routes ces foncrions
this topic (Plank
syntaxiques soient au meme degre essen
6
II
rielles ou caracterisriques de Ia partie du
lexikalische
discours en question . . . Autrement dit, le
unter die gleiche oder unter eine sehr ahnliche Bedeutungskategorie fallen, ist
Si le changement de Ia foncrion syn
die gleichen oder ahnliche Verhaltensei genschaften als total verschiedene Eigens
I2
chaften besitzen' (Plank '
. . •
I 984: 5 I6).
dazu pradesriniert . . . in referierenden
Ausdriicken verwendet zu werden, deren kommunikarive
foncrion syntaxique primaire celle qui
Situarionsbeteiligte als Thema von Pradi
correspond a Ia forme-base, et foncrion
karionen zu idenrifizieren' (Plank
I 936: 42).
I3
Hauptfunkrion es ist,
5 1 7).
I 984:
'[Da] lndividuierung und eventuell Quan rifizierung . . . bei solchen Lexembedeu
'. . . les mots possedent une foncrion syn
tungen perzeptuell gut moglich und in
taxique primaire en vertu meme de leur
dieser Funkrion kommunikariv am ehe
sens lexical . . .' (Kurylowicz
I 936: 43 );
sten norwendig [ist, gehort] die Ausdriick
'. . . les foncrions syntaxiques primaires
barkeit von Numerus-Opposirionen . . . zu
decoulent des valeurs lexicales des parries
den bevorzugten grammarischen Verhal
du discours . . .' (ibid.).
tenseigenschaften
'Tout comme Ia derivation syntaxique se
Klasse von Lexemen' (ibid.).
deroule a l'interieur d'une seule et meme valeur
lexicale
(par
exemple
I4
dieser
semanrischen
We would accordingly have to observe
adjecrif
terminological distinctions berween the
epithete-adjecrif anaphorique, Ia valeur
rwo levels: e.g. on the semantic rype level
lexicale restant Ia meme), tout ainsi Ia
there would be 'individuals' and 'pro
derivation lexicale suppose que le mot
perties', on the conceptual level there
base et le derive sont idenriques quant a leur
foncrion
(Kurylowicz
syntaxique
I 936: 45 ).
primaire'
I5
would be 'objects' and 'qualities', etc. A remark is necessary regarding the 'in region': the relation 'x in y' in itself is
One might doubt whether the direction
neuter with respect to the distinction
of derivation actually goes from 'event' to
berween external and internal region; c£
'physical object'. But it can easily be shown that this analysis is correct: all
there is an apple in the bowl vs. there are holes in the cheese. But the French prepositions
deverbal nouns formed with the suffix
for 'in' behave very much like the other
-age
prepositions, which all specify external
can denote an event, whereas not all
nouns of this rype can denote a physical object. Examples are
massage
regions. It 1s thus convenient, if not
'massage',
language 'swaying movement (of a vessel)', abattage 'felling (of trees)', rodage 'running
entirely justified, to include the in-region
I6
among the external regions. Lyons did not formulate a proper theory of prepositions. The above qualification is
in (a new car)'. In his book on morphology he remarks ·
meant to express a result of recent work
that the semantics of sufflxless derivation
on rhe lexicon of space; c£ Bierwisch
( I 988: 8 f£),
might be built upon a more elaborate semantic theory of the parts of speech, and he points out that already Hermann Paul
I7
Wunderlich
The status of a
&
Herweg
J.4). (droite, gauche) de
(forthcoming: section
X
as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(Ia foncrion lexicale restant Ia meme), est
pond a Ia forme derivee' (Kurylowicz
IO
die
es erstens vie! wahrscheinlicher, dal3 sic
syntaxique secondaire celle qui corres
9
haben,
praricien s'en riendra a Ia notion de
entraine le changement forme! de A en B
8
Bedeutungen
Jonction syntaxique prima ire (et Jonctions syntaxiques secondaires) . . .' (Kurylowicz I 936: 42). taxique d'une forme (ou d'un mot) A
7
1 98 I : I I4).
'Wenn mehrere Lexeme einer Sprache
360 Concept Types and Parts of Speech: the Lexicon of Space in French
I9
20 2I
22
23
24 The system of region adjectives of modern French seems to be relatively recent. There is some evidence that the present system has replaced a system in which internal regions were specified by prepositions or adjectives of dimension. Compounds like avant-bras , 'fore-arm', arriere-train 'rear end', bas-ventre 'lower abdomen' represent this state of evolution; cf also the occur rence of haul 'high' and bas 'low' in geographical proper names such as les
Hautes-Pyrenees , Ia Basse-Bretagne . 25 Lyons ( I 989: I S7 f) emphasizes rhar, for a
26
27
28 29
universal theory of the grammatical expression ofontological srructure, phrasal distinctions (concerning constituents, like 'noun phrase') are more relevant than lexical distinctions (concerning lexical categories, like 'noun'). This ofcourse does not affect the analysis of parr of speech systems of particular languages like French, which have a clear formal distinc tion berween nouns and noun phrases. In Lyons's theory second-order entities are 'events, processes, stares-of-affairs, etc., which are located in time and which, in English, are said to occur or take place' (Lyons I 977: 443); third-order entities are 'such abstract entities as propositions, which are outside space and rime' (ibid.). There is no general noun for the rear region, because rhe words which one would expect here have undergone semantic change, due to euphemistic usage: le posterieur, le derriere both mean 'buttocks'. cote originally meant 'ribs', bord comes from a meaning which is still present, namely the board of a boat. Lyons reasonably postulates such a domain of arbirrariness: 'That ir is difficult to draw rhe line precisely berween what counts as a discrete physical object and what is nor is unimportant . . . Ir is the lexical and grammatical srructure of particular languages that draws the line for us in the unclear instances (e.g., with respect ro the ontological status of moun tains, rivers, etc.' (Lyons I 977: 442).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I8
prepositions is doubtful. x cannot be a personal pronoun: .a droite de moi is clearly agrammarical, whereas a cotede moi is fully grammatical. One has ro use rhe possessive pronoun instead: a ma droite. In this expression droite behaves like an ordinary noun. One might conclude from rhis rhar a {droite, gauche) de x are nor complex prepositions, but prepositional phrases which contain a modifier. In any event, ir is clear that a { droite,gauche) de X have gone less far in rhe process oflexicalizarion rhan rhe other complex expressions in the present list. The use of avec as a pronominal adverb, like in le camion est parti , et moi avec 'the truck starred off, and I with ir' is con demned by purists, but is fully gram matical in spoken French. This also holds for other words in rhe present table. The use of pour as a pronominal adverb is resrricred to the expression of an opinion, i.e. to contexts in which the incorporated zero pronoun refers to an absrract enriry such as a suggestion, or a plan. The exception is parcourir 'to move within an object, reaching most or all of irs parts'. Theoretically, region concepts can also be amalgamated with the concept of loca tion. But there are only very few and rather marginal verbs which show this conceptual structure in their basic mean ing; cf desJossesjouxtent Ia route 'there are ditches along the road', un rocher surplombe Ia route 'there is a rock hanging over the road'. Bur this small group of verbs is extended by secondary readings of verbs of motion; cf (8) and (9) above. The only exception is traverser which is derived from (a ) travers. longer and le long de both are derived from rhe adjective long . survoler is a compound; it is formed following a pattern which French has almost completely given up. This whole discussion confirms an old methodological experience: rests and descriptive criteria are certainly useful, but they cannot be used without inter pretacion.
C. Schwarze 361
RE FE RE N CE S Bierwisch, M. (1 988), 'On the grammar of local prepositions', in M. Bierwisch et a/. (eds), Syntax, Semantik und Lexikon (Studia Grammatica, 29 ), Akademie Verlag, Berlin,
I-65.
Cresswell, M. J. ( 1 97 3), Logics and Languages, Methuen & Co Ltd, London. Croft, W. (1 990), 'A conceptual framework for grammatical categories (or: A taxon omy of propositional acts)', Journal of
Societe Linxuistique de Paris , 37 (1 936), 7992. Lyons, ]. ( 1 966), 'Towards a "notional" theory of rhe "parts of speech" ',journal oJLinguis tics, :z, 209-3 S · Lyons,]. (1 977), Semantics, vol. II, Cambr.idge
University Press, Cambridge. Lyons,]. (1 989), 'Semantic ascent: a neglected aspect of syntactic typology', in D. Arnold et a/. (cds), Essays on Grammatical Theory and Universal Grammar, Clarendon Press, Oxford, pp. 1 s 3 -8·6. Plank, F. ( 1 98 1 ), Morphologische (Ir-)Regulari
tiiten: Aspekte der
Wortstrukturtheorie ,
Gunter Narr Verlag, Tiibingen. Plank, F. (1 9H4), '24 grundsarzlichc Bcmcr kungcn zur Wortartcn-Fragc', Leuvense
Bijdragen , 73, 489-520.
Semantics from Different Points of View, Springer Verlag, Berlin, pp. 304-23. Schwarze, Ch. ( 1 985), ' Uscire e andare Juori: srrurrura sintartica e semantica lessicale', in Franchi de Bellis & L. M. Savoia (eds),
A
Sintassi e moifologia della lingua itaIiana d'uso: teorie ed applicazioni descrittive , Bulzoni, Roma, pp. 3 S 5-7 1 . Schwarze, Ch. ( 1 989), 'Polysemie als Pro zedur, am Beispiel von frz. a travers und chez ', in Ch. Habel et a/. (eds), Raumkon zepte in Verstehensprozessen: Interdiszipliniire Beitriige zu Sprache und Raum , Niemeyer, Tiibingen, pp. 3 1 0-38. Talmy, L. (1 985), 'Lexicalization patterns: semantic structure in lexical forms', in T. Shopen (ed.), Language Typology and Syn tactic Description , Vol. 3, Cambridge Uni versity Press, Cambridge, pp. 5 7-149. Wienold, G. & Ch. Schwarze ( 1 989), 'Lexical structure and the description of motion events in Japanese, Korean, Italian and French', Konstanz, 1 989 (- Universitar Konstanz, Department of Linguistics, Working Paper No. s). Wunderlich, D. & M. Herweg (forthcoming), 'Lokale und Diektionale', in A. von Stechow & D. Wunderlich (eds), Handbuclz Semantik, De Gruyter, Berlin.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Semantics, 7, 245-79.
Kurylowicz, J. (1 936), 'Derivation lexicale et derivation syntaxique: contribution a Ia theorie des parties du discours', Esquisses linguistiques, Wrodaw-Krak6w, 1 960, 41so; presented originally i n Bulletin de Ia
Schwarze, Ch. ( 1 979), 'Reparer-Reparieren : a contrastive study', in R. Bauerle eta/. (eds),
Journal ofSemantics 8: 363-402
© N.J.S. Foundation (1991)
A Critical Examination of Two Classical Approaches to Aspect M I C HAEL
HERWEG
University ofHamburg
Abstract
1
THE S C O P E O F T H I S P A P E R
Any formal semantic theory of temporal expressions needs a viable account of the fundamental aspectual distinction between perfective and imperfective sentences, which, as a semantic opposition, shows up in most if not all languages of the world.1 Perfective sentences express propositions about durative or non durative events: they say that in a given period of time, an event of a certain type occurs in its entirety. Imperfective sentences express propositions about states (in the global sense, including e.g. habitual states and states of iteration) or ongoing processes: they say that at a given time a state holds or a process is going on. This correspondence between the aspects and the sorts of situations described is widely accepted; therefore, I will do without further justification here.2 Moreover, since it was demonstrated by Galton (1 984) that, considering those properties relevant from the point of view of aspectual semantics, there is no need to distinguish between sentences about states and sentences about processes-the former being sentences about 'states of no change' and the latter being sentences about 'states of change'-we may use as alternatives the terms
event sentence and state sentence to refer to sentences ofthe two respective aspects. The reasons for demanding an account ofaspect within a theory of temporal
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This paper discusses the merits and shortcomings of the two leading paradigms in the model theoretic account of the perfective and the imperfective aspect the proposition-based approach in the tradition of Tense Logic, and the eventuality-based approach in the tradition of Donald Davidson's semantics for event expressions. It is shown that neither approach may claim general validity for their respective format ofanalysis, as their theoretical means suit one particular aspect only: those · of the proposition-based approach are confined to the imperfective aspect, and those of the eventuality-based approach are confined to the perfective aspect. Contrary to what is suggested by their advocates, neither format of analysis can be generalized to account for the other aspect. Rather, it is imperative to have a theory which integrates the two complementary approaches to one unified account. The basic features of such an integrated theory are outlined in the final part of the paper.
364 Two Classical Approaches ro Aspect
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
expressions in general are manifold; let me here name but a few. In the first place, a number of temporal adverbials and conjunctions are restricted to argument expressions of one particular aspect; this must be reflected in their semantic representations. For example, time-span adverbials like in one month are restricted to event expressions, whereas durational adverbials like for one month combine with state expressions only (Peter wrote a book in one month vs. •Peter wrote a bookfor one month -Peter was writing a bookfor one month vs. • Peter was writing a book in one month ; see e.g. Dowty 1 979; see also below, section 2). In the second place, the temporal reference of the tenses may depend on the aspect of their argument expressions. Well-known examples are the various forms of non-actual interpretations of the present tense when it is applied to an event expression. Consider, e.g., the future time reference of the present tense in German (morgen kommt Peter nach Hamburg 'tomorrow come-PRES Peter to Hamburg')3 and the habitual interpretation of the present tense in English (Peter drives his car to work , c£ Peter is driving his car to work). Finally, event sentences and state sentences have different effects on the temporal sequencing of the situations reported in particular types of texts and discourse. Most notably, only event sentences bur not state sentences are prone to give rise to a progression of the internal time of a narration, to which the reported course of events is linked (see. e.g. Partee 1 984). In the following, I will touch upon the first of these special topics only, since the purpose of this paper is a more fundamental topic in the semantics of the aspects: the question of the proper account of events and states as the subject matters of perfective and imperfective sentences. To this end, I will examine the particular accounts of the aspectual distinction between event expressions and state expressions within two competing paradigms in the model-theoretic approach to the semantics of temporal expressions. The bulk of theories which dominate the research in temporal semantics may globally be divided into two groups: on the one hand those approaches which align their basic pattern of analysis with the im perfective constellation, and on the other hand those approaches which shape their account of the aspects according to the perfective constellation. Theories of the imperfective paradigm (e.g. Reichenbach 1 947; Bennett & Partee 1 972/ 1 978; Bauerle 1 979; Dowty 1 979, 1 982; Cresswell 1985; Nerbonne 1 98 5; Fabricius-Hausen 1 986; Hinrichs 1 987; Ballweg 1988) folio� the tradition of classical Tense Logic (see e.g. Prior 1 957, 1 967; Rescher & Urquhart 1 97 1 ). Their basic picture is that sentences express that a proposition is true or not at a given time, which in general is considered as a time interval. The proposi tion is intended to represent the event or state the sentence is about. The role of the tenses is to specify the temporal relation between a time of evaluation, e.g. the time of utterance, and a time at which the proposition is true-in short, a truth interval of the proposition. Event sentences and state sentences
M. Herweg 365
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
are distinguished by characteristic heredity features of the respective proposi tions with regard to their truth intervals: propositions about states possess the subinterval property, whereas propositions about events are subject to the anti-subinterval property. Theories of the perfective paradigm (e.g. Wunderlich 1970; Saurer i984; and all approaches within the framework of Discourse Representation Theory (DRT); see e.g. Kamp & Rohrer 1983; Partee 1984; Reyle 1987, Bauerle 1 988) follow the Davidsonian tradition of event-based semantics (Davidson 1 967, 1970). Their basic picture is that sentences express the occurrences of events in time-or rather, adopting the terminology suggested by Emmon Bach, the occurrences of eventualities in time (Bach 1 986), since not only events in the narrow sense but also quantities ('bits') of states and processes are to be covered. The role of tenses is to specify the relation between a time of evaluation and the time of the reported eventuality. The aspectual distinction between event expressions and state expressions is captured by a distinction among the referential properties of type-predicates about eventualities, namely the properties of heterogeneous reference (for event expressions) and homogene ous reference (for state expressions). I will label the two types of theories the proposition-based approach and the eventuality-based approach , respectively. Both approaches generalize their basic scheme of analysis, which is essentially directed toward one particular aspect, to the opposite aspect. I will argue that both types of theories can give adequate analyses of their target aspect only (c£ Lohner 1 988). Therefore, they cannot be regarded as theories of temporal expressions which may claim any general validity. Neither can the notion of an event (or eventuality, if you like) occurring in time explain the semantics of the imperfective aspect, nor can the notion of a proposition being true at a time account for the semantics of the perfective aspect. I will demonstrate in particular that neither theory alone has the theoretical means necessary to facilitate an adequate formal reconstruction of the different logical status of the subject matters of perfective and imperfective expressions, namely situations of the sorts 'event' and 'state', which reveals itself from the semantic properties of the two aspects. The remainder of this paper is organized as follows. In section 2, I will sketch the semantic properties of event sentences and state sentences relevant for the ensuing discussion. From these the logical features of events and states will be determined which will serve as criteria for the evaluation of the proposition based and the eventuality-based approach to the aspects in the following section. Section 3 forms the core part of the paper. Here, the merits and short comings of the respective treatments of aspect in the two classical approaches are discussed in some detail. The discussion will result in the demand for a the ory which integrates the two complementary approaches into one unified account, providing the theoretical tools necessary for the appropriate formal
366 Two Classical Approaches to Aspect
treatment of both aspects. The most relevant characteristics of such a theory, which is set out in more detail in Herweg (1 990, 1 992), are briefly sketched in section 4·
2 T HE S E M A N T I C P R O PE R T I E S O F E VE N T SE N TE N C E S A N D S T ATE SE N T E N CE S
( I)
a.
b.
c. (2)
a.
b.
c. (3)
a.
b. c.
Peter ran two miles in ten minutes. *Peter ran two miles for ten minutes. Peter ran two miles three times. *Peter was running in the park in ten minutes. Peter was running in the park for ten minutes. *Peter was running in the park three times. *Peter stood on the lawn in ten minutes. Peter stood on the lawn for ten minutes. *Peter stood on the lawn three times.
Event sentences like Peter ran two miles are compatible with time-span adverbials like in ten minutes but not with durational adverbials like for ten minutes . Furthermore, they may be combined with count adverbials like three times (see ( I )). For state sentences, such as sentences reporting ongoing processes like Peter was running in thepark (see (2)) and states like Peterstood on the lawn (see (3) ), it is just the other way around: they combine with durational adverbials, but not-or at best marginally-with time-span adverbials, and they do not go together well with count adverbials.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The most notable semantic properties o f event sentences and state sentences, which will serve as clues for determining the logical properties of their respective subject matters, i.e. events and states, can be ascertained by exploring the combinatorial potential of these expressions with regard to time-span adverbials, durational adverbials, and count adverbials. The results will give us the background against which the competing theories of the aspects will be evaluated in the following sections. Since there has been extensive discussion in the literature (see e.g. Vendler I967, Verkuyl I 972, Mourelatos I978, Dowty I 979, Krifka I 987 and many others), let me simply give an overview ofthe well established facts and my interpretation of these (for a more detailed discussion see Herweg 1 990, 1 991. As far as possible, my interpretation of the linguistic facts is stated in terms which do not already show a bias towards one of the competing theoretical accounts to be discussed in the ensuing sections. The following English sample sentences provide us with the necessary linguistic data:
M. Herweg 367
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I will say more on the combination of state expressions with count adverbials below. A number of cases seem to argue at first sight against the position that state expressions do not combine felicitously with count adverbials; consider e.g. Yesterday, Peter was in the park three times . Furthermore, the combination of a state expression with a count adverbial (and time-span adverbials) is often not strictly excluded but rather merely acceptable to a limited extent. All these combinations will be shown to involve a non-compositional reinterpretation of the state expression to the effect that the expression does not characterize a state but rather quantities of that state, the number of which is specified by the count adverbial. This reinterpretation results in an expression which exhibits all the logical properties of an event expression. Therefore, quantities of a state must be seen as events, in fact as events which consist in chat the state holds for a while. The reinterpretation is induced by the context. It is in particular forced by the aspectual requirements of count adverbials, which can be directly traced back to the nature of counting (see below). So, what we are really going for are the semantic properties of expressions which have a state or event meaning in a given context. The commonly cited distinctive features of events and states are the features of telicity and non-telicity. Events possess an inherent goal or set terminal point (Krifka 1987). Only when this set terminal point is reached, can the event be said to have occurred. The feature of inherent telicity gives events a definite temporal bound, which allows them to be temporally located within a time span (see ( 1 a)). In contrast, states are inherently a relic. A state may be said to have held (or a process to have been going on) irrespectively of any constitutive terminal point having been reached. Ofcourse, most states actually end at some rime, but this end is not a constitutive part of the state, but merely an extrinsic property. This lack of inherent telicity disallows states to be assigned a definite temporal bound. Therefore, states cannot be located within a rime span (see (2a) ). Due to their relic character, each event can be assigned a definite time, its unique time of occurrence: no proper part of the overall rime occupied by an event is also a rime at which the event occurs (Vendler I ¢7). In contrast, the relation between states and the rimes at which they hold is indefinite: if a state holds over some period of rime it also holds at each part of this period. That a state holds is a continuous property of an uninterrupted stretch of time; this property holds for all parts of that period.4 We may sum up this feature by saying chat states are homogeneous with respect to the time at which they hold. In contrast, events may be said to be heterogeneous with respect to the time they occupy. The different modes in which states and events are related to their times show up in the interpretation of the tenses in state sentences and event sentences. Let me illustrate this with the example of the past tense.5
368
Two Classical Approaches to Aspect
(4.) a. The book was on the table. b. The book was on the table, where it still is now.
(s) a. Peter put the book on the table.
b. *Peter put the book on the table, where he is still putting it.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In the case of states, only some indefinite time at which the state holds is located in the past, which leaves it open whether or not the state stiil holds at the time of utterance. (4b) shows that there are no semantic implications as to whether the state actually ceased to hold before the time of utterance or still continues to hold. Imagine (4a) as an answer to someone asking Where didyou last see the book? with the intention of finding out where the book is now. By answering (4a), the person asked makes a statement about a past time only. It is neither implied that the book is still on the table nor that it is not. Due to their indefinite relation to the times at which they hold, it is an essential characteristic of state expressions in the past tense that the actual duration of the state beyond the past time in question is simply left open. In contrast, since events, due to their definite relation to the time they occupy, possess a unique time of occurrence, the past tense locates the whole time occupied by an event in the past (see (sa) ). This is shown by (sb), which is unacceptable for semantic reasons. The proper treatment of the aspects within the setting of a semantic theory of temporal expressions must reflect this difference in the interpretation of the tenses in state sentences and event sentences. I will come back to this demand in the evaluation of the proposition-based and the eventuality-based approach to aspect. The distinction between two particular modes in which situations may be related to time explains the complementary distribution of state sentences and event sentences in the context of durational adverbials and time span adverbials. Durational adverbials like for ten minutes are expressions of duration measure ment: they specify a lower bound for the duration of a situation, i.e. they fix the minimum amount of the time the situation occupies. Durational adverbials are downward-entailing but not upward-entailing (c£ Krifka I 987): if someone was running in the park for ten minutes, he or she necessarily was also running in the park for nine, eight, . . . minutes; whether or not he or she has been running even for eleven, twelve, . . . minutes is left open by the duration specification. Due to this feature, durational adverbials are confined to situations which are homogeneous with respect to the times they occupy. The applicability of a durational adverbial requires that the time in question can be arbitrarily narrowed down to quantities each of which has the same property as the entire time, in fact the property that it is a time of the situation in question. Clearly, only states allow this kind of division of the time at which they hold. Since an event is heterogeneous with respect to the time it occupies, its unique time of occurrence cannot be divided into distinct quantities which also are times at
M. Herweg 369
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
which the event occurs. Of course, an event of Peter running two miles as a whole can take ten minutes of time. But this must be expressed by means of a time-span adverbial like in ten minutes , which specifies an upper bound for the time assigned to a situation. Time-span adverbials are upward-entailing but not downward-entailing: if someone ran a mile in ten minutes, he or she necessarily ran a mile in eleven, twelve, . . . minutes as well; the time-span specification does not however imply anything about whether or not the person managed to run a mile in an even shorter time. Due to this feature, time-span adverbials are confined to situations which are heterogeneous with respect to the time they occupy: there must be a unique time of the situation, i.e. a time which cannot be divided into proper sub-times which in turn are times of the situation in question. Only events show this definite relation to their times. Count adverbials put the same semantic restrictions onto their argument expressions as time-span adverbials. The combinatorial restrictions of count adverbials to event sentences can be shown to follow from the requirement of heterogeneity that count adverbials put on their argument expressions. Counting means to specifY the number of possible applications of a predicate to entities which satisfY that predicate.6 Thus, there are two shirts in the wardrobe means that the predicate 'shirt' can twice be applied to distinct entities in a limited area which is specified by the local adverbial. In order to make counting possible, the predicates must provide a criterion ofindividuation for the entities which satisfY the predicate, since individuality is a prerequisite for countability. Only predicates of a special class provide such a criterion, namely predicates which in the philosophical tradition have been called sortal predicates (e.g. Strawson 1 959) and which in the terms of modern formal semantics usually are called heterogeneous predicates (cf Krifka 1 987, 1 989 on 'quantized predicates'). As Frege ( 1 8 84: 66) put it, these predicates do not apply to any part of an entity in their extension, and they separate each entity they apply to from any other ' entity. Thus, these predicates individuate the entities in their extension. As a matter of fact, something is an individual only in virtue of being subject to a heterogeneous predication. From their countability we conclude that events are individuals from a logical point of view. The heterogeneous predicate which secures their logical status as countable individuals is provided by the event sentence itself Let us call the sentence minus its temporal specifications by means of tense and temporal adverbials a radical. I will indicate radicals by using bare forms of verbs without the inflectional component (e.g. Peter run- two miles , Peter be running in thepark, Peterstand- on the lawn ) We may then say that the radical of an event sentence-in short, the event radical -is a heterogeneous type-predicate about events, which means that it does not apply to any proper temporal part of the events in its extension. As such, it includes the criterion of individuation and counting for the events it applies to. Each type-predicate about events-in short,
370 Two Classical Approaches to Aspect
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
each event-type predicate-carries its own criterion of individuation and counting. So, one and the same situation may be viewed as being composed of a different number of events, if it may be characterized by different event-type predicates which carry distinct criteria of individuation and counting. Take e.g. a situation in which a musician tunes his or her violin. This situation may be viewed either as one event of the type 'a musician tune- his/her violin' or as being composed of four events of the type 'a musician tune-a string of his/her violin'? As a state holds homogeneously over periods of time, it must be considered to have an indefinite number of proper parts which also are instances of that state. This property is clearly incompatible with counting, since one could not make sense of specifying a definite number of possible applications of a predicate to something which has an indefinite number of parts, each of which can be characterized by the same predicate. In contrast to heterogeneous event radicals we classify state radicals, i.e. radicals of state sentences, as homogeneous predicates , which means that they are closed with respect to the part-of relation defined on the entities in their extensions: if a homogeneous predicate applies to an entity, it applies to all of its parts as well. Due to being homogeneous predicates, state radicals do not involve any criteria of individuation. Therefore, a state radical cannot individuate the entities in its extension. So, states must not be considered as situational individuals from a logical point ofview; this implies that they cannot be counted. It is exactly this property of homogeneity of predicates about situations which durational adverbials require from their argument expressions (see above). So, the essential opposition of criteria! contexts by which the logical properties of states and events can be determined is the one between count adverbials and durational adverbials. Count adverbials are restricted to hetero geneous argument expressions; by contrast, durational adverbials demand homogeneous argument expressions. These aspectual requirements are based directly in the nature of counting and duration specification (i.e. duration measurements), respectively. Events are situations characterized by a hetero geneous predicate; as such, they are individuals, which means that they can be counted but cannot appear as arguments to lower-bound duration speci fications. States are situations characterized by a homogeneous predicate; as such, they must not be treated as individuals from a logical point of view. Therefore, states cannot be counted; they can however be specified with regard to their minimum duration. The distributional pattern of event expressions and state expressions in the context of time-span adverbials does not give us any substantial new evidence. Time-span adverbials demand situations which are assigned a definite time; a definite time can only be assigned to situational individuals, i.e. events.8 Although a state is not an individual from the start, it may be turned into an
M. Herweg 3 7 1
individual by providing it with an external criterion of individuation. We do so by splitting up the time at which a state homogeneously holds into continuous stretches of time at which the state holds just for a while. Thus, we single out a number of quantities of a state (Krifka 1 987). One way of quantizing a state is to explicitly assign a duration to the state (c£ the above exposition on duration specification). The resulting predicate is a heterogeneous one, as can be seen from the count construction (6), which presupposes that no proper temporal part ofPeter's running in the park for ten minutes is also Peter's running in the park for ten minutes. From this we conclude that quantities of a state are simational individuals, i.e. events. Another, implicit way of quantizing a state is achieved by taking the maximum period at which the state continuously holds. Let us call this period a phase of the state (cf Lohner 1 988). In German and English, this way of implicitly quantizing a state is totally conventionalized for copula construc tions like (7a) and (7b). In German, the perfect form of the auxiliary can freely be used as an alternative to the simple form of the auxiliary (see (7b) ); in English, the use of the perfect form is more constrained; it is e.g. excluded by particular temporal adverbials. (7) a. Im letzten Jahr war Peter dreimal in London. (in-the last year be-PAST Peter three-times in London) b. Peter was in London three times last year. c. Im letzten Jahr ist Peter dreimal in London gewesen. (in-the last year be-PRES Peter three-times in London be-PARTPERF) d. Peter has been to London three times in the past two weeks/*last year. The sentence radicals underlying the tensed sentences in (7) are predicates about occurrences of phases of the state in question. These radicals are heterogeneous predicates, i.e. event radicals. Otherwise, we could not make sense of the count construction since there would be no way to justify the claim that Peter was in London exactly three times within the last year and not indefinitely often, namely that indefinite number of times that is the number of parts the state possesses. So, the occurrence of a phase of a state is an event (cf Partee 1 984 on 'holistic states'). This concludes our survey of the feamres of event sentences and state sentences and the corresponding sorts of simations relevant for the following evaluation of the two predominant model-theoretic accounts of the aspects introduced at the outset. It has often been noted in the literamre (see Mourelatos 1 978; Bach 1 986; Krifka 1 987, 1 989, to name but a few) that the distinction between events and states and the corresponding predicates, i.e. event radicals and state radicals,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(6) Yesterday, Peter was running in the park for ten minutes three times.
372 Two Classical Approaches to Aspect
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
parallels a distinction among concrete entities and their predicates, namely the distinction between objects and masses and the corresponding predicates, i.e. count nouns and mass nouns. Objects may be counted but not be arguments to downward-entailing measurement specifications. So we have two shirts but *two pounds ofshirt , since the noun shirt is a heterogeneous predicate. Because of this, a shirt is a countable individual. As such, it cannot be divided into equal units, each with a weight of one pound, which in turn are all shirts. By contrast, masses may be subject to measurement specifications but cannot be counted, which gives us two pounds ofcotton but *two cotton (s ). In the same way a state can be turned into an event by providing it with an external criterion of individuation which defines a quantity of the state, a mass can be turned into an object by means of quantizing it; consider pound ofcotton and bale ofcotton . These parallels have especially been exploited within the eventuality-based approach to aspect, which employs theoretical notions originally introduced in Link's model-theoretic semantics of count nouns and mass nouns (Link 1 98 3). In particular, the eventuality-based approach subscribes to a domain of quantification comprising events and quantities of states in very much the same way Link's theory makes use of a domain of objects (the denotations of count terms) and quantities of stuff (the denotations of mass terms). Thereby, the eventuality-based approach fits into a fairly recent development within the model-theoretic semantics of natural language. Instead of dwelling upon the long prevalent model structures known from classical intentional approaches to natural language semantics in the tradition of Max Cresswell (Cresswell 1973) and Richard Montague (Montague 1 974), which by and large get by with set-theoretic constructions out of standard objects, truth values, and possible worlds, the theorists participating in this current development employ domains of entities of a variety of sorts with a rich internal structure, including domains of non-standard individuals like quantities of masses, times, events, and quantities of states. By contrast, the theoretical tools used in the proposition based approach are much more conventional in that one exclusively relies on the traditional notion of a proposition, for which a semantics in the style of ordinary modal logics is given. So, instead of assuming a domain of quantifica tion for abstract entities like events and states, natural language talk about events and states and their location in time is reconstructed solely by the notion of the truth of propositions at times. I will now present and evaluate in some detail the treatments of the aspects proposed in those two leading theoretical paradigms within formal temporal semantics. As I indicated above (section 1 ), I will demonstrate that neither approach may claim general validity for their respective pattern of analysis. Rather, each of the two paradigms is essentially confined to one aspect and obtains inadequate results for the opposite aspect. The major difficulty for both is the formal reconstruction of the difference between events and states
M. Herweg 3 7 3
3 T W O C L A S S I C A L A P P R O A C H E S TO A S P E C T 3.1
The proposition-based approach
For quite some time, most if not all formal approaches to temporal semantics have subscribed to various conservative modifications of the inventory of classical Tense Logic (see e.g. Prior 1 957, 1 967; Rescher & Urquhart 1 97 1 ) in order formally to represent temporal expressions, and give a model-theoretic semantics to them which takes into account the aspectual distinctions discussed in section 2. These theories in the tradition ofTense Logic constitute what I call the proposition-based approach to temporal semantics. The modifications of the original system of Tense Logic are especially motivated by linguistic considerations. Since there has been extensive discussion in the literature (e.g. Bennett & Partee 1 9721 l 978; Dowry 1 979, 1 982; Galton 1 984), I am not going to bother with a comparison of the original system and its various successors. Instead, I will simply list the features of what appears to be the core version of a proposition-based approach to temporal semantics on which most theories working within that paradigm can currently agree. The basic idea of the proposition-based approach is the following: natural language sentences express that a proposition, which represents the event or state the sentence is about, is true or false at given times of evaluation. These times are formally intervals which represent extended periods of time. The times at which a proposition is true may be called its truth intervals. The tenses
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
regarding their logical status. The proposition-based approach will turn out to reconstruct successfully the logical properties of states as non-individuals. It fails, however, to treat events as individuals of their own, in particular as individuals logically independent from the times they occupy. By contrast, the eventuality-based approach gives a correct analysis of events. However, it cannot avoid treating states on a par with events, since it is irrevocably confined to quantities of states, which are individuals under a logical perspective. I will present the general characteristics of both accounts from a fairly global point of view. Specific proposals by individual representatives will be con sidered in detail only if they form substantial progress compared with the mainstream within their respective framework. I will pay special attention to the interplay between the proposed semantics of the tenses and the aspects. The reason for this is that in judging the two competing approaches, we will have to consider how they cope with the peculiarities of tensed state expressions and event expressions demonstrated above with the sample sentences (4) and (s). Especially in the eventuality-based approach, considerable effort has been made just to achieve the correct analyses of state sentences in the past tense (see below, section 3.2).
374 Two Classical Approaches to Aspect
( I ) The book was on the table. (2) The tenseless proposition expressed by the sentence radical the book be- on the table is true at a time interval which is earlier than the time of utterance (and which is included in the time of reference). A very simple and elegant formal system along these lines is the one proposed by Hinrichs (I 987). In the following, I confine myself to presenting only those features of the system which are relevant for our purpose. In Hinrichs' system, tenseless and tensed propositions are evaluated with respect to a time of utterance t and a time of reference r. PAST is an operator on tenseless propositions p, representing the past tense. Besides locating the time of evalua tion t ' for the argument proposition p before the time of utterance t, this opera tor posits t ' within the time of reference r, which in general is contextually specified. Adverbials like yesterday , which serve to fix the time of reference, are also treated as propositional operators. YEST shifts the time of reference to the day before the time of utterance. The truth conditions are defined as follows; < is the relation of temporal precedence;
t and t '