JOURNAL OF SEMANTICS
INTERNATIONAL JoURNAL fOR rns iNTERDISCIPLINARY STIJDY OF rnB S EMANTics oF NATIJRAL LANGUAGE AN
MANAGING EDITOR: Plrru Bosai (IBM Sctennfic Centre, Hetdelberg and Uruversity of Osnabnick) ASSOCIATE EDITORS: MAHFuD KRIFIA (Uruvemuy ofTexas, A wn n) Roa VAH DBII SAHur (Uruvernty ofNtJ megen) REVIEW EDITOR: TIBOII Kiss (IBM Scteonfic Centre. He�delberg) ASSISTANT EDITOR: BIANL\ BU$CHIIECII:-WOLP (IBM Scteonfic Centre, He�delberg) EDITORIAL BOARD:
N. /um!ll (Uruvemty ofTexas, A wnn) R BARnCH (Uruventty of .Amsterdam) M. BtEIIWISOI (MPG and Humboldt Uruvernty
Berlm)
B. BocuaAuv (Apple Computet Inc) M. BoilllO (U ruvern ty ofToulouse) G. BROWN (Uruvernty ofCambndge) K. BROWN (Uruvernty of Essex) G. CHIBIICHIA (Uruversity of Mtlan) 0. DAHL (Uruvernty ofStockholm) S. C GARIIOD (Univernty of G�) B. GE1J11T1 (Uruvernty of Osnabriic.k) M. HnWBG (IBM Saennfic Centre, He�delberg) L R HoRN (Yale Uruvemty) j.JACOJIS (Uruvemty of-Wuppertal).-- P.N.joKHSON-LAuw ( Pnnceton Uruven:tty) M. K.utBvAMA (SRI Internanonal, Menlo Park)
H. KAwP (Uruvemty ofStuttgm) S l..aVINSON (MPI NtJmegen) S UI11NH11 (Uruvemty ofDiisse ldorf) S111joHN LYONS (Uru=ty ofCambndge) A. MANASI"BII-R..uum (Wayne State Uruvemty) J. McCAWLEY (Uruvenuy ofClucago) M. MoBNs (Uruvernty ofEdwburgh) F. J Pm.umEll (Uruven:t ty of Alberta) M. PmiAJ. (Uruvernty of Surbnicken) T. SANPo110 (Uruvern ty ofGlasgow) R SoiA (Uruvernty of .Amsterdam) A. voN STI!Oiow (Uruvemty ofTiibwgen) M. STEEDMAN (Uruvernty of Pennsylvarua) W. WAJ�I.STD (DFKJ, Surbriick en ) B.-WBIIBBII (Uruvernty.of.Pennsylvarua) . H. ZIIBVAT (Uruvemty of .Amsterda m) T. E. ZtMMI!1lMAHH (U ruven� ty ofStuttgart)
EDITORIAL ADDRESS Journal ofSemanncs, c/o DT P. Bosch, IBM Germany Sctennfic Centre, Vaogerowstr. 18, 0-69• 1 s He�dd� Germany. Phone: (4�1:;11-) S9-.f;IS 1/4-483. Tekfax: (.f�lll-) S9-JWO. Enwl: boschCvneubm.com, bssCIImeubm.com
C Oxford Uruvernty Press ngha reserved; no part of dus publtanon IIUY be reproduced, stored 1n a rcmcval sy1<em. or cransnuttfil m any form or by any mnm, eleccrom.c, rnedwuc:.J, phorocopymg. =rdtng. or otherwut Without e1ther the pnor wntten penrumon of the Pubbshers. or a 1.tcence pemumng r=ncted copytng wued m the UK by the Copyngbt Lu:erutng Agency Ltd, 90 Toctenham Court Road, London W1P 9HE. or m tbe USA by !he Copynght deannce Onter, All
�7 Congreu Street, Salem, Mass o1970-
J�rrwl of� (ISSN 0167 S IJJ) u publubed quarterly m February, Apnl. August. and N�mber by Oxford UDJventry Prtu, Oxford UK. Annual mbsatpaon u US$1w per year. frvrMI of&-"1la u dismbuted by Virgm Mulmg and Dumbuoon, I 0 Camprown Road, lrrutgton, New jeney 07 I I 1-1 I OS, lJSA. S«ond class postage p:ud at Ncwuk, N"'Jeney, USA and at addmonal entry poma.. US POSTMASTER: send address corncoons to ]OtUMI of &-"lies, c/o Vugm � and Dumbuoon, 10 Camptown� lmngton. New Jersey07111-1105, USA. The wue date u No...,mber.
JOURNAL OF SEMANTICS Volume
12
Number 4
CONTENTS MARKUs EGG
The Intergressive as a New Category ofVerbal Aknonsan
311
EMIEL KRAHMER AND REINHARD M USKENS
Negation and Disjunction in Discourse Representation Theory
3s7
INGRID KAuFMANN
0- and 0-Predicates: A Semantic Approach to the Unaccusanve-Unergative Distinction
377
C Oxford Uru�rnry Press 1995
Juu,} of&mMttus IJ.. 3II-356
The Intergressive as a New Category ofVerbal Aktionsart M A RKUS
EGG
Instituufor Logic & Linguistics, mM Drotschland
Abstract The coprc of
tlus paper are prechcatel; hke 15
tc play a sonata or tc cough , whose classificanon m
unsansfactory I wtll present an augmentation of Dowty's
system of aknonsart to clas51fy these prechcates adequately. Tlus revised classilicanon employs can be evaluated With respect to extended penods of orne only),
the features 'mterval-based' (
-
'bounded' and 'teltc'. Roundedness and tehmy are different features. Bounded predrcates apply to limited stretches of time only, teltc predrcates mtroduce ordered (tehc predrcates
are
a
change of state. The features
are
bounded, bounded predrcatel; are mterval-based) and disnngmsh
four mutually non-overlappmg groups of aknonsarten: state, process, mtergressrves and change. Intergr=ve predrcates are represented as m Dowty (1979), Le. more complex predrcates are analysed
as
less complex ones
hnked
by surrable operators. Two mte�rve
operators account for the whole range of mtergressrve predrcates. The close parallel berween the classrficaoon of akootlS:Irten and the cla5Slficacion of noJiliruU expressiOns carries over to the d!Snnction of lntergressrve and change predrcates. Next, an extensron of one mtergressrve operator to non-temporal domains Will be sketched and. finally. I will show that the assumpnon of an mtergressive aknomart has Wide repercussroru m other semannc fields.
1 I N T RO DUC TIO N Most predicates1 (the term
is meant
to include verbs, verbal projections, and
sentences) can be assigned to an aktionsarc in the classtfications of aktionsarten as proposed in the hteramre. By the term aktionsarc I refer to inherent characteristics of predicates that determine temporal properties of the entities denoted by those predicates. However, there is a residuum of examples that cause problems for these classifications. These predicates are the topic of the present paper. Examples of this group are: (1
)
Ariadne played a sonata
(2) Fritz coughed (3) Amelie sang for five hours Further examples are
toflash, to Juwe a bath, to talkfor a while, to run a mil(. Some
authors have considered the existence of such predicates in their classifications, bur so far none of the proposed classifications of aktionsarcen has managed to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
proposed systems of akoomarten
312 The lntergresstve
as
a New Category of Verbal Aknonsart
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
classify the members of this group satisfactorily. This is due to the peculiar nature of these predicates: in their original, not reinterpreted meaning they hold for entioes with fixed boundaries, but do not mtroduce a change of state. Therefore these two features must be distinguished. I will label the one 'telic' (-'denotes changes of state') and the other 'bounded'(-'denotes entities with boundaries'). Together with predicates that denote changes of state, predicates (I}-{3) belong to the major group of predicates denoting entities With boundaries. I will refer to the elements of this major group as 'event predicates' and their denotata as 'events'. As a general term for the denotata of predicates of every aktionsart I will use 'situation'. The difficuloes in classifying(I)-{3) hinge on the two properties tehciry and boundedness: event predicates are often defined in one of two ways, either as denoting changes of state or as denoong enoties With boundaries. If the first definition of event predicates Is employed, predicates(I)-(3) cannot be classified as they do not introduce a change of state. Such classifications(e.g. Dowry I979) must be augmented in order to include predicates like (I)-{3). Classifications that employ the feature 'bounded' to distinguish event prediCates(like Verkuyl I972 or Krifka I9892) can classify predicates like (1)-(3); they group them together with events that mvolve a change of state. But in such classifications an important diStinction is nussing: event predicates are lumped together regardless of whether they mvolve a change of state or not. There are addinonal classificanons which do not restrict themselves to one way of describing event predicates, e.g. the classifications of Moens & Steedman (I986, I988) or of Carlson(I98 1 ). However, these classificaoons, too, cannot account for all of the problematic predicates. Section 2 is devoted to the problems caused by predicates like(I)-{3) for these kinds of classifications. Because of these problems another group must be introduced into classifications of verbal aknonsart. I will extend Dowry's system so as to include this group. The extended system disnnguishes boundedness from tehciry. Therefore it includes a new bounded yet atelic aktionsart, which is called 'intergressive'. This aktionsart comprises examples (I)-{3) and fills in the gap between tehc and bounded predicates and atelic and unbounded predicates.3 Having ouclmed the formal apparatus on which the representation of the findings of this paper is based in section 3, I intend to establiSh a classification of verbal aktionsart in terms of three features in section 4· Two have already been mentioned; the third one is 'interval-based' (- evaluable with respect to extended periods of time only). They build up a hierarchy (possessiOn of 'higher' features implies possession of 'lower' ones) and group predicates into four mutually non-overlapping classes, namely state, process, intergressive, and change predicates. It will be shown that the definitions of the features in this paper logtcally entail the hierarchical relations between them I Will also discuss the feature'punctual'(- applicable to points of time only) and Vendler's(1967)
Markus Egg 3 I 3
closely related distinction in achievements and accomplishments (see 2.1). Neither is included in the proposed system of aktionsarren. The aim of section 5 is to formulate an appropriate semantics of mtergresstve predicates which carries the features attnbuted to this aktionsart. This enterprise will be conducted in the spint of Dowty's 'aspect calculus'. The meanings of predicates are not regarded as unanalysable wholes but broken down into more basic predicates linked by smtable operators. Thus, an operator charactenstic of (and part of) the semantic decomposmons of mtergresstve predicates must be defined. Proposals of Galton I984) and Herweg I990) for
(
(
such an operator Will be discussed. It wt!l be demonstrated that two intergressive operators are needed.
(
that it IS posstble to extend the domain of an intergressive operator to non temporal predicates. Finally, the modification of intergressive predicates by
again wt!l illustrate that the acknowledgement of an intergressive aktionsart has repercussions elsewhere in the semanocs of a language.
2 I N TE RG RE S SIVE P RE D I C A TE S I N O T HE R CL A S S I F I C A TIO N S O F AK T IO N S A R T Before presenting yet another suggesnon for a classification of akoonsart, I will first motivate this undertaking by shoWing the problems intergressive predicates pose for other systems of classification. As these problems are different for different kinds of classification, each kind of classificanon Will be discussed separately.
2.1
Classifications that describe event predicates as introducing a change of state
If one tries to account for intergressive predicates in terms of classifications that do not assign these predicates a group of their own, one must-due to their boundedness-fit in intergressive predicates with the other bounded predicates. This IS not possible if one characterizes all bounded predicates in terms of change of state, which Will be shown for one member of this group, the Vendler- Dowty classification as proposed by vendler (I¢7) and elaborated by Dowty (I979). It distinguiShes four aktionsarten: state (completely homo geneous, like
to know, to be in the pub, and to love), process (homogeneous to a
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In section 6 I will show that the parallel between the classifications in the nominal and the verbal domam as dtscussed by Mourelatos (I978), Bach I986), and others is even closer if the new classification is adopted. Secoon 7 shows
314
The Imergress1ve as a New Caregory of Verbal Aknonsart
(4) Question: Was Nigel in the pub for five hours? Answer: Yes, he was in the pub for five hours, in fact, he was there for six hours
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
certain extent, e.g. to run , to talk, and to rain), accomplishment (extended change of state wuh a preparatory process causing the change, like to build a house, to go to thepub, and to wash one's car), and achievement predicates (simple durationless change of state, such as to win the race, to die, and tofind). Two kinds of bounded predicates are distinguished, but both involve a change of state. This classification is based on a whole battery of tests, some of which I will enumerate briefly here for purposes of illustration. The tests employed in my classification will be outlined in detail in section 4· State as opposed to process and accomplishment predicates may not occur in the progressive (except those that express a volitional act, as e.g. to watch) and as complements of toforce and to persuade and have no habitual reading in normal contexts. Activity predicates differ from accomplishment predicates by their ability to be modified by durarive adverbials like for an hour bur not by time-frame adverbials like in an hour, by their inability to be subcategonzed by to finish, and by the missing ambiguity in the case of a modification by almost. E.g. I almost washed my car is ambiguous-it can be read as 'I never undertook to wash it' or 'I started but did not complete it'-while I almost sang has only the reading that I did not sing. Finally, accomplishments but not achievement predicates can be subcategor ized by to finish and to stop and only the modification of an accomplishment with almost results in ambiguity (e.g. He almost noticed me is unambiguous). How could this system classify the sentences (I}-(3)? Let us begin with (I). Dowty classifies VPs like to play a sonata as accomplishments, which is fully compatible with the tests he employs. For instance, the VP of (I) can be the complement of tofinish. However, Dowry admits that it is difficult to assign a result state to such predicates (I 979: 69£, I 86£). He tentatively assigns them a result state-'an abstract performance object exists'. The somewhat strained nature of this mterpretation shows that it is highly problematic to subsume predicates like {I) under predicates that introduce a change of state. If predicates like (2) were accomplishment or achievement predicates, they would introduce a change of state. But as these predicates can rather be regarded as denoting interruptions of an unmarked state (LOhner I988: ISs). they cannot be subsumed under either class. Neither do predicates like (3) introduce a change of state: to singforfive hours does no more denote a change of state than to sing does. It is only a pragmatic implicature to infer from this predicate that the singing started at the beginning of the five-hour interval and stopped at the end of this interval. This implicature can be cancelled without contradiction. e.g. in (4). See Krifka (I989: I 68) and Herweg (I 99 I: 67) for a discussion of this implicature.
M:arkus Egg
31 S
In sum, if event predicates are defined as predicates that introduce a change of state (I}-(3) cannot be subsumed under this class. 2.2
Classifications that describe event predicates as bounded
2.3
The classification ofMoens & Steedman
Finally, some classifications of aktionsanen do not define event predicates in tenns of boundedness or in tenns of telicity only. As an example, I will discuss the system ofMoens & Steedman(1986, 1988). I want to show that this system, too, cannot classify all intergressive predicates. Moens & Steedman distinguish non-stative aktionsanen by a cross-classification using the features 'atomic' and 'consequence'. 'Atomicity' is their name for the feature punctuality. atomic
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Classifications of aktionsan that describe event predicates in tenns of boundedness (see ( I 2)) have no problem including mtergressive predicates. Examples for bounded predicates are to walk to the store or to eat three sandwiches. Consider, for instance, the discussion of to play a sonata and to drink a glass ofwint in Krifka ( I 989: I S 8 fi): he concentrates on the way the boundedness of such predicates can be calculated. In both cases he assumes the same relation ('SUK': a specification of the patient theta-role) between the sitution denoted by the verb and the object. SUK, which can be reconstructed as (partial) homo morphism from objects to events, makes possible the determination of the (un)boundedness of the prediate by the (un)boundedness of its patient NP. But this classification neglects the differences in tehaty (though Krifka mentions it briefly): to drink a glass ofwine but not to play a sonata involves a change of state (the disappearance of a cenain quantity of wine). The same holds for Verkuyl ( I 972) and Verkuyl & Zwarts (1992): they discuss similar examples but do not consider the question of telicity. What matters for them is only the question of boundedness. In Herweg's (1 990) framework, too, event predicates are distinguished in the same way.• The problem intergressive predicates pose for these classifications is different these predicates are not outside the range of these classifications, but the distinction between their properties of boundedness and atelicity is neglected in these frameworks. This results in two undesirable consequences: first, the evidence for this distinction is disregarded; second, if telic predicates are characterized only as bounded, pan of their meaning is neglected. E.g. eating an apple is not just a bounded event of processing an apple; it also includes the disappearance of the apple. Acknowledging two clases of event predicates (intergressive and change predicates) makes it possible to descnbe such relic predicates more closely.
3 1 6 The Incergressive
as
a Ne-w Category of Verbal Akoonsart
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
predicates apply to points of time only. Predicates with the fearure 'consequence' introduce a well-defmed rernunal point at which a change of state takes place (I986: II £).The authors characterize process predicates like to run, to swim, or to play the piano as neither atomic nor mvolving a consequence, culminating process predicates like to build a house or to eat a sandwich as involving a consequence but not atomic, culnunation predicates hke to recogniz� or to win the race as both mvolving a consequence and atomic. As puncrual predicates are a special form of bounded predicates (see section 4.2), not all bounded predicates are considered to be relic by Moens & Steedman. Hence, lt is possible to classify at least some mtergressive predicates in their system: they discuss examples hke (2) and assign them to the fifth aktionsart 'point', wruch does not appear in Vendler's claSSlfication and comprises atomic predicates without a consequence like to hiccough, to tap, and to wink. Other classifications of this kind are proposed by Smith (I99I) and Carlson (I98I). But there still remains the quesnon of how to classify predicates like the non-puncrual (I) and (3). Moens & Steedman consider (3) as culminated process predicate (I986: 2I £).However, this clashes with theu defininon of the fearure 'consequence'. To state the defininon m my terminology, the fearure 'consequence' comprises both boundedness and telicity and Identifies them with each other, ie. predicates with a positive value for the fearure 'con sequence' must have an endpoint at which a change of state takes place. While this definioon is appropriate for non-puncrual change predtates, it precludes the classification of (3): although (3) involves a terminal point (the end of the five-hour interval), there need not be a change of state at this terminal point, as (3) does not entail that Amelie stopped singing at this point (See section 5·3 for the semantics of for-adverbs). This proves that the fearure 'consequence' is not adequate for the description of tills predicate. Moens & Steedman (I988: 20) classify predicates hke (I) as process predicates. However, this contradicts their observation that process predicates may only be modified by time-frame adverbials like in an hour after a remterpretation that either mtroduces a preparatory process for the siruation denoted by the predicate (and whose length is expressed by the adverbial, e.g. in John ran in aJew minutes, i.e. he started running in a few minutes) or after a reinterpretation that introduces a terminal point that is associated With the denoted siruation (e.g. in John ran infour minu� today, if we know that John runs the same distance every day). But no such reinterpretation is necessary to understand (I). In sum, other systems of classification cannot satisfactorily classify inter gressive predicates. If event predicates are defined as changes of state, these predicates cannot be classified at all; if one regards all bounded predicates as event predicates, the difference between relic and atelic event predicates is lost. Other systems which do not completely identify boundedness and telicity, like
Markus Egg 3 1 7
the one of Moens & Steedman, cannot account for all of these predicates either. These findings do not constitute a complete disproof of these classificanons but suggest necessary extensions of these frameworks. Such an extension will be presented in section 4· 3 RE PRESENT ING T I MES AND PRED I CATES
3 .1
The representation of time
Following van Benthem (1983), Herweg (1990, 1991), Krifka (1989, 1992) and others, the structure of time is represented in a mereolog1cal model (see also Link 1983). We assume a set of temporal entities T and a join operator 'u'. This operator 1s idempotent, commutative, associanve, and complete:
(s ) a. Vt[t u t- t]
b. Vt Vt'[t u t' - t' u t] c. Vt Vt' V'[t u (t' u t')- (t u t') u t'] d. Vt Vt' 3t'[t u t' - t']
Hence, (T, u) is a complete join-sernilattice. Such a structure can likewise be assumed for the domam of the objects. (This is presupposed in the discussion of the preposition through in secnon 7.) The operator 'v' can be used for the defirution of a transitive, reflexive, and antisymmetric inclus10n relation 'k'. The sign 'C' denotes the relation of proper part. (6)
Vt Vt'[t � t'- t u t'- t'] b. Vt Vt' Vt'[t � t' 1\ t' � r·- t � t'] c. Vt[t � t] d. Vt Vr'[r � r' 1\ r' � r- r- r'] e. Vt Vr'[r c r'- r � r' 1\ r � r'] a.
(T, �) is a partial order. In addition, there is no zero element within the lattice (T, u):
(7)
--.3r Vt'(t � r')
Following Herweg (1 990: 97), points of time are defined as elements of (I, Q whose parts are always identical to themselves. Le. they are lattice elements without internal structure:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This section contains the formal defirutions underlying my formalization of my theory on aknonsarten. First I wtll mtroduce the presupposed theory of nme, which is basically a mereological theory in the sp1rit of van Benthem (1983). Then I will talk about the formalization of the semanncs of predicates and other linguisoc expressions.
318 The lntergressive
as a
New Category of Verbal Akrionsarc
(8} Vt[PT(t}- Vt'(t' � t- t'- t}] There is a {transitive, irreflexive, and linear) precedence relation '
( w)
Vt Vt'[t At'- (t < t' 1\ Vt• (t < t• 1\ t• < t'- PT(t.))}] b. Vx VP[ATOM(x.P}-P(x) 1\ --.3y(y c x 1\ P(y))] a.
Finally, we must single out the group of temporal entities that are of interest in the present paper. Being a semilatrice, (T, v) allows disjointed temporal entities. But we are concerned here with continuous temporal enrines only, hence, we restrict our attention to convex temporal entities from now on, which fall under the predicate CONV: (I 1) Vt[CONV(t}- Vt' Vt• Vt'"(t' � t 1\ t• � t 1\ t' < t'" < t•- t'" � t}) Note that it �possible to regard points of rime as basic and mduce the above lattice structure on the basis of a point structure (T, �), with T being the set of points of rime and '
The representation ofpredicates and other linguistic expressions
Predicates are represented as functors that express relations between individu als. The semantic relanon between a functor and its argument is represented in the standard way, i.e. with the help of lambda-calculus. Sansfacrion of a functor by an argument is represented as functional application.. Following the intuition of Davidson (I¢7} and others, I assign predicates-apart from the arguments subcategorized in the syntax-an additional argument that stands for the situation or eventuality involved in the predicate. This argument is the referential argument of the predicate in Bierwisch's (I988} terminology. Predicates all of whose arguments but one have been applied are called 'sentence radicals' (and written with the verb in the infinitive) as they represent sentences without temporal infonnarion and definiteness. In this paper, I interpret sentence radicals as properties of convex (con-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Ais the conjunction of precedence and adjacency. Two intervals are adjacent if all intervals (if any} in between them are points of rime (Herweg I990: 98}. A entity is mmimal or atomic with respect to P if it itself but none of its proper pans belong to the extension ofP:
Markus Egg 319
tinuous) temporal intervals. Here I cannot address the question of whether sortal differences between different kinds of predicates exist, that is, whether predicates apply to different kinds of entities. (E.g. Herweg and Krifk.a assume such a difference between event predicates and non-event predicates and between state predicates and other predicates, respectively.) But none of my conclusions depends on the ontological status of predicates and their denotata. I use a predicate logic with sorted variables. Variables for times and for objects are distingmshed. Variables for times are represented by t, t', t• and so on. Variables whose son is clear from the context or which encompass both individuals and times (like in the definition of boundedness below) are
can be singled out by second-order predicates. For instance, a very prominent distinction is the propeny of boundedness (Krifka 1992: J2):
(12)
(BD). Formally, it is defined thus
VP[BD(P)- Vx Vy[P(x) 1\ y c x-+ --.P(y)]]
In other words, bounded predicates that hold for an entity x hold for no part of x. Hence,
they do not hold for entities x is a pan of, either.
Another important feature of the logical framework employed in this paper is that it is three-valued. Following Herweg (1990, 1991), I assume that all unbounded predicates have a contrary predicate. In between intervals for which such a predicate holds and its contrary holds, respectively, there may be intermediate intervals for which neither the predicate nor its contrary is defined. Hence, the fact that an unbounded predicateP does not hold for a time t (written as '--.P(t)') may be subclassified with respect to whether the contrary
of P holds good for t (represented as '- P(t)') or whether P is undefined for t. This feature is cruaal for an adequate defirution of predicates denonng a change of state. In addition, no time may be both in the extension of a predicate P and its contrary P (propeny of contrarity).. -
4 THE REVISED CLASSI FICATION OF AKTIONSARTEN In this section, I will propose a new classification of aktionsanen in the form of an enlargement of Dowty's system. The kind of classification advocated here is an ideal ordering. As far as I know, no such system has been presented in the literature on aktionsanen up to now. (For a typology of the different types of classifications, see Verkuyl 1989.) The characteristic of an ideal ordering is that its elements are ordered by a group of features which each partition the set of elements in two (according to whether the feature applies to an element or not).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
represented by x, y, x and their variants x', x·, etc. The classification presented here draws on the fact that groups of predicates
320 The
lntergresstve as a New Category of Verbal Akoonsart
In add1tion, the fearures form a luerarchy; i.e. an element cannot have fearures 'higher' on the hierarchy unless it has all the lower ones.5 The set of predicates is ordered by the already mentioned three fearures:
(IJ)
interval-based bounded relic
These fearures each divide the set of predicates m two, which results in four murually non-overlapping groups:
(•4) state predicates process predicates intergress1ve predicates change predicates
to love, to be in the pub,
and
to have a car,
process predicates, to walk, to sing, and to dance, of mrergressive predicates, flash, to sing a song, and to cough, of change predicates, to die, togo to the pub, and
of
to to
build a house. The order of the fearures m {I 3) is important the hierarchy goes from left to right; an aktionsart cannot possess the fearures 'bounded' or 'relic' unless It is interval-based. In other words, bounded prediCates are a subgroup of the interval-based predicates; rehc predicates are a subgroup of the bounded (and of the interval-based) predicates. {Note that no such set-theoretic relation holds for the members of different aktionsarten.) Every predicate has a value for each of the features; no predicates which are unmarked for one or more of the features are assumed. This claim rests on the observation that the tests adduced for these fearures are decisive for all predicates; that is, the tests clearly indicate whether a predicate has a positive or a negative value for one of these features. For instance, modifying a predicate with adverbs likefor one hour without reinterpretation of the predicate is either possible (which signals unboundedness of the modified predicate) or it is not; there is no third option. This is not so with the tests e.g. for punctuality, where It is there may be considerable leeway in interpreting the results (see section also worth pointing our that the fearures in I 3) are propernes of predicates, not
.p).
(
of situations. Different predicates may describe one single siruation from different points of view, for instance, a situanon described by the bounded predicate { I Sa) may also be referred to by the unbounded {I 5b):6
(Is)
a.
Bruno watched an Inspector Columbo serial
b. Bruno watched 1V
(
(
The feature hierarchy I 3) and the classificanon 1 4) also restrict the possible means of a language of changing the aktionsart of a predicate: they can apply to a predicate only if they add or take away coherent parts of the hierarchy without creating gaps in the hierarchy. E.g. the application of the English progressive operator on a change predicate results in a predicate that is nor only unbounded, but also non-interval-based (hence, a state predicate).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Examples of state predicates are
Markus Egg
321
The classllication is illustrated in the following schema. Fields with a cross denote a positive value, empty fields a negative value: {16)
interval-based state predicate process predicate intergressive predicate change predicate
bounded
telic
X X
X
X
X
Thefeatures
Here the features wtll be expounded and linguistic relevance of the classes and thetr features will be demonstrated. Evidence for these distinctions will be adduced by employtng tests (for a substantive hst, see Dowty 1979: 51 ff). I will also show that my formal definitions of the features allow the logtcal derivanon of the hierarchical relations between them? 4.I.I The feature 'interval-based'
The validity of interval-based predicates can only be calculated with respect to at least two points of time. Non-interval-based predicates can be evaluated for single points of time, too. This features distinguishes state predicates from all other aktionsarteiL The interpretation of English sentences in the simple present can be put down to this feature; hence, an attempt to explain these data need not follow Galton (1984) in assuming a dichotomy of the denotata of predicates (he regards unbounded predicates as sets of times, bounded ones as sets of events, which are ontologically different entities): interval-based predicates in the simple present (except performarive verbs as to promise or special narrative techniques as in sports reports) cannot be understood as direct predicanons about the moment of utterance. This is not so for non-interval-based predicates: they merely state that a certain predicate holds for the moment of utterance. Interval-based predicates in the simple present are reinterpreted as habituals (as m ( I 8)) or they are understood as pertaining to a time in the future (especially in the presence of a future adverb, as in Mary starts school tomorrow; Galton 1984: ro). Habitual readings of a predicate P are non-basic in that they involve the notion of a certain frequency of occurrence for denotata of P rather than simply denoting sets of entities for which P holds (see Krifka 1988). Consider the following examples (predicates chat are only acceptable in a reinterpreted meaning or in special contexts are indicated by'#'):
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
4.1
X
322 The
(I7)
(I8)
lntergresstve as a New Category of Verbal Akoonsan
Nigel is in the kitchen # Nigel sits in the most comfortable chair b. # Nigel runs c. # Nigel plays Greensleeves d. # Nigel drinks a pint of beer a.
The only sentence which is not reinterpreted as habitual is the non-interval
(
based I 7). For members of all other aktionsarten, some reanalysis is compulsory. (See also Dowty I979: I 7 3 ff. on the habitual interpretation of
verbs like to lie in the simple present) There are two reasons for this: if predicates apply to extended intervals only, they cannot apply to the (punctual) predicates in the simple present must be understood differently, e.g. as non
interval-based predicates (habituals are stative). However, some intergtessive and change predicates can apply to points of time. The fact that they must be reinterpreted if they are used in the simple present needs a different explanation. The reason for this reinterpretation is that these predicates are interval-based. They make prediCations about more than the one point of time to which they apply in that they also restrict the environment of tlus temporal entity. E.g. if a change predicate P applies to a point of nme t, there is a certain predicate P' whose contrary holds for an interval immediately preceding t, while P' holds for an interval t' unmediately following tHence, one can only attribute P to a temporal entity t If one is certain that t is followed by such a t'. Thus, only after clus t' has set in can one truthfully attribute P to t But then t is already past and the present tense no longer adequate (Galton I98+ I J). Therefore, these predicates must be understood in a different way, too, if they appear in the simple present8 Formally, interval-based predicates are charactenzed by the property of propertiesm:
(I9) VP(ffi(P) - 3P' Vt D(P(t) 1\ PT(t) ..... 3t'[P'(t') 1\ t � t' 1\ qe, P', P)])] In prose, a predicate P is interval-based iff there is a predicate P' such that necessarily the validity of P for points of time t entails that P' holds for a temporal entity t' different from t such that qt', P', P) is true. That is, interval based predicates either apply to non-atomic temporal entities or make predic ations about at least two temporal entities. The necessity operator is
)
indispensable in ( 19 , as in a certain world the validity of a non-interval-based predicate P for points of time may always happen to co-occur with the validity of a predicate P' for another time. The restriction C excludes trivial fulftlment ofm: P' must be different from any property that necessarily holds for t' (say, by its being a time) and from any property of t' that presupposes the existence of a time t• that falls under P. E.g. P' cannot be the property of being different from a time t• such that P holds for t•.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
moment of utterance.Hence, process, non-punctual intergressive, and change
Markus Egg 32 3
(w)
C- ).p ).Q J.t'[O-.Q(t') A
O(Q(t') A -.3t.(P(t.))))
This definition is compatible with the traditional analysis of predicates like to
flash or to explod� as pertainmg to points of time only. Predicates may invariably dmote sets of points of time and yet always make predications about larger temporal entities. Consider e.g. toflash. While it applies to points of time only, it invariably makes predications not only about these pomts of time (namely, that a certain emanation of light takes place there) but also about the immediate vicinity of these points of time (namely, that no such emanation takes place there). Note that
(19)
implies that mterval-based predicates have a nonempty
time. This restriction captures the intuition (which
has
been the basis of the
discussion of strict cumulatiV1ty as opposed to cumulativiry; see note I I ) that the distinction of aktionsarren makes sense only if one restricts oneself to worlds in which there is more than one point of time. In a world in which there IS only one point of time the distincnon between the aknonsarren breaks down. The first group of interval-based predicates are dynamic, i.e. they denote or Imply a change of state of some kind (this change does not necessarily show up directly in the semantics of the predicate). Evaluating a change of state involves the comparison of the state of affairs at two points in time at least. Many process
predicates are interval-based, as they involve a continuous but indefinite change of state (Dowry 1979: 169).9 Change predicates denote a definite change of state (in a definite change of state the predicate whose validity changes is explicitly specified in the semantic decomposition of the predicate). On the other hand, other predicates (e.g. toflash and to runfor exactlyfive hours) express the validity of a specific unbounded predicate for maximal temporal intervals, which by definition implies the invalidity of this second predicate for the intervals directly preceding and following the maximal interval see section s.2).
(
But, obviously, in between an interval for which a predicate holds and an adjacent interval for which it does not hold, a change of state must take place. The exclusion of a change of state likewise entails a positive value for the feature interval-based: the non-occurrence of a change of state can be
calculated with respect to intervals of time only. Consider e.g. predicates like to sit or to lie: they are valid if the state of affairs at different points of time does not differ with respect to the validity of an embedded predicate which is claimed to
connnue (Fabricius-Hausen 1975; Dowry based, too.
1979: 176). Hence, they are interval
A third group of interval-based predicates is exemplified by sentence
(2 1) :
(21)
Felix was in the pub for five hours
(21)
cannot be evaluated with respect to single points of time but only with
respect to the whole temporal interval introduced by the adverbial. Thus, (21 ) is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
extension only in worlds in which there are more times than just one point of
324
The lntergressive as a New Category of Verbal Akno�rt
likeWlSe an interval-based predicate. The non-interval-based group of predicates comprises only state predicates.
4.1.2 The fearure 'bounded' Boundedness indicates that the validity of a predicate is temporally limited. Change and intergressive predicates have a positive value for boundedness. State and process predicates, on the other hand, are unbounded. This feature is related to two other well-known properoes: bounded predicates are netther and divtStvity mean cumulative nor divisive.1° Cumulativity
(CUM)
closedness with respect to a suitable join operation
(22)
a.
b.
( ) and with respect to an
3.1 ); e.g. in Krifka ( 1992 : 3 2):11
VP(CUM(P)- Vx Vy(P(x) A P(y)- P(x v y))) VP[DIV(P)- Vx Vy((P(x) A y � x) - P (y)))
With the help of the axioms (6a), (sc), and (sa), both properties can be merged mto 2 3):
( (23 ) VP((CUM(P) A DIV(P))- Vx Vy(P(x) A P(y)- P(x
v
y)) )
However, process predicates are only divisive down to a predicate-specific minimal interval. Boundedness results from the internal verbal structure or from ex ternal operanons. This feature appears both in the verbal and in the
nominal domain. Thts point has been worked out by and will be taken up m section 6.
Krifka and Bach ( 1 986)
Bounded predicates are interval-based (VP(BD)- m(P))). Proof assume to the contrary a bounded bur non-interval-based predicate P. Then, accordmg to
( r9), there is no predicate P' such that the validity of P for a point of orne t necessarily entails that P' applies to a t' different from t such that qt', P', P) is true. But due to r 2), any t' that properly contains such a t must fall under -.P.
(
As ( 12) holds good in every possible world, and as for all t" and t'", t" c t"' entails t" "# t"', we have found for every possible world in which there is more than one point of time a P' (i.e., -.P), whose validtty for a t' different from t follows from P applying to a point of time t. qt', -.P, P) is true as -.p may hold for a t' even if there is no t" that falls under P. This contradicts the assumption that Pis not interval-based. Possible worlds in which there is one point of time only have been excluded from consideration m this proof. as no sensible aktionsart distinction can be drawn in these worlds. Due to its dominant status, the presence or absence of the feature 'bounded' is illustrated by several tests. The most important ones are compatibihty with adverbs of repetition or with durative adverbs. Bounded predicates denote countable entities because the members of their extensions have boundaries. Hence, only bounded predicates allow modification by adverbs of repetition like twice or repeat�ly (without reinterpretation or a special context).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
appropriate part-whole relation (�; see section
(DIV)
v
Markus Egg
{24)
J2 5
a.
Ariadne played a sonata twice Felix emptied his tankard twice b. c. # Felix ran twice
Barring reinterpretanon. only elements of the extension of predicates that are both cumulative and divisive are measurable by durative adverbs. This explains the nonacceptability of (zsb) (as opposed to (zsa)) m its literal meaning. This te'it does not apply to mtergressive predicates like (3), as the mcom patibihty of (3) with a duranve verb can be explamed differently: two duranve adverbs in one sentence are either conflicting or redundant.
The results of these tests parallel the subdivision in the nommal domain: count nouns are bounded, hence their denotata are countable; mass nouns are not bounded, therefore they can be combined wuh measure phrases. Mass nouns become bounded by mensuratioiL Sirrularly, the predicate (zsa), the result of a mensuration of the unbounded predicate to be ill by a duranve adverbial, is bounded (see sections 5·3 and 6). There are further tests to distinguish event predicates from other predicates. They single out cumulative or divisive predicates, so they fail for the bounded event predicates. I will gtve two examples: the first is the well-known (shghdy adapted)12 implication from X is V-ing to X V-ed {Dowty 1979: 57· 172), which holds for all divtsive predicates. E.g. if one can say Ariadne is reading at some moment of rime t then there must be a stretch of time Immediately preceding t at which Anadne was readmg, too; hence Ariadne read is also true at t. The second is Verkuyl's (1989: so) coordtnation test: if two coordinated temporal adverbs that modify a predicate allow a reading which mvolves only one situation, the predicate must be cumulative. E.g. (26a) may denote one single situation, whereas (z6b) must describe two. (26)
run on Monday and on Tuesday b. dehver a parcel on Monday and on Tuesday
a.
4. 1 . 3 The feature 'relic'
Predicates are relic if they introduce a definite change of state. Change predicates constitute the sole group of relic predicates. Formally, these predicates are characterized by an operator that denotes a change from the val idity of a state or process predicate to the validity of its contrary. Since change predicates are bounded, they apply to limited temporal intervals only. Intervals denoted by telic predicates are inherently preceded and followed (and also limited) by other intervals for which the preceding and the consequent state or
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(25) a. Konstantin was ill for three days b. # Konstantin played a sonata for three days
326
The Intergresstve
as a
New Category
of Verbal Akrionsart
process hold. E.g. the interval of the l�ving of a house is inherently preceded and followed and bounded by intervals during which one is inside the house and outside the house, respectively. The domain of relic predicates comprises the Vendlerian classes accomplishments and achievements. Fonnally, telicity can be defined thus:
(27)
VP(TEL(P)- 3P' V t D(P(t) .... ATOM(t, A.t 3t' 3r·(-P'(t') 1\ P'(r•) 1\ t' At 1\ t A r·) 1\ qr·,P', P))))
......,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
According to (27), a predicate P is relic iff there is a predicate P' such that the validity ofP for a t necessarily implies that t is a mimmal interval With respect to a second, complex predicate: this second predicate states that there is a t' which immediately precedes t such that -P'(t) (i.e. the contrary ofP' holds for t') while P' holds for a r· immediately folloWing t Hence, P' must be undefined for any such t In addition. C (see (20)) restricts P'.t In (27), the necessity operator is indispensable: in a certain world, the contrary of a predic ate P' may happen to be valid immediately before, and P' may happen to be valid immediately after any interval in the extension of an atelic predicateP. (27) poses conditions on the definition of the telicity operator BECOME, which characterizes all tehc predicates. This operator resembles closely Dowty's BECOME (1979: 141), but uses a distinction between times for which the con trary of a state or process predicate P holds (the set of t such that -P(t)) and times for which P is either false or undefined (the denotation of A.t P(t)). BECOME(a) is valid for smallest intervals t such that there is a t' immediately preceding t such that -a(t') while a holds for a r· immediately following t Definitions (12) and (27) are in accordance with the feature hierarchy (13): telicity entails boundedness, i.e. VP(TEL(P) .... BD(P)). The proof is very simple: assume to the contrary a relic unbounded predicateP. Then, according to (12), there are two intervals t and t' m the extension ofP such that t c t'. However, this violates the definition of teliClty:P cannot apply to both t and t', because in that case t' would not be minimal with respect to A.t 3t' 3r· (-P'(t') 1\P'(r•) 1\ t' A t 1\ t A r·) for the given predicate P'. Hence, relic predicates must be bounded. The following tests for telicity separate intergressive and change predicates. The main test for telicity is a comparison of intervals immediately preceding and following the interval for which a predicate holds. The means of describing such intervals in terms of the involved predicate vary from language to language; these intervals can be denoted either by appropriate forms of the respective predicate or by periphrastic constructions that involve the predicate. In English, both possibilities are made use of immediately preceding intervals can be expressed by the to begoing to-construction. immediately following intervals by the perfect of the relevant predicate.13 If the comparison of the meanings of the two predicates forth� rwo intervals reveals that one of them entails a state or process predicate
Markus Egg 327
while the other entails the contrary of this state or process predicate, then the predicate is relic. Consider e.g. the following pair ofsentences: (28) a. b. (29) a. b.
Fritz 1s going to enter the pub Fritz has entered the pub Fritz is going to run a mile Fritz has run a rrule
(3o) Fntz left the pub The semantics of (30) gives no clue about what happens during an interval for which (30) holds. (Our conjecture that Fritz moved dunng this interval is based on world knowledge, not on the semantics of to leave.) On the other hand, the meaning of to runfor an hour or to cough entails that during the denoted intervals there is running or a kind of noise, respectively. These concluswns must follow from an appropriate semantics of intergressive predicates. Another test for telicity is the mterpretation offor-phrases. If these phrases modify relic predicates, a reinterpretation is needed in order to avoid a clash between the boundedness of the predicate and the selection restriction (cumulativity and divisivity) of the modifier. For relic predicates (except those denoting irreversible change predicates like to dit), there is a special way of reinterpreting the adverbial. As an example, consider (3 I ): (3 I )
#
Amelie went to Oxford for one year
The most natural interpretation of this sentence would be 'Amelie went to Oxford and stayed there for one year'. In other words, the for-adverbial may specify the length of the consequent state. Whenever such a reinterpretation is
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The comparison of (28a) and (28b) shows that there is such a stare: in one interpretation of (28a), the state of Fritz being outside the pub follows from its meaning, whereas (28b) has an interpretation that entails the contrary of his being outside the pub, namely, that he 1s inside the pub. But no such state emerges from the companson of the semantics of (29a) and (29b): these two predicates do not entail predicates that are contrary to each other. Neither are they contrary to each other themselves. Another test distinguishes intergressive and change predicates by trying to derive what is going on during the nme for which the event predicate holds. If the semantic description of the predicate entails no such information, the predicate is relic. (The reverse does not hold, as some--change predicates include such information, e.g. to walk to the station.) The reason ipr -fhis implication is that changes of state are definable in terms of their irrimed1ately preceding and following intervals. E.g. (3o) 1s analysed as the smalleSt interval in _between mtervals for which the predicates 'Fritz is inside the pub' an4 'Fritz is outside the pub', respectively, hold.
328 The Intergress�v e as a New Category of Verbal Aknonsart possible, the modified predicate is relic. For predicates of all other aknonsarten, this reinterpretation is ruled out In these cases, the adverbial must always specify the length of the mterval denoted by lts (possibly reinterpreted) modified predicate. The reason for the special interpretanon of sentences like
(3 I)
is chat relic predicates explicitly make reference to a consequent state in
their semanncs, which distingmshes them from predicates of all ocher aktionsarten I will conclude this section with some comments on the status of the intergressive predicates. The relative scarcity of intergressive verbs rrughc suggest the treatment of this phenomenon as something not on a par with
coercions are postlexical processes wluch change the aktionsart of a predicate to the aktionsart demanded by its context or its mcxhfier so as to av01d ungrammancahty. For instance, the fact chat (32) denotes five bounded spans of Fritz's staytng in the pub could be put down to a posdexical bounding of the VP
to be in the pub, which enables the adverbial phrase five times (which lS accordmgly regarded as applicable to bounded predicates only) to modtfy the VP. (See Egg I 994: I 54 for an analysis of adverbtals hkefive times.)
(32) Fritz was in the pub five times However, intergressive predicates occur m the leXIcon as well as on the level of VPs and sentences (for instance, to cough, to flash , or to wink). All these mtergressive lexemes belong to one of the two groups of intergresstve predicates that I Will distinguish in section S · (They belong to the intergressive
predicates that are bounded by maximality [group I ] ; no leXIcal intergresstve predicate belongs to the intergressive predicates whose boundedness results from the measunng of their duranon [group z].)
Further evidence for the intergressive aktionsart is provided by many ocher lexical devices that involve the intergressive aktionsart in their semantics although they are not intergressive predicates themselves. Besides lexemes that are ambtguous between a stative and an intergressive reading (like through, which I Will analyse in section 7, or the verb to cross), there are modifiers which impose intergressive aktionsart on their modificanda. Most prominent among these are durative adverbs (likeforfive hours) [group 2],14 bur intergressivity may also be introduced by verbal morphology (the French passe simple, the Ancient Greek aorist and aspectual morphology in Slavtc languages [group I ] ) and processes of predicate nominalizations (to take a walk as opposed to to walk [group 1 ]). These phenomena will be taken up m section S · Considering chis amount of lexical means of expressing intergressive aktionsarc, one cannot regard mtergressive predicates as minor postlexical phenomenon
Now chat the features which are employed in the present classification have
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
aknonsarten One might consider analysing these verbs in terms of the so called 'type coercion' as presented by Moens & Steedman (I 986, I 98 8). Type
Markus Egg 329
been described and motivated, I will turn to the feature punctuality, which is often used in other classifications of verbal akrionsart, but is not a major feature in the present classification. 4.2
Thefeature 'punctuality '
(3 3) Ariadne stopped writing her CV (34) # Ariadne stopped winning the scholarship But the same phenomenon shows up for puncrual and not puncrual intergressive predicates:
(3 5)
Ariadne stopped playing the sonata
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Here I will discuss the additional feature punctuality and justify its omission from the proposed classificanon. While I do not question the reahty of this feature, I regard it as a minor phenomenon m the system of aknonsanen. Closely related to this issue IS the distmcrion in accomplishments and achieve ments which partitions the group of change predicates. This distinction figures prominently in the Vendler-Dowty dassificanon (see section 2. 1 ). I will show that this distinction does not back up the assumpnon of a feature 'punctuahry' in akrionsanen classifications, as it hes outside the range of akrionsanen classification. Fmt, I will argue against Dowty (1 979), who assigns to puncrualiry_ far reachmg influence on the semantic decomposition of a predicate. But refuting this analysis is in itself no proof for the rnmor status of punctuality. The argument I will adduce for this view is taken from Verkuyl (1 989) and shows that certain predicates cannot be assigned a value for this fearure sernanncally. In this respect punetuahty differs from the other fearures: any predicate is specified for the fearures interval-based, bounded, and telic. This difference shows that puncruahty is not on a par with the other fearures. The feature puncruahty cuts across the distinction between telic and atelic predicates and represents the resrricrion of the extenswn of a prediate to pomts of rime. For change predicates, this distinction Is included m the Vendler Dowty dassificanon: puncrual change predicates are called achievements, non puncrual ones accomplishments. Puncrual change predicates can be characterized by the tests adduced by Dowry ( 1979: sS £) for the disrincnon of achievements. But these tests also partition intergressive predicates into puncrual (toflash, to wink) and extended ones (to run a mile, to sleepfor an hour). Consider, e.g. the test that distinguishes accomplishments and achievements by the ability to occur as complement of to stop. As opposed to accomplish ments, achievements cannot be the complement of to stop (unless they are reinterpreted). Hence, (3 3) is fine, whereas (34) is only acceptable under an iterative reinterpretation.
3 30
The lntergreSSive
as a
New Category of Verbal Aknonsart
(36) ( # ) Ariadne stopped winking13 One more test for puncrualiry is the modification of a predicate with adverbs like suddenly: in order to allow for this modification, all non-puncrual predicates must be reinterpreted as ingressive (predicates that denote the commencement of the siruanon expressed by the uninterpreted predicate). (37) a. b. c. d. e.
# # # #
f.
Sentences (37a}-{37d) are understood as ingressives only; hence, they are marked by '#'. In other words, they can be paraphrased as Suddenly, Heinrich began to . . without change in meaning. No such rephrasing is necessary for (37e) and (37f). This test once more demonstrates that the distinction in punctual and non-puncrual predicates cuts across both the group of change predicates and the group of mtergressive predicates. In Dowry ( 1 979), decomposttions of non-puncrual change predicates (which are equated wtth accomplishments) differ radically from those for puncrual change predicates (which are equated with achievements). The former are analysed as causatives; omitting agentiviry operators, their decomposition consists of the embedding of two predtcates under an operator CAUSE. The first argument of CAUSE is a predicate which expresses the cause, its second argument a predicate that plays the role of result. Achievements, on the other hand, denote only changes of state, which ts expressed as application of the operator BECOME to a state or process predicate. (See Dowry 1 979: 1 o8 for the exact definition of CAUSE) I do not question the value of this analysis in the field oflexical semantics nor deny that accomplishments and achievements differ with respect to causation. The point I WISh to make is that thts difference is independent of the feature punctuality. It is of no avatl for any temporal property of predicates: hence, it does not belong to the realm of akoonsart. The strucrure of the decomposition of change predicates represents the inner structure of the siruations denoted by these predicates but does not deter mine their value for the fearure puncruality: there are puncrual accomplish ments and non-puncrual achievements as well as the familiar non-puncrual accomplishments and puncrual achievements: the verb to shoot is Qike all the verbs m the lexical field 'kill') a protoryptcal example of an accomplishment. Yet it is puncrual, which can be seen from the fact that it must be reinterpreted iteratively in (38a), while it cannot and need not be reinterpreted in (38b): .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Suddenly, Heinrich was asleep Suddenly, Heinrich ran Suddenly, Hetnrich built a house Suddenly, Heinrich ran a mile Suddenly, Heinrich coughed Suddenly, Heinrich vanished
Markus Egg
331
(38) a. # John stopped shooting b.
Suddenly, John shot
On the other hand, many predicates which describe only a change of state and
leak out. This can be shown by its modification by an adverb like in an hour. These
involve no causation may be non-punctual, for instance the predicate to
adverbs apply to non-punctual bounded predicates only and specify their (maximal) temporal duration. If they modify other predicates, these predicates must be reinterpreted, usually as ingressives. In that case, the adverbial specifies the time which elapses before the ingression rakes place: (39) a. # John was dead in an hour The water leaked out in one hour
While the only feasible interpretation of (39a) is 'After one hour had elapsed, the state that John was dead began', sentence (39b) can be paraphrased as 'The leaking out of the water lasted one hour'. Hence, to leak out 1s non-punctual . In sum, the structural distinction between achievements and accomplish ments as proposed by Dowry is independent of punctuality; hence, this distinction does not in itself specify a value for this feature. On the other hand, punctuality does not decennine whether the structure of the decompositions of relic predicates must include a causative operator or not The minor status of the feature punctuality is evident from the fact that not all predicates must have a value for it The temporal extension of the denotata of some predicates is not fixed and therefore dependent on the context of the situations they denote. A change in the context results in a corresponding change in the temporal extension. Changes in the context of situations, of course, are extralinguistic properties of situations, and not linguistic features. (Nevertheless there are punctual or non-punctual predicates like to enter or to
mate a work ofart, respectively.)16 The following example has been presented by Verkuyl (1 989: s6fi). It concerns the typing ofletters (symbols and texts): in a mor traditional context, typing a letter symbol is considered to be an instantaneous event, but the writing of a letter (text) is regarded as an extended event: (4o) a. # John typed the letter p in five minutes b.
John typed the business letter in five minutes
Therefore, in this rraditional context the adverbial measures the time elapsing before the event in (4oa); in (4ob), it measures the duration of the event17 Hence, (4oa) must be puncrual and (4ob) non-puncrual. But within a more computer-oriented context the reverse is also possible: the typing of a standard letter may involve hitting one key only, while on the other hand the typing of a letter symbol may take some time
(if the
used word processor has no free
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
b.
332
The lntergressive as
a
New Category of Verbal Aknonsan
memory available at the moment of the typing). In that case, the mterpretanon of rhe adverbs and the resulnng classificanon of the predicates would be exacrly the ocher way round. Hence, the predicates to type a business letter/the letter p cannot be assigned a value for the feature punctuality in their semantics. The mmor status of punctuality is represented in Dowry's 'aspect calculus' by the fact that it cannot be expressed like the other aktionsarten: it cannot be represented by or read off a certain structure of the decomposition of a predicate; there is no operator for this feature. Instead, the value for this feature must be speofied for each (lexical) prediCate in the metalanguage. (4 I ) stares chat punctual predicates only hold for points or rime: All punctual predicates are bounded, i.e. VP[PUN(P) - BD(P)]. Proof assume to the contrary the eXIStence of a punctual unbounded predicate P. Then, accordmg to ( I 2), there exist r and t' such char P(t), P(r') and r C r'. As punctual predicates apply to points of time only (see (4I )), both t and t' must be points of time. However, as t C r', this contradicts the defimrion of points of time in (8). On the other hand, punctuality and rehciry are independent of each ocher. Accordingly, the properties PUN and TEL as defined in (4 I) and (27) are compatible with each ocher as well as PUN and TEL -.PUN and TEL, or -.PUN and TEL In sum, there are three features m my classtficanon (interval-based, bounded, and relic) which Impose an ideal ordering on the set of predicates. The additional feature punctuality is considered ro be a minor feature within aknonsarten systems because it cannot be equated with the disrincnon between the decompositions of achievements and accomplishments (in Dowry's sense) and because nor every predicate has a value for this feature semantically: m certain cases, this value is determined by extralinguistic, contextual influence. -.
-.
,
.
s I NTERGRE S S I VE O PE RATOR S Now that I have established and described the three features of my classification, this section IS devoted ro the formalization of inrergressive predicates, followmg Dowry's ( I 979) approach: more complex decompositions fall into less complex ones linked by suitable operators. The goal of this approach is to make the decomposition of predicates and of adverbs directly reveal their akrionsart or their potennal to Influence aktionsart. Eg. the operator BECOME is part of the decomposition of every rehc verb and the decomposition of adverbs likefor an hour must indicate char it bounds unbounded predicates (see section s.J). Taken together, the rwo goals aim at deriving the aknonsarr of a complex predicate compositionally.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(4 I) VP(PUN(P) - Vx(P(x) - PT(x))J
Markus Egg 3 3 3
There IS a characterisnc operator m the semantic decomposition of every intergresstve. This operator takes as tts argument an unbounded predicate (state or process predicate) and yields a bounded atelic predicate. I will first diScuss Galton's ( 1 984) and Herweg's ( 1 990) suggestions for operators which may be used as intergressive operators in my framework. Herweg's operator will be employed as a first intergressive operator. But as some mtergressive predicates cannot be represented with this operator, I will suggest another intergresstve operator.
(42) a. Perf Po p - P*(P(P -.p 1\ -.p) 1\ -.p) b. Pros Po p - f*(-.p 1\ F(p 1\ F -.p)) c. Prog Po p - P -.p 1\ p 1\ F -.p P*p is shorthand for p V Pp, F*p abbreviates Fp V p. F and P are operators on predicates taken from standard Priorian terue logic and mean 'there is a orne m the past/future for which the argument ofP/F IS valid'. Perf, Pros, and Prog are properties of event predicates which mean that an event (their argument) took place, will take place, or is taking place, respecnvely. E.g. (42a) mearu that an event Po p took place if and only if there was a maximal interval for which p held that ended (at least immediately) before the time toAccording to (42), predtcates on times p may be false or undefined for the vicinity of situatioru of type Po p, as '....,p' is defined as in section 3.2. This gives the right result for cases like (43): (43) Max was in London twice
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
operator which might be employed as intergressive operator has been described by Galton ( 1 984). This operator was to account for a group of corurructioru like to Vfor a while, to have a V, and the Russian verbal prefix per 'V for a while'. Galton has christened it 'Po' (for 'pofective', an allusion to po -). 18 In his framework, there are predicates on nmes and predtcates on events, which are related by certain operators, among them Po. Its argument is a predicate on nmes p; it returru an event predtcate Po p that is based on maxtmal phases for which predicate p holds. But Galton cannot tntroduce thts operator compositionally: it is imposstble to define the operator Po in ISolanon; for each of the three posstble relatioru of speech nme and duration time of an event Po p (precedence, overlap, and succession), -a separate operator is needed. Thts is due to the fact that Galton tries to define Po solely in terms of the Priorian terue operators plus conJunction and negation. The common core of all three definitioru is that the interval on which the event predicate is based must have a beginning and an end (Galton 1984: So ff): An
3 34 The lntergressive as a New Category of Verbal Akoonsart
5.2
Herweg (1990)
The first operator which will be used as a suitable operator for the formal description of intergressive predicates has been proposed by Herweg ( I 990) in order to describe phenomena like the reinterpretation of unbounded predicates as bounded ones. This process is forced by certain contexts, e.g. by frequency adverbs as in (43). Herweg considers state and process predicates as predicates on temporal intervals, other predicates as applying to events. Events and times are different entities, which are nevertheless related by operators and functions. One of them is the operator PO, which will be discussed in this section. Herweg (I 990: I03) first defines maximal phases (maximal intervals of validity) of a state or process predicate:
(44) VS Vt(S-phase(t) - S(t) 1\ Vr '(t ' � t - S(t ')) 1\ Vr "(t c r · - -.S(t "))) For maximal temporal intervals r and their parts the respective predicate must hold, but it must fail for all superintervals oft. One may model maximality with the help of (44) even in the case of sentences like (43) as sketched above. The second conjunct of the definition (Vr '(t' � r - S(r '))) makes the S-phase applicable to state and process predicates only (it rules our non-divisive predicates). S-phases are atelic according to the definitions of relicity (2 7): no predicate is necessarily valid for an interval immediately preceding an S-phase while irs contrary must hold for an interval immediately following this phase.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Due to the frequency adverbial twice, (43) involves two maximal phases for whtch Max be in London holds (Herweg 1 990: 6 1). Bur (43) is also felicitous if Max was born in London and died there, bur was never in London during the rest of his hfe. In other words, the two phases during which Max stayed in London are maximal bur nor bounded on both sides by phases during which he was not in London: before his birth and after his death, the predicate Max be in London is neither wrong nor right, bur has no truth value at all. The operator Po is flexible enough to model the maximality of Max's stays in London even in such cases. However, there are-in addition to the imposs1btlity of isolating the meaning of the operator Po-rwo more problems with Po. The first has been mentioned by Herweg ( I 990: 78): for given p, Po p does nor only denote maximal phases of the type p; its extension also mcludes proper parts of these phases. The second problem is the validity of the argument p of Po for the whole interval between the commencement and the termination of Po p. The definition (42) is roo weak to prevent an interruption of the validity of p during the relevant interval. Since Galton's operator is too weak and must be defined differently with respect to the context m which it occurs, it is not necessary for the analysis of the intergressive operator.
Markus Egg
33s
For both these intervals, -.S holds. (For a similar defirution of a maximality
operator see Krifka
I 989: I 79 £)
Next, Herweg introduces the operator PO by the following posrulates
64). PO links S-phases and events: (4 S )
(I 99I :
a. VS Vt(S-phase(t) - 3!e(PO(SXe) 1\ r(e) - t)) b. VS Ve(PO(SXe) -+ S-phase(r(e)))
For every maximal phase t of a state or process predicate S, there is exactly one corresponding event (this is indicated by the quantifier 3!) of the type PO(S) , whose time of duration is the interval t 1' maps events onto their duranon time) . On the other hand, each event of the type PO(S) has a duration time
(
who do not make this distinction can regard the S-phase itself as intergressive operator.
I will employ the S-phase as intergressive operator in my classification. Now
we are in a position to give an adequate semanncs for lexemes like toflash or to cough: (46) represents the sentence radical the lightflash. (46) stands for the set of times that are maximal with respect to the predicate the light shine (L abbreVIates the representation of the light).
(46)
J. t(A. t ' shine' (LXt '))-phase(t)
The S-phase is also adequate for the representation of sentences like (43). There are two maximal phases of Max's staying in London involved in (43). This sentence cannot be interpreted in the sense that Max stayed in London during two periods of time which were not interrupted by a period of Max's absence from London. If either phase is represented by (47), this follows from the definition of the S-phase: maximal phases of a state or process predicate p must be surrounded by mtervals for which p is false or undefined.
(47)
A.t (Max be in London)-phase(t)
Modificators of predicates which involve the application of such an mtergressive operator are the Russian verbal prefix po- 'for a while' and the
nominalizations of verbs in English. Consider for instance to take a walk, which is a bounded event predicate while to walk is a process predicate. The French
passe simple and the Ancient Greek aorist denote bounded predicates, too. The countability of such predicates, which has been noted in Krifka (I 989: I 8 I) and
Armstrong (I 98 1 ), proves their boundedness. They cite the following sentences:
(48)
a.
Charlotte dormit trois fois
'Charlotte slept three times' b. "Of.UJwv & roiiro � u ellreiv Kai del Uyt:cv (ZENO B r
)
'It is the same to say (aor.) this once and to say (pres.) it always'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
which is an S-phase. The operator PO is necessary in Herweg's approach because he distinguishes predicates on times and predicates on events. Those
336
The lntergress1ve
as
a New Category
of Verbal Aktionsart
5·3 A
second intergress ive operator
In this section it will be demonstrated that not all intergressive predicates can be described in terms of Herweg's definition of PO and S-phase. The examples around which the discussion of this section centres are state and process predicates modified by durative adverbs. These complex predicates lead to problems for the first intergressive operator and will be accounted for in terms of a second intergressive operator.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Both phenomena can be described in terms of maximal phases. However, a description of passe simple and aorist in terms of (44) would necessitate the onussion of the second conjunct of the definition, since these two operators apply to divisive as well as to non-dlVlsive predicates. If a predicate is bounded, the predicate is by itself maximal; thus, Its maximal phase is the same as the predicate itsel£ A further point must be made. In the next section, I will introduce an additional mtergressive operator, whose bounding effect is not based on the criterion of maximahry. While the bounded status of the predicates in (48a) and (48b) is obvious, the maximaliry of the involved phases has not been shown yet. In Ancient Greek, there is a phenomenon supporting the interpretation of the aorist in terms of maximal phases. It can be described as 'resulrative effect': some process predicates descnbe situanons that may also function as the cause involved in a causatioiL These predicates denote the whole causanon (both cause and result) if they are pur into the aorist One member of this group is the verb m{8w. Its present in£nitive m{8etv means 'be in the process of persuading', which of course may also be the cause for someone's conviction. Accordingly, its aorist counterpart miaa. t means 'convmce'. This phenomenon can be explamed m terms of the intergressive. The final boundary of the aorist is reinterpreted as the start of the result state associated with the verb. Bur as the cause cannot continue after the beginning of the resulting state in a causation, this reinterpretation presupposes that the validity of the predicate in its non-aorist form terminates at the final boundary. Therefore, the bounding was effected by maximahry of the predicate. In our example, the phase of m{8etv is bounded by the success of the persuasion: we cannot persuade someone of something which he already believes. Once we have convinced him, persuading is no longer possible. Note that this is not the only feasible way of interpreting aorist forms: the semantics of the aorist itself would not rule out other interpretations of these forms as bounded situations (for miaa.t , 'a [possibly futile) act of persuading happened' would be an interpretation compatible with the aorist mearung, too). Hence, I cannot explam why these forms were so mterpreted. I argue the other way round and try to account for an already established perfective use of these forms.
Markus Egg
33 7
Expressions that consist of a predicate modified by a durative adverbtal with a measure phrase that is not itself modified (e.g. forfive hours, bur notfor at least
five hours) are intergessive predicates. Herweg reconstructs
them wtth the help
of his operator PO: durative adverbs not only specify the minimal duration of a situation in his analysis; they also indicate that the situation is
(TO)
maximal with respect to the involved state or process prediate. Thus, (49) is the
common denominator of durative adverbs like for two hours:
{49)
A T A e(PO(T){e) A Qu( r(e)) ;;2
TD)
for the predicate T; furthermore the minimum duration of the events is given by the value TD. Qu is a suitable temporal measure funcnon on times. {Herweg
assumes a partial order in the domain of the values of Qu. This order lS expressed by the sign ';;2', which mdicates (im-)proper inclusion Hence, the meaning of 'Qu(r(e))
:2 TO' 1s 'the value of Qu(r(e)) includes the value TO'.)
Maximal phases of a state or process predicate p are preceded and followed by intervals for which p is false or undefined. Since the duration indicated by the adverb may only pragmatically be assumed as the maximal duration of the
event, this analysis must consider the value TD to be only an (1m-)proper part of the duration of the whole event.
However, the analysis presented m (49) as the semantics of durative adverbials is both too weak and too strong. Constder for instance (so): (so) Alois walked for five hours Intuitively, the sentence radical of this predicate is an atelic bounded predicate: it lS applicable to neither parts nor joins of situations to which it applies. Moreover, tlus predicate may apply to rwo or more adjacent events or temporal intervals: at the end of a five hours' walk. nothing (apart from his condition) can prevent Alois from walking for another five hours. Followtng
(5 I ):
(49), the representation of the sentence
(5 I ) Ae(PO(Aiois walk Xe) A Qu( r(e)) :2
radical of (so) would
be
sh)
In prose, ( 5 1 ) is the set of all events which consist of a maximal phase of Alois's walking and last at least five hours. However, the join of rwo elements in the extension of (5 1 ) would also qualify as a member of (5 1 ), as (5 1 ) specifies only the nunimal length of the events in its extension This entails that there are also
elements in the extension of (5 1 ) (for instance the join of two or more elements of the extemion of (5 1 )) which have proper parts which also fall under
(5 1).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This formula maps state and process prediCates (predicates for nmes, which is expressed by the sorted vanable T) on to event predicates. The members of the sets of events which are denoted by these predicates involve a maximal phase
3 3 8 The lntergress1ve as a New Category of Verbal Aktionsan Hence (s 1 ) lS no bounded predicate, which shows that it is too weak for the definition of durative adverbials. This is due to the fact that (49) tries to account for duranve adverbials, which do not restrict the environment of the intervals to which predicates modified by them apply, with the help of the operator PO, which does restrict the
environment of events to which predicates of the rype PO(P) apply (see also section 54). The restrictiorts of PO are cancelled by specifying Qu(r(e)) only as
minimal duration of the denoted event, which makes the combination of durative adverbial and predicate unbounded. On the other hand (49) is too strong: the representation of for fivt
hours
according to (49) wrongly implies that there must be a break in between two
)
involve a maximal phase of Alois's walking. Hence (s 1 falsely stipulates that in between any two such walks of Alois there must be a phase in which he did not walk.
)
I will take these two problems with (49 as a motivation to interpret durative adverbs m terms of another intergressive operator, which I will call 'S-quantiry'. It is meant to complement, not to replace Herweg's intergressive operator. This operator bounds its argument not by restricting the environment of the phase in quesnon but by referring to Its duration. The adverb measures the length of an interval during which some unbounded predicate holds. The mertsuration bounds the predicate, which entails the bounded nature of the resulting event prechcate. The environment of such a bounded predicate does not influence its akriortsart. One may illustrate this point by the parallel phenomenon in the nominal domain: as soon as a mass noun is modified by a measure phrase (for irtstance
five pints ofbeer), it becomes an individual nominal. This nominal can or cannot refer to a quantiry of beer regardless of whether thts quantiry is surrounded by other quantities of beer or not. This bounded nature of a predicate lS represented as S-quantiry PH(S, t, n):
(5 2)
V t VS Vn(PH(S, t, n) - 'Vt '(t '
� t .... S(t ')) 1\ Qu(t) - n)
According to this definition, an interval t is an S-quanriry if and only ifS holds for t and its parts and, if measured by an appropriate measure function Qu, t 1s mapped on to the value n (e.g. five hours or ten seconds). Such numerative constructiorts are (under certain assumptiorts, see Krifka 1 989: 64 £) always bounded; hence the interval t has a fixed extertsion regardless of whether S holds or does not hold for intervals immediately adjacent to t. As n denotes a fixed value, A. t(PH(S, t, n)) is for given S and n neither a cumulative nor divisive predicate on rimes. The S-quantiry operator apphes to unbounded predicates only, which is secured by the validiry ofS for all (im-)proper subintervals of t. This operator is based on the usual twofold way of defining durative adverbs
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
events or temporal intervals for which (so) holds: both events or mtervals
Markus Egg
3 39
(c£ e.g. Dowty 1 979: 3 3 3; K.rifka 1 989: 1 67; or Vlach 1993: 25 5): ( 1 ) Durative adverbs impose boundaries on a situation for which a predicate holds by specifying its length. Hence, the resulnng predicate is bounded. (2) Durative adverbs are applicable to unbounded predicates only. However, there are different ways of expressing these properties. While Dowty regards expressions like two hours as (bounded) predicates on times (52) assumes (like Herweg 1 990) a general measure function whose values are the entities denoted by two hours and similar phrases. If K.rifka's ( 1 989: 1 67) definition of duraove adverbs, nouns like hour and �ar are themselves interpreted as measure functions: (5 3) maps cumulative predicates P on the set of events that belong to the extension of P and whose temporal duration -r(e) is mapped on to 2 if the measure function hour is applied to -r(e) (hour maps intervals on numbers that represent the length of the mterval in hours). The third conjunct in (5 3) allows only cumulative predicates as input for (53). Durative adverbs are now defined in terms of the S-quannty. The common denominator of these adverbs and the sentence radical of (so) are given the representations (54): (54) a. A.S A. t PH(S, t, TD) b. A. t PH(Alois walk, t, sh) (54) represents the central part of the desired representation of (so): it denotes the set of temporal intervals which last five hours and during which Alois walked. The sentence radical does not imply that Alois's walk lasted no longer than five hours: neither does it indicate that Alois started and stopped walking at the start and the end of the five-hour-interval, respectively. On the other hand, these two pragmatic implicatures are compatible with (54b). It is obvious that this second operator is not applicable for sentences like (s s), as it would not rule out that (s s) could be understood in the sense of'Max spent two adjacent penods of a fixed yet unspecified length in London'. Therefore, both intergressive operators (Herweg's and mine) are necessary, neither one can replace the other. (ss) [-(43)] Max was in London twice The different ways Herweg's and my own intergressive operator bound unbounded predicates (either by measuring a phase or by posrnlating its maximality} correspond to the two kinds of bounding of mass noun terms which will be presented in section 6. This yields further evidence for the relevance of the intergressive aktionsart.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(5 3) A.P A.e[P(e) 1\ hour(-r(e}} - 2 1\ CUM(P}]
340 The lntergressive
5 ·4
as a
A
New Category of Verbal Aknonsart
unification ofboth intergressive operators?
All intergressive predicates could theoretically be represented in terms of one single intergressive operator. I will nevertheless adhere to the distinction of two different intergressive operators because of the fundamentally different nature of the boundaries inrroduced by the two kind of operators. I will first show a uruform rrearment of intergressive predicates. Durative adverbs involving the length of time n can be described in terms of the S-phase: {s6) J.P l t [{l t [Qu{t) :!:;; n])-Phase{ t) 1\ P{t)]
(57)
VP Vt Vn (PH(P, t, n) - {J. t [Qu(t} :!:;; n])-Phase{t} 1\ P(t}]
Although this rule allows for a uniform rrearment of intergressive predtcates, lt smooches out an important distincnon: maximal intergress1ve predicates are bounded from outstde, that is, the mtervals surroundmg the intervals for which such intergress1ve predicates hold are restricted. But measured mtergressiVe predicates are bounded from within, i.e. by restricting the interval itself for which the measured intergresstve predicate h<.?lds. Trus difference is obvtous if one tries to give the decomposttion of temporal adverbs in terms of (s6): in order to fit the meaning of such adverbs into the frame given by the S-phase operator, the simple statement 'the duration of the interval in question is n' (bounding from within) is reformulated in (57) as 'no superinterval of the relevant interval has a duration of more than n' (bounding from without). Therefore, I will adhere to both different intergressive operators. In sum, there are two intergressive operators with the help of which it is possible to represent intergressive predicates adequately. One operator bounds its argument (an unbouned predtcate} by maxirnahry, the other one by measuring its duration. Theorencally, one might do wtth one intergresstve operator only, but this would yield complicated and not intuitive representa tions. Now that I have presented my proposal for a classification of aktionsart in detail, I shall devote the next secnon to the parallels between aktionsarten and the distinctions of different types of nouru. My aim will be to fit in the intergressive aktionsart with these parallels.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
{s6) maps predicates on times P on the set of times t such that P holds for t and t is m:oornal with respect to the predicate 'its length is smaller than or equals a given value n'. In this decomposition the bound1ng effect on the S-quantity is mimicked by the maximal phase t of the predicate l t [Qu(t) :!:;; n] : as the length of all submtervals oft is smaller than n while the length of all supenntervals oft is larger than n, t's length must be exacdy n. In addition, { s6) refrains from restricting the environment of t P may be valid immediately before and after t but need not be. Hence, (s6) 1S not subject to the cnticisrns levelled against (s I ). S-quantities can be rranslated into expressions of the type { s6} by rule (57):
�rkus Egg
6
341
I N T E R G RE S S I VE P RE D I C A T E S A N D T H E A L G E B R A
OF Here I will demonstrate that
EVENTS
ifw e assume a n mtergressive aktionsan the match
between the partition of verbal predicates in aktionsanen and the classification of nouns becomes closer. Since (as Bach
I 986 has shown) the parallels between
the classifications in the nominal and the verbal domain go very far, one could expect there to be a subdivisiOn in the nominal domain that corresponds to the intergress1ve aktionsan. The existence of such a group of nominals would be additional evidence for the intergressive aktionsart.
mass nouns, three groups of nominal predicates can be distinguished: - All membets of the first group E ' denote individuals of a group E They are atomic, that is, they apply only to individuals which do not fall below a cenain predicate-specific lower bound. E is closed with respect to a suitable jom operation which represents the fusion of (sums of) atoms only. - The second group A' is the set of all nominals which denote atoms in E (for instance, bear). E '\A' is the set of all pluralic nominals like bears, which represent sums of single things (plural objects). - The third group D ' is a subset of A' whose members represent 'portions of matter'. The members of A'\D ' stand for single thmgs with inherent boundaries (e.g.
bear, bur nor a pint ofbeer). D constitutes a complete join
semilartice wtth material pan relation � and material join um·
Unlike mass nouns, the elements of D ' denote individuals. In the ontology on which this classification is based one can individualize matter by singling it out The two ways of singling our matter are mensuration or postulating maximality. Under my definition, D ' is the group corresponding to the intergressive predicates. Both groups share a number of properoes. Elements of the group of norrunals D, like five pounds of meat, are bounded (- denote ennties with boundaries). In their semanncs, a bounding operator was imposed on an unbounded nominal. E.g. meat itself introduces no boundary, but five pounds does. Consider the decomposition offive pounds in Krifka's analysis
( I 989: 64) (pound ' denotes a function mapping objects on to
the number that is their weight m pounds):
( s8)
A.P A.x(P(x)
1\ pound'(x) -
s)
The same procedure can be assumed for one group of intergressive predicates. This has been shown by
Kri£ka ( I 992: 8 J, 98);
it is reflected in the parallel
structures of his decompositions for durative adverbs like in
(S3) and
measure
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Consider for instance a classification of nominals that follows the classifica tion of the domain of individuals in Lmk ( I 98 3). Apan from nominals denoting
34-2 The lntergresstve
as a
New Category of Verbal Aknonsan
(59) Max drank two beers/many beers (6o) (- (43)) Max was in London twice/repeatedly The noun of the object NP in (59) must be interpreted as bounded; in the same way (6o) is understood as denoting a number ofstays of Max in London each of which is bounded. In both cases, the exact boundaries remains unspecified. There is yet another dlSnnction of the nominal domain which carries over to the verbal predicates, the distinction of different kmds of bounding 'portions of matter'. Apart from the set of predicates represennng measured 'portions of matter' there is an additional group of predicates denoting bounded 'portions of matter', namely, maximal objects consisting of all quantities of matter of a certain kind (in the respective context). The characteristic means of referring to such an object is the definite article applied to a mass noun. (6 I ) is a representa tion of the definite article for mass nouns in the style of Link ( I 98 3: 3 I 8). The relation x T y is true iff the denotation of x is nonempty and a material part of the denotation of y. (6 I) A.Q A.P 3x(Q(x) I\ Vy(Q(y)
_.
y T x) I\ P(x))
{6 I) involves the maximal object belonging to the extension of Q (the representation of some mass noun) in the given situation. But both kinds of predicates denoting 'portions of matter' have a verbal counterpart as has been argued for in section s. there are two different kinds of intergressive predicates, which denote either a measured or a maximal period for which a state or process predicate holds. Thus, by assuming an intergressive aktionsart one can extent the parallels between nominal and verbal classifications even to these very subtle details. In the light of all these parallels, I conclude that intergressive predicates are the verbal counterpart to the set D' of predicates representing 'portioru ofmatter'. Coruidering the close parallel between the classifications in
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
phrases as ( s 8). An example of this group is (3) above: this predicate consists of an unbounded predicate (Ame1ie sing) and an operator (forfive hours) which imposes a boundary on the predicate. Due to their boundedness, both the intergressive predicates and the elements of D' denote countable things. If one pushes the analogy even further and distinguishes events and the states or processes they are 'made from', inter gressives also constitute a complete join semilattice (the 'bits of process With join uP and partial ordering �· mentioned by Bach I 986: 8). Moreover, state and process predicates are interpreted as bounded predicates (as predicates denoting maximal periods for which the uninterpreted predicate holds) if they appear in count or iteration contexts. The same applies to mass nouns: in count contexts or if modified by appropriate determiners they are understood as referring to 'portions of matter' bounded in one way or other. Compare the following sentences:
Markus Egg
343
the domain of verbal and nominal predicates, the match between D' and the intergressive predicates is funher evidence for the existence of the class of intergressive predicates. 7
EXTE N S I O N O F THE I NTERGRE S S I VE O PE R A T O R TO OTHER D O M A I N S
(6 2) A. y A. x A. t LOK(x, IN•(y))-phase( t) Due to the defirution of the S-phase, (62) indicates for given y, x, and t that there are intervals of time before and after the phase t during which the external object x is not yet or no longer located in y. This entails that the object x enters, traverses, and leaves the region derived from the internal object y by the function IN• (which usually maps entities on the convex closure of the space occupied by these entities) . Restrictions on the path covered by x or constituted by the shape of x as e.g. in Kaufmann's (1 990) analysis of the German preposition durch , the equivalent of through , are not necessary.20 But as soon as one tries to imagine possible scenes whose descnption mvolves a through-PP, as, say, in to walk through the park, the representation (62) seems to be too simple. Should not one include an additional restriction to the point that the length ofthe moving object's path must not fall below a certain length or that the path must divide the object in two halves of approximately equal size? I do not question these intuitions, but I want to argue that they represent pragmatic infer ences: they can be cancelled by appropriate hnguistic means without getting into a contradiction. E.g. the VP (6 3 ) is semantically well-formed. (6 3) to walk a short distance through the park Like all other directional prepositions, this prepos1oon demands a path somewhere in a dominating phrase conceptually; following Kaufmann (1 990:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This secnon presents an extension of the domain of the intergressive operators. It is possible to generalize the intergressive operator that expresses maximality of its argument. The extension of the domain of this operator is motivated by lmguistic data like the preposition through and verbs like to cross and their wide range of temporal and non-temporal usages. I will discuss the preposinon through in the reading of the preposition that denotes the complete intersection or crossing of its internal (governed) argument by its external argument.19 The temporal reading of this variant of through is a prototypical intergress1ve predicate: it denotes the maximal phase of a locative relation between its external and its internal object. This reading can be given the following decomposition. For the sake of readability, I will omit the additional A.-abstracted parameter of the locative relation in this section:
344 The lntergressive as a New Category of Verbal Akrionsart
(64)
a.
A.P lx l �MOVE(x) A WALK*(x) A P(x))(t) I +DIR
b. AP Ax l�(MOVE(x) A WALK*(x))(t) A P(xXt)) I
+DIR
c. Ax A�(MOVE(x) A WALK*(x))(t) A LOK(x. IN*(P))- phase(t) d. Ax J.�MOVE(x) A WALK*(x) A LOK{x. IN *(P))-phaseXt) For given x. (64d) denotes the set of all intervals which are maximal with respect to the locative relation of x in a gtven park, and hence involve a complete traversal of this park. In addinon, the process of x's walking must hold for the intervals. It ts irrelevant for the meaning of (64d) whether x walks before it enters the park or whether it keeps walking after lt has left the park; the only thing that matters is that it is walking while it is localized inside the park. On the other hand, there are also non-temporal readings of through. For instance, through -PPs may modify nouns which denote objects whose shape constitutes a path: (65) the road through the park (65) cannot stand for maximal mtervals of time during wluch a whole road is in a (contextually given) park. Rather, u denotes the pan of a road whose extension is maximal wuh respect to the property of being located withm the park, i.e. the pan that completely traverses the park. Readings like (65) suggest that mtergressive operators are not resrricted to the temporal domain. If it is possible to find a suitable generahzation of the operator, this will yield funher evidence for the nature of bounding as a pervasive phenomenon in language. In order to account for examples like (6 s), a generalization of the decomposition (62) will be proposed.22 Firstly, I will generalize the S-phase operator in a suitable way in order to extend its applicability to non-temporal
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
1 8), it is the head of the next dominating maximal phrase. Verbs of movement conceptually involve such a path. (64a) is the decomposition of to walk. Note that, according to tlus decomposition, verbs of movement hke to walk subcategorize for a directional preposition. This is expressed by a categorial resrriction '+ DIR' on the abstracted predicate P. (This analysis of movement verbs of course merits further discussion, but in this paper I will not pursue this topic.) P abbreviates tx(park'(x)). Via the reformulation of the verbal decomposinon (64b), the functional application of the decomposition of through the park and reformulation of the result (64c) we arrive at (64d). The step from (64a) to (64b) and from (64c) to (64d) presupposes that the extension of A�a(t) 1\ P(t)) is the same as the extension of A�a 1\ P)(t).21
Markus Egg 34 5
predicates. The temporal variable t of definition (44) is substituted for by a variable representing entities of an unspecified dimension. In addition. the hidden quanofic ation over the parts of the dimenswn m the defrmnon of the S-phase is made explicit.23 The new operator is called I NTER: {66) V� VS(INTER(d1, S) - Vd(d � di .... S(d)) 1\ Vd.,(d1 -.Vd(d � d., -+ S{d))))
c
d., ....
This decomposition can be paraphrased as follows: (67) is valid for given x, y, t iff during t the dimension D has a maximal phase d1 for which and for whose parts a location relation between an {Im-)proper part of x and the internal argument of the preposition y holds. {67) is given lts fully fledged Interpretation in the context only. To illustrate this decomposinon. the derivations of to walk through the park and (6 s) will be expounded. Through modifies {either by functional application. as in {68), or by 0-role-idennfication. as in {69)) unbounded verbal and nominal predicates whose semantics contains a path conceprually. Walk and road both belong to this group of predicates. Consider first the derivation of walk through
the park:
{68)
a.
A.P A.x A.r(MOVE(x) 1\ WALK* (x) 1\ P{x)Xt) I +DIR
b. A.P A.x A. r((MOVE(x) 1\ WALK* (x)){t) 1\ P(xXt)) I +DIR
c. A.x A.r({MOVE(x) 1\ WALK*{x)Xt) 1\ A.x ). t I NTER(D, 3z{z � x 1\ LOK{z, I N* (P)))Xt){xXt)) d. A.x A.r({MOVE(x) 1\ WALK*(x)Xt) 1\ INTER(D, 3z{z � x 1\ LOK{z, I N *(P)))Xt)) The head of the maximal phrase which immediately dominates the PP is the verb itself It specifies the relation � as identity; hence, the quantification becomes redundant and can be omitted in (68e). The verb also determines D as temporal, which yields {68D [- {64d)): e. A.x A.r({MOVE(x) 1\ WALK*(x)Xt) 1\ I NTER(D, LOK{x, I N *{P))){t)) £ A.x ). r(MOVE(x) 1\ WALK* (x) 1\ LOK{x, I N * {P))-phase)(t)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In prose, d1 is a maximal phase ofS in the dimension D, ifTS holds for d1 and Its parts whereas every d., d1 is a proper part of has some parts for which S does not hold or is undefined. For temporal D, I NTER(D, S) can be reduced to the temporal operator S-phase. With this new defirution. we an define the decomposition of through directly: {67) A.y A.x A. t INTER(D, 3z(z � x 1\ LOK{z, IN * {y)))Xt)
346 The lntergressJVe
as a
New Category of Verbal Akcionsart
Now consider the derivation of (6s). The decomposition of
road is
(69a).
Nouns are assigned a temporal parameter, too; otherwise, lexemes as former cannot be analysed. (69) a. lx l t(road '(x))(t) b. lx l t(road '(x) 1\ INTER(O, 3z(z � x 1\ LOK (z, IN* (P)))))(t)
As the conjunction of two predicates
is
valid only at times at which both
predicates are simultaneously valid, wide scope of t over both predicates must
be assumed. One may achieve this by lmking (69a) and (67) (after specification of the direct object of the preposition. that is) by a generalized conjunction in the sense of Rooth & Partee ( 1 982) (see note 2 1 ).
()
Hence an object x is a road through the park at t iff I x is a road at 4 (2) for each part of the spatial extension of the road there is a part of the road that is located inside the park, and (3) any elongation of the spatial extension of the road has parts for which there
is no part of the road which is inside the park.z..
It is undetermined whether the road (of which road through tkpark
is only an
(im-)proper part) continues outside the park. It may but need not terminate at
the boundaries of the park. Formula (69b) manages to capture the two possibilities. This is achieved by the defininon of the operator INTER. Let us first consider the case in which the road continues outside the park. This case is depicted in drawing (7o): (7o)
�A -
� o_ · yo
\. .___.....;
D'
_
____,/
�
o· ____v ,...---'� �-----...Dr..; '.._
_____J
D'
The figure with the rounded edges stands for the park, the other figure for the road. The dimension 0 selected in (69b) is indicated in (7o) and is the area with the double hatching. Now, according to the meaning of through, 0 is maximal with respect to the predicate 3z(z c x 1\ LOK(z, IN*(P))), where x is the road.
Hence, there cannot be an elongation 0' of O for every part of which there is a part of the road located in the park. As can be seen from (70), this is indeed the case; although there is a part of the road for every part of O ', this part of the road is not in every case situated in the park.Hence, 0 ' is not maximal with respect to this predicate, as the predicate does not hold for all of the pares of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
This rime, the head of the next maximal phrase dominating the PP is the noun road. It fixes the conceptual parameter 0 and the relation between z and x as the spatial extension of the road and the relation of proper part, respectively.
Markus Egg 347
8 I NTE R G RE S S I VE PRE D I C ATE S A N D A GA IN
In this section, I will present one of the consequences involved in the recogni tion of bounded yet atelic predicates. The reclassification of verbal aktionsan is no isolated phenomenon penaining merely to the behaviour of predicates with respect to cenain tests for aktionsan but has repercussions elsewhere in the language system. E.g. established semantic interpretations of other lexemes may be in need of revision in order to extend their applicability to intergressive predicates. As an illustration, consier the adverb again and its semantic analyses. Accounts of the meaning of again in terms of semantic decompositions as a rule involve three intervals. The argument of again, a predicate, must hold for the first and the third interval but must be invalid for the second interval (called the 'intermediate interval' from now on). Consider e.g. the analysis of again by Dowry (I 979: 26 I ): (72) VP D(again ' (P) - {"P A P(-.vp A PYp))) Following (72), again P is true iff p holds now, did not hold at some time before now, but held at a still earlier time.26 Others have imposed funher restrictions on this decomposition. Consider e.g. Fabricius-Hansen's (I 975: 48) (simplified)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
dimension. On the other hand, it would not be possible for through to select a dimension D' smaller than D in (7o): D' IS not maximal with respect to the predicate 3z(z C x A LOK(z, IN*{P))); there are larger dimensions (e.g. D) for all of whose parts the predicate holds. In other words, with the help of INTER we carve out exactly that pan of the road which traverses the park. Now we turn to the second case, in which there is no road outside the park. The situation is illustrated by drawing (71 ). The figures represent the park and the road, just as in (7o). Again, the meaning of through selects the dimension D and thus singles out exactly the pan of the road which traverses the park: D is maximal with respect to the predicate 3z(z c x A LOK(z, IN*(P))), where x is the road. Any elongation D' of D has parts for which this predicate is invalid because for them there exists no pan of the road. So the proposed decomposition of through is in accordance with our intuitiotlS on (71 ). Note that a o· smaller than D cannot be selected by through in (71 ): D ' is not maximal with respect to 3z(z c x A LOK(z, IN*{P))), as this predicate applies to the larger dimension D, too.n To sum up the discussion of this section, a formerly purely temporal intergressive operator may be extended to the nominal domain. This operator bounds temporal and nominal predicates with respect to maxtmaliry. Therefore, this operator is a good candidate for one realization of a 'universal packager', which bounds ennties both in the verbal and nominal domain.
348 The lntergresstve
as a
New Category of Verbal Aknonsart
analysis of again. Here again P is true for an interval t iffP holds for t, did not hold at the interval t ' which immediately precedes t, but 15 true at another interval t• which precedes t '. A denotes adjacency (see ( wa)): (73) A. P A.t[P(t) 1\ 3t '(t ' A t 1\ -.P(t ') 1\ 3t .[t • < t ' 1\ P(t .)] )] Obviously, the motivation for the inclusion of such an mtermediate interval m the analysis of again was the applicability of the adverb to unbounded predicates: (74) Ariadne 15 happy agam (75) Felix played cricket again
(76) Korutantin ran a mile again The sentence is true ifKorutantin ran a mile sometime in the past, and if there was an earlier time in the past during which Korutantin ran a mile, too. It is irrelevant for the truth of (76� whether Korutantin took a break between the two ruru or not no intermediate interval must distinguish two mtervals (for which the argument of again holds) as opposed to the case of unbounded predicates, since each interval for which run a mile holds is bounded. On the other hand, no intermediate interval follows semantically if again modifies an intergressive predicate as in (76), as this predicate does not restrict the environ ment of intervals in tts exteruioiL
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
These sentences are not true if there are two adjacent intervals of Ariadne's happiness and Fehx's playing cricket, respectively. The intervals must be mterrupted by another interval during which Ariadne was not happy or durmg which Felix took a break. Since again is also applicable to change predicates, the decompositioru of both change predicates and adverb must be compatible, too. The specilication of an intermediate interval is redundant in these cases: every change predicate expresses a change of state, which can be characterized as, say, a change from of -P to P. This change predicate holds for a temporal interval t only if -P holds for a t ' that lies immediately before t (see the definition ofBECOME in section 4-1.3). Immediately after the change has taken place, the predicate P must hold. Hence, a change predicate can only apply to two different intervals of time if the validity of A.t(-P(t)) is restored between these intervals. That is, there must be an interval between them for which BECOME (A. t(-P(t))) holds. Hence, this intermediate interval trivially fulfils the condition stated in the analyses of again that there be an intermediate interval for which the argument of again is not true: if BECOME(A.t(-P(t))) is true for an interval t, BECOME(P) is false for t But for the one group ofintergressive predicates the analyses of again are too strong:
Markus Egg 349
(77) J.P l t(P - Phase(c) 1\ 3t'{t' < t 1\ P - Phase(t'))) This decomposition is for given P the set of all times t such that a maximal phase ofP holds at t and at a t' before t. Since bounded predicates are themselves maximal phases, (77) does not force or preclude a gap in between t and t' in this case. On the other hand, for all unbounded predicates there must be a gap in between t and t'. Note that the two maximal phases cannot overlap, which is due to the definition of the precedence of intervals in terms of the precedence of their points of time (see (9)). To sum up this secnon, it has been possible to adapt the interpretation of again to deal wtth intergressive predicates as well. However, in order to account for the interactions of the adverb with intergressive predicates, the resulting semantic representation of again differs constderably from the decompositions assumed m other analyses. This shows that the intergressive predicate is by no means a peripheral phenomenon oflanguage.
9 CONCLUSION In clus paper I discussed expressions like to play a sonata, to hiccup, and to singfor five hours. As the denotata of such predicates (verbs, VPs, and sentences) have fixed boundaries but introduce no change of state, these predicates pose difficulties for other systems of classtfication: if classifications define event predicates as involving a change of state, these predicates cannot be classified at all, in classifications that define event predicates as predicates whose denotata have fixed boundaries; they are grouped together with predicates that involve a change of state in spite of the distinctions between these two kinds of event predicates. I suggest an extension of Dowry's (1 979) framework that evades these problems. This extension is based on three features (interval-based, bounded, and relic).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
lntergressive predicates that consist of a maximal phase pose no problem for the analysis of again in terms of (73): according to (44), each maximal phase is surrounded by intervals for which the respective argument of the S-phase operator does not hold. Hence, any two maximal phases of the same type are separated by an interval which is not such a maximal phase. Thus, the first type of intergressive predicates presents a challenge to the analysis of adverbs like again. My suggestion for an adequate way of accounting for the data employs once again the definition of the S-phase, but with the conjunct which tests the divistvlty of the predicate (Vt'(t' � t -+ S(t'))) left out. Then the operator is applicable to.OOunded predicates, too {as idennty mapping). Now the decomposition of again can be formulated easily and succmccly:
3 50
The lntergresstve as
a
New Category of Verbal Almonsan
Each feature partitions the group of predicates with respect to whether a predicate possesses the respective feature or does not As the features build up a hierarchy (possession of a higher feature entails possession of all the lower ones), and as there are no predicates which are unmarked for one or more features, the result is an ideal ordering. The feature hierarchy is reflected in the formal definitions of the features, which allows for the logical derivation of these relations. The crucial factor is the distinction between the two features 'teliciry' (the introduction of a change of state) and 'boundedness' (the inrroducnon of boundanes). These features distinguish four �,purually non-overlapping groups of
predicates on the other a new group of predicates has been assumed. This group
have been labelled 'intergressive' and constitute the link between relic and unbounded predicates. Punctuality is regarded as minor feature in the system since not all predicates possess a fixed value for this feature. In addition. the distinction in achievements and accomplishments is independent of punctuality or any other temporal property of predicates; hence, it is not included in the present system of aktionsarten. The semantic representation of the intergressive followed Dowry's (1 979) proposal: more complex predtcates are broken down into less complex ones which are linked by suitable operators. To this end, two operators (of Galton 1984 and Herweg 1 990), which turn unbounded predicates into bounded ones by maximizing the interval for which they hold, have been expounded. Herweg's operator
has
been adopted as intergresstve operator.
An additional
intergressive operator (the S-quantiry), which bounds unbounded predicates by measuring the interval for which they hold, is introduced. The parallel between the distinction between different kinds of norninals and between different kinds of verbal predicates carries over to the distinction between relic and atelic bounded predicates: The group D', which comprises norninals denoting bounded 'portions of matter' (elements of Link's ( 1 983) subgroup of mdividuals D), correspond to the intergressive predicates. Moreover, the two kinds of bounding matter (mensuration vs. postulating maximality) reappear in the set of intergressive predicates. For the first intergressive operator it has been shown that its domain may be extended beyond temporal intervals in order to account for the different uses of the preposition through in terms of one single decomposition. Finally, as an example of the far-reaching consequences of the recognition of an intergressive group of predicates, a semantic analysis of the adverb again has been sketched. In the light of all the evidence discussed in this paper, I conclude that the intergressive is not merely an indispensable member of the set of aktionsarten but additionally a pervasive phenomenon of natural language.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
predicates (state, process, intergressive, and change predicates). In between the unbounded state and process predicates on the one side and the relic change
Markus Egg 3 S I
Acknowledgements I WJSh to thank Michael Herweg, Adin Lalun, Peter Pause, Amim von Stechow, and two anonymous reviewers for therr valuable comments. Tlus work was supported by Grant I S 341' 93 &om the Deursche Forschungsgememschaft. Received: I 2.05.94 Rcvtsed version recetved: 27.02.95
N O TES The tenninology takes mto account the fact that verbal projections and sentences have their own aktionsan that may dtffer &om the aktionsan of their respective verb. Note that often the term 'aspect' is used to refer to aktionsanen. 2 Note that Kri£ka uses 'event' rufferendy: m his work, the group of event predicates excludes only stative predtcates. His analogue to my 'event predicate' IS 'quannzed predicate'. The expression 'mtergressive' IS taken &om LOhner {I988: I S S� who used it for predicates denoting 'mterruptions of an unmarked state'. These predicates belong to the group of predicates which denote rruoomal phases of the validity of a state or process predicate (see note s� 4 Herweg's terrru nology differs &om the one used in this paper. He defines 'hetero geneity' as a combinaoon of 'anocumu larivtty' and 'antidutributivtty' ( I 99
chies. In these hierarchies, features need apply only to pans of the hierarchy that have yet to undergo further dtvislOn. Hence, such features may not apply to pans of the hierarchy that have already been ideno6ed and separated. For instance, Mourdatos (1 978) separates process predicates like to walk &om event predicates such as to win by the critenon of boundedness, but it is of no avatl whether this cnterion applies to state predicates at all (which have already been tdenti6ed as a separate group� Therefore, It would be mlSleading to present my classification in the form of a hierarchical diagram. 6 ThiS follows &om my analysu of all sentence radtcals as predicates on times. Some analyses Qike Bach 1986) distin guish events and the 'process matter' events are made o£ In these analyses dif ferent viewpomts as exemplified by {16a) and (16b) are not the comequence of dtfferent conceptualizaoons of one single Situation but result from the choice between refernng to two materially tdenrical yet different entities. Such an ontology allows for a closer match between the properties of predtcates and the properties of the entines they denote.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
MARKUS EGG lnstiiiiUfor Logrc & Lirrguistrcs r IBM IXut.schlarrd Irformationssystmle GmbH Varrgerowstr. 1 8 691 1 5 Heidtlbng Gm!larry e-mail:�hridtlbg ibm.com
3 5 2 The Intergress1ve as a New Category of Verbal Aktionsan 7
In the followtng secnons, the definioons are
every relevant time dunng the move
given for propos�
ment. However, some process predicates
tional predicates (properties of nmes� All
cannot be analysed in terms of cononu
of akrion.sarten other
predicates P
must
undergo
ous changes mtumvely, e.g to
a
wluch bnngs out thetr mterval-based
akoonsarten de6ruoons to apply to them:
character
ifP IS an n-place predicate, 611 in the first n- 1 arguments wtth distinct vanables of the
type
appropnate
(proposmonal
varubles must be st:aove). line of argumentanon
8 Thu
style of to
and
the
Boundedness differs from non-divmvtty and non-cumulativtty: boundedness implies either of these properties but
may be
neither of them Implies boundedness. Indeed, some unbounded predicates are
performaove
predicates, too. In the first case, the repor ter uses a 6ctioous present (wluch lags
this pomt m orne. Thus, when he makes
divmve and not cumulative Qike to drink fnv bouks of b«r� not diVISIVe and cumulaove (to drink much b«r� or, worse soli, netther cumulaove nor diVISive (to workfrom two tofiw hours� The diSoncoon
has mformaoon about the orne
cates by the feature 'bounded' apphes to
behind the actual present a couple of seconds) and pertams all ius utterances to pred1caoons about this fictitious present,
he
Immediately
followtng
this
of event predicates and non-event predi these predicates
ficooous
present, wluch enables !urn to attribute
as
well and asstgns them
punctual mterval-based predicates m the
to the latter group. ThiS counters one of Moltmann's (I99I: 647fi) arguments
sunple present to
aga�nst the relevance of diSttnguulung
this 6cooous present.
two groups of predicates by such features.
For performaove predicates, perta�rung them to the present by uttering them in
their simple present tense form IS possible
because of their special nature:
as
their
aftermath (e.g. a certam obhgaoon for the speaker m the
case
of
to prom�)
IS
brought about by the very utterance of
II
Addmonal resmcoons on cumulaovtty and dlVlslVlty are necessa ry, namely, the eXIstence of suffiaently many (at least two) elements m the do=n of a cumula nve predicare and the eXIStence of proper parts of en noes that belong to the domam
the predicate, the speaker has full control of this aftermath and knows that It Will
of a diVISIVe predicate By these restric
set m mvanably by uttering the predicate.
oons Knfh (I989: 45. 195) diSonguiShes stnct diVISivtty (SDIV) and strict cumu
Therefore, the truth of the predicate
lativity
apphed to the present can be assessed
fulfilment of cumulanvtty and diVISivtty:
wtthout having to walt for the aftermath
(I) VP (SCUM(P) - CUM(P) 1\ 3x 3y(P(x) 1\ P(y) 1\ x y)]
of the predicate to set
9
to decompose them in the
rrmain, that IS, as the exclusiOn
m.
Thus, no other
(SCUM)
and excludes a mvial
understanding of perforrnaove predicates in the sunpie present is necessary.
(u)
an mterval t, then a change of state that
Otherwue, cumularivtty and diVISivtty
If an mdefinite change of sme holds for
carmot be spect6ed exactly rakes place at every
part of t (down to a predicate
specific lower bound). E.g. movement breaks down mto a series of changes of locanon of an object, but It IS impossible to specify the exact location that IS left at
VP (SDIV(P) - DIV(P) 1\ Vx(P(x) 3y(y C x))]
would
mvtally be saosfied even by
change predicates hie
to dit.
any two
dytng situaoons of John are idenocal, hence conjuncoons denotmg
of the predicate
thiS siruaoon boil down to the
predicate Itself (due to the Idempotency
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
tense'
IS
of a change away from being asleep
10
extended to cover the 'sports reports present
sl«p. The
only analysu of these verbs I can thmk of
Taylorian ( 1 977: 2o6) process of filhng in all but one argument posmon for the
Markus Egg 3 5 3
14
15
16
17
18
from the state 'the predtcate has never been the case' to the state 'the predicate has been the case at least once' (as in Haw you (WT bmt to London?� Durative adverbs mrroduce the bound anes on thetr modtf1ed unbounded predicates by themselves. One cannot assume an intervening process of type coercion that bounds these predicates and thereby makes possible the modification of unbounded predicates by durative adverbs: accordmg to such an :usumpnon, duranve adverbs would apply to loncally bounded predicates, too, which they do not. The tteranve readtng of some atehc punctual predicates (Lke (36}} has become fatrly convennonahzed. Therefore I put the I stgn m parentheses. But even some of these predtcates rrught have only a preferred value for punctu ahry Constder e.g. sentences hke Tk army mtmd tM tarit.ory of tM mnny in two days and Suddmly, tM magtcian cr(a((d a work of art by a mag�c 1p(ll. The orne frame adverbtal can measure the length of the struanon m the first sentence; m the second sentence, the predicate need not be remterpreted tngresstvely Hence, m these contexts the predicates have exactly the opposite value for punctuality as m thetr preferred readmgs. Note that both events share a causanve structure (something hke Jolut rut some keys, wluch brought about the existence of . . .'). Tlus clearly shows the classifica tory and terrru nologtcal mix-up, which resulted &om the confusion of temporal extension and complex structure of a predtcate on the one stde and puncruahry and stmple internal structure on the other. The group of these :md sirruJar construc tions differs from the group of mtergres stve predtcates as proposed in this paper. Hence, Gal ton's ( 1984= 66 £) and my use of the term 'atelic event predicate' must be diSnnguished. (Galton wes the term for this group of constructions.)
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of the conJuncnon� Thus, the predicate would wrongly be assigned cumulanvity. As dying sttwnons are not percetved as mvolving proper parts, the predicate holds for every pan of ruch muations (as there IS none), which would entail divistvity (Molrmann 1 99 1 : 637� 12 One cannot denve the present perfect form from the present progresstve form, stnce the baste mearung of the perfect of an unbounded predtcate 1mphes that the progresstve of the predicate does not hold for the moment of utterance. E.g. Fritz has run tmphes that he 1s not runrung now, wruch IS mcomparible With Fritz IS running. 1 3 The perfect of a tehc predicate P may speafy dtfferent kinds of mtervals tmmedtately following an mterval for wluch P holds. While the perfect of P usually stands for mtervals for which the consequence of P holds, tt may also denote longer mtervals, for parts of wluch this consequence does not hold Consider e.g. the predtcate john fall, whose conse quent state IS John It( on th( floor While the perfect of John fall usually denotes nmes for wluch the consequence of John's falhng holds, there are other opnons: if a wound on John's forehead ts pragmancally connected to a fall ofJolut, john has Jalkn down apphes to mtervals, too, for parts ofwhtch the consequence of Johnfall does not hold; hence, the perfect does not always express the consequent state of the event denoted by the predt cate. (The example IS taken from Herweg 1990: 1 82 £). The same caunon IS called for in the interpretation of the to M going to-prospecnve. Hence, the charactensoc of tehc (as against atehc} predicates IS not that a companson of the penphrasnc prospecnve and the perfect of the predi cate must reveal a change of state involved m the predicate meanmg but that 1t can do so. The 'expenmental' or 'existential' perfect IS of no avail here. This use of the perfect apphes to all predicates and is Interpreted as rransirion
3 5 4 The Intergress�ve as a New Category of Verbal Akrionsart 19 It is alro possible to genetahze the intergressive operator S-quanory. ru in the case of the S-phase, there IS lmguistic evidence for such an extension of the domain of dus operator, for mstance the reading of the lexemefor that subcategor izes for a measure phrase. Such measure phrases may apply to non-temporal dimensions, too, as m for fiw miks. For lack of space, however, I can only present the general.iz.toon of the S-phase m dus 20
paper.
(1) If the two arguments ofn are of type t (true or false), n IS the same as the standard conJuncoon 'A'. (ii) Iff and g are two functions of type (a, b), fn g is a function of type (a, b) that maps every element x of type a onco an element f\x) n g(x) of type b, that IS ro say: f n g - A.x(f\x) n g(x)�
22 The generalization of the S-phase to INTER IS mnilir to Wunderlich's ( 199I ) generaliZation of the operator BECOME. Wunderlich took the resulting operator CHANGE (wluch denotes all kinds of change of state) to descnbe both temporal and non-temporal readings of the pre posmon into in one single composinon (to wallt into thr city, tM road into thr city� 23 Thu quanrificanon shows up m the tlurd pan of the definition of the S-phase: the predicate must fail for all t' larger chan t This fatlure can be explained thus: Every t' has a patt t" for wluch the predicate does not hold. But since t IS also pan of every t', t' Ius also a pan for wluch the predicate holds. Hence, the reason why
(i) \IS 'Vt(S-phase(t) - S(t) A 'Vt"(t" � t - S(t")) A 'Vt'(t c t' - -.\ft"(t" � t' - S(t"))))
24 One rrught thmk that NPs like thr road through thr park refer to the whole road, regardless of whether It extends outside the park or not However, dus would entail that a sentence hke (1) meant that the speed lirrut extended beyond the Vlllage, which IS not the case: (1) The road through the Vlilage speed litrut
has
a
25 To rule out inacceptable examples branch through the window, an addmonal conceptual resmcoon IS necessary wluch blocks the modification of nouns by directional PPs unless the path consn tuted by the shape of the object denoted by the noun is involved m the funcoon of the noun. 26 Dowty's formula has the operator H Instead of P. But this formula IS contra dictory accordmg to the standard de6ru tion of H;(-.P--.,) ObVlously, the two Hs are rruspnnts since Dowty's prose translaoon of the formula involves exist enoal and not uruversal quannficanon over rimes m the past t C (whose first COil.JUOct IS, smcdy speak ing. superfluous here) excludes trivial ful6ilment of TEL. E.g., following Herweg (I 99 I: 64 £� the property of immed12tely following an interval in the extension of an event predicate P would be a P' that could make rme TEL even for an atehc P. The corresponding P would be the property of immediately precedmg a member of the exteruion of P. I do not think that these two propemes as defined by Herweg model the semantics of the prospective and the perfect tense, respect ively, as it is fine to utter (i), maybe in the •
-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Admittedly though, her decomposition of durchlthrough takes into account more readmgs of the prepostnon than (62� E.g. her analysis also encompasses the reading of the preposmon that denotes mcom plete mtersecrions. 21 The formal details of this more general nonon of conJuncoon have been spelt out by Rooth & Partee (1982� where the generahzed conjuncoon 'n' IS defined recursively:
the predicate does not hold for t' IS that the predicate does not hold for all pans of t'. Therefore one rrught reformulate the definition of the S-phase equivalendy as (!):
Markus Egg 3 55 context of a sweanng and panting runner who gets ready for a race:
(1) Nigel h as ru n 10,000 metres and he IS gomg to run (another) 10,000 merres.
RE FERE N CE S Armstrong, DaVId ( I 9 8 I� 'The Anaent Greek aonst as the aspect of countable acnon', 10 Philip Tedeschi & Anrue Zaenen (eds� Syntax and &mantics 14: T= and��. Acaderruc Press, New York, I l l.
3 1 -64-
Davidson, Donald (1967� 'The logtcal form of acnon sentences', m Nicholas Rescher (ed.), T� Logic of Dtcision and Action , Uruversiry of Plrnburgh Press, Plrnburgh,
8 1-95· Dowry, DaVId (1979), Word M(aning and Monl4gu( Grammar, Reidel, Dordrechr. Egg, Markus (1994� Aktionsart und Kompo sitionalittit. Zur kompositiondlrn Abkitung dn Aktionsart kompkxn- Kaugonrn , Akade rrue-Verlag, Berhn. Fabncms-Hansen, Carhrine (1975), Transfor rnatiw, intransforrnatiw und kursiw V"brn, Niemeyer, Tiibingen. Galton, Anthony (1984� 1k Logtc ofAsp«t: An Axiornaflc Approach, Clarendon Press, Oxford. Herweg, Michael (1990� ZLrlaSp(ku: Dr( &rkutung wn T(tnpus, �Itt und tmrpor alrn Kanjunktionrn , Deutscher Universi cits-Verlag, Wiesbaden. Herweg. Michael (1991� 'Temporale Kon junknonen und Aspekr: Der sprachliche
90-
Kaufinann, Ingnd (I990), 'Semannsche und konzepruelle Aspekte der Weg-Priiposition durch', Kogmtionswissmschaft, 1, I 5-26. Krifh, Manfred (1 988� 'The relanonal theory of genencuy', 10 Manfred Knth (eel), GrnmCJty 1n Natural Langua�, Pro c�ings of t� 1 988 Tubmgrn Conf"rna, Uruversicit Tiibmgen, 285-3 12. Kn£ka, Manfred (I989� Nominalrif(m!Z und Zdtlwnstitut1on: Zur &mantiIt von Massrn((r rnrn, Plural/((rrnrn und �ktklassrn , Fmk, Miinchen. Kn&a, Manfred (I992� 'Thematic relatioru as holes between norrunal reference and temporal corutirunon', in Ivan Sag & Anna Sabolcsi (ech� Ltxical Mattm, CSLI, Stanford, 29-5 3· Lmk, Godehard (1983), 'The logtcal analysiS of plurals and mass tenns: a laroce theoretical approach', 10 Ramer Bauerle (t al. (eds), M(aning, u� and Intnpmation of Language, de Gruyter, Berhn, 302-23. LOhner, Sebasnan (I988), 'Ansatze zu emer mregralen semannschen Theone von Tempus, Aspekr und Aktionsarren', in Veronika Ehrich & Heinz Vater (ech� T(tnporalsmrantik: &itr.igt zur Linguistik dn Zdtrtj"tnZ, Niemeyer, Tiibingen, I 6J-9I. Moens, Marc & Steedman, Mark (1986� T(tnporal Information and Natural Langua� Proctssing, Research Paper RP-2, Centre for Cognitive Saence, University of Edin burgh. Moens, Marc & Steedman, Mark (1988� 'Temporal ontology and temporal refer ence', Comput4tional Linguistics, 14, J-14Molrmann, Friedenkt (1991� 'Measure adverbials', Linguistics & Philosophy, 14,
62<)-6o.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Bach, Emmon (I986), 'The algebra of events', Lmguistics & Philosophy, 9, 5-16. van Benthem, Johan ( 198 3 � 1M Logic ofTim(, Reidel, Dordrechr. B1eiW1SCh, M:mfred (1988� 'On the grammar of local preposinoru', m Manfred Bier wisch n a/. (eds� Syntax, &mantik und das Ltxikon, Akaderrue-Verlag, Berhn, 1-65. Carlson, Lauri (1981), 'Aspect and quann fication', 10 Philip Tedesclu & Annie Zaenen (eds� Syntax and &mantics 1 4: T= and �t, Acaderruc Press, New York,
Ausdruck von Zeirrelanonen ZWISChen Siruanonen', KogrlltJonswissmschafi, �. 5 1 -
3 56 The lntergremve as a New Category ofVerbal Akrionsan Verkuyl, Henk (1 989), '&pectual cl= and aspecrual composinon', Linguistics & Philo sophy. l l, 3 9-94Verkuyl, Henk & Joost Zwarts ( I 992� 'Time and space m conceptual and logical semanncs: the nonon of Path', Lmguutics, J O, 48 3-5 1 I . Vlach, Frank {I993� 'Temporal adverb1als, tenses and the perfect', Lmguutics & Philo sophy, 16, 23 I-83 Wunderlich, Dieter ( I 99I}, 'How do preposl oonal phrases fit mto composmonal synrax and semanncs?' Linguistics, 19, 59 1-
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Mourelatos, Alexander ( 1 978� 'Events, pro cesses and mtes', Linguut1cs & Phdosophy. :a, 97- 1 2 1 . Rooth, Mats & Pmee, Barbara (1 982), 'Con juncnon, rype ambiguity and wtde scope "or" ', m Daruel A1ckinger tt al. (eds� Procudings oftht First Wm Coast Ccnftrmu on Formal Linguistics, Stanford Uruvernry, 3 53-
joamt11l ofSnMntlCS
1 2.
3 57-3 ']6
C Oxford Untwrmy Pr= 1 995
Negation and Disj unction Theory
m
Discourse Representation
E M I E L K R A H M E R and R E I N H A R D M U S K E N S
Tilburg University
Abstract cannot :mtecede an anaphonc element 1f the noun phrase 15, but the anaphonc element IS not,
m
the scope of a neganon; the theory also pred1crs that no :maphonc lmks are poss1ble between
the two pans of a diSJUncnon. However, 1t
IS
well known that these predictions meet Wlth
counterexamples. In pamcular, anaphon 1s often posstble 1f a
doublr
negation intervenes
between antecedent and anaphonc element, and also 1f the antecedent not only occurs m the first part of a diSJUnction but also WJclun the scope of a neganon, wlule the anaphonc element the second part of the same diSJUnctlotL In clus paper we argue that these recalorrant
� m
phenomena are related and that a solunon to the double neganon problem Wlll also proVIde us wnh a solunon to the diSjunction problem. We reVIew the bas1c set-up of claSSical DRT and otTer an extensiOn (called 'Double Negation DRT) wh1ch vahdates the law of double negation An adaptation of the standard DRT consrructlon algonthm wh1ch transforms texrs mto Discourse Representation Srrucrures
IS
sketched and It IS shown that the problems Wlth
negation and diSJ Uncnon that led to the defirutlon of our new version of DRT are properly dealt Wlth
I
TWO PROBLEMS F O R D R T , A N D A RE D U CT I O N I.I
The double negationproblem
In a now classic paper (Karttunen I976) Karttunen noted that while a discourse referent cannot outlive a single negation or a single verb with an inherently neganve implication (such as fail, n eglect, or forget) it will not be blocked by a double negation. While in ( I ) the pronoun it cannot be mterpreted as dependent on a question and in (2) the pronoun cannot depend on an answer, the definite in (3) may depend on the precedmg indefinite and the it in {4) can be taken to refer to an umbrella. The anaphonc pronouns in (s) can likewise be interpreted as depending on the indefinite that precedes them, even though the latter is within the scope of two negations.1 ( 1) Bill didn't dare to ask a question. ---rhe lecturer answered it. {2) John failed to find an answer. *Ic was wrong. (3) John didn't fail to find an answer. The answer was even right.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Classt�l Discourse Represenranon Theory (DRT) prechcrs that an mdefirute noun phrase
358 Neganon and Dtsjuncnon m Discourse Representation Theory
(4) John didn't remember not to bring an umbrella, although we had no room for it. (5) It is not true thatJohn didn't bring an umbrella. It was purple and it stood in the hallway.
(6)
X
x - john y �
(7)
X
�
umbrella y x brought y
y
x - john umbrella y x brought y
Other formulations of the dynamic perspective are confronted with essentially the same difficulty. In DPL, the negation of a formula cp will act as a 'test', irrespective of the internal structure of cp, and so, since the first clause of(s) is of the form -.cp, the anaphoric lmk between an umbrella and it is predicted to be impossible. In FCS we have that the first sentence in (s) does not succeed in extending the domain of the current file, while a new card for an umbrella is needed in order to establish the link between antecedent and anaphoric pronoUJL In chis paper we shall discuss the double negation problem (and the disjunction problem-see below) from a DRT perspective, but the reader will
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Vanous authors2 have pointed out that examples such as (3), (4), and (s) are a problem for dynamic theories of discourse such as Discourse Representation Theory (DRT, K.amp 1 98 1 ; K.amp & Reyle 1 993), File Change Semantics (FCS, Heim 1 98.2, 1 98 3), or Dynamic Predicate Logic (DPL, Groenendijk & Stokhof 1 991). These theories correctly predict negation to be a plug with respect to anaphoric binding and thus fit the factS in ( 1 ) and (2),3 but they also incorrectly predict a double neganon to be a double plug, not a plug unplugged as the factS in ( 3Hs ) would suggest.• In DRT, for example, the discourse referent that is connected to an umbrella in the first sentence of(s) will land up in a Discourse Representation Structure (DRS) that is twice embedded to the main DRS and that will thus not be accessible for future anaphoric reference. An apphcation of the DRT construcnon algorithm to the first sentence of(s) gives the DRS in (6) as an output, while it is the s1mpler DRS in (7) that would give the right predictions here. In the latter, but not in the former, the discourse referent y, which is connected to an umbrella, will be accessible from conditions in the main DRS.
Emiel Krahmer and Remhard Muskens
3 59
have no difficulty in translating our proposed solutions to her favourite dynamic semantic framework. While we chink chat Karttunen's data essentially show chat double negations are holes for anaphoric linking and thus form a problem for standard dynamic accounts of anaphora, it should be noted chat ocher data apparently point in an opposite direction. Consider (Sa), for example, a text chat is decidedly odd if the anaphoric pronoun is interpreted as depending on no guest, even though the latter occurs within the scope of a negation and no ocher relevant operators intervene between the would-be antecedent and its dependent element.
The oddity of (Sa) should be contrasted, however, with the complete acceptability of (Sb) and is due, we conjecture, to a uniqueness effect (c£ Evans 1 977; Kadmon I 9S7). Given some highly unlikely context in which it is understood chat at most one guest can be present at a weddmg (Sa) would be fine. We feel chat it is precisely the unlikelihood of such contexts which explains the markedness of (Sa). Note, however, chat (Sa) is still better chan (9), its counterpart With one negation only. (9) There is no guest at this wedding. •He
is standing right behind you.
Another category of prima facie counterexamples to the double neganon rule is formed by cases where the only plugs intervening between a possible antecedent and an anaphoric element are indeed two negations, but where the two still do not conspire to form an authentic double negation because they sandwich ocher material. We have in mind cases like ( 1 0), whose first sentence should be rendered as the DRS in (I I). (10) No man didn't bring an umbrella. •rt was purple and it stood in the hallway.
(u )
-
.X
man .x
-
y
umbrella y
.x brought y
The truth conditions of (I I ) are exactly chose of the predicate logical formula -.3x(man(x) 1\ -.3y(umbrdla(y) 1\ brought(x, y))) and (I I ) is as much a case of double negations as chis predicate logical formula is. Since such apparent
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(Sa) It is not true chat there is no guest at this wedding. ??He is standing right behind you. (Sb) It is not true chat there is no bride at this wedding. She 1s standing right behind you.
36o
Neganon and Disjunction m Discourse Representanon Theory
counterexamples on closer examination thus rum out to be no counter examples at all, it seems that we can take it to be a general rule that, as far as truth conditions and the possibility of anaphora are concerned, double negations m standard English behave as if no negation were present. 1.2
The disjunction problem
( 1 2) Either there's no bathroom in this house, or it's m a funny place X �
bathroom x in this house x
v 1t 's m a
funny place
Kamp & Reyle (1993)6 remark that it is in fact the presence of a negative element in the first disjunct which seems to license the anaphora in (1 2),7 even though negations m themselves usually block the possibility of linking. If there is no such negative element, as in (1 4), coreference is impossible. ( 14.) ??Jones owns a car or he hides it A second observation made by Kamp & Reyle is that sentences of the form A or B can in general be felicitously paraphrased as A or otherwise B and this leads to a proposal to let the DRT construction algorithm provide for the 'other case'. In ( 14) the 'other case' IS the case whereJones does not own a car, and thus a revised form of the construction algorithm adds a condition to this effect to the second disj unct of the DRS for the sentence. The result is shown in ( 1 s� ( 1 s)
X
x • jones z y v car y x owns y
�
y car y x owns y
z hides z•x
it
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The double negation problem seems to be related to another problem that is also generally thought to be a hard nut for DRT and related theories. In ( 1 2)5 the pronoun it is naturally linked to no bathroom, while DRT and other dynamic theories predict no antecedent in one part of a disjunction to be accessible for a pronoun in the other part. If we apply the standard construction rules to this sentence we get the DRS which is g�ven in (1 3), but in this DRS the pronoun it cannot be resolved as the referent x.
Ernie! Krahmer and Reinhard Muskens 36I
Here, since it cannot be resolved as y, the revised construction algorithm does not lead to predictions different from the original one, but as soon as we turn to sentences like ( 1 2) we see that Kamp & Reyle's remion pays off The 'other case' to be considered now is the case where a bathroom is present and if this information is added to that of the second disjunct we get ( 1 6) at a crucial stage of the DRS construction. This time it is possible to resolve it as x and the link between anaphor and antecedent can be established. (r6)
X
X �
bathroom x in thishous� x
v
bathroom x in this house x
Kamp & Reyle's treatment of'bathroom' sentences can perhaps be cnticized for not being entirely precise, in the sense that their new construction rule does not seem to prescribe exactly what material is to be added to the second disjunct Suppose that we take the rule to be that m constructing the DRS for a disjunction we should add the negation of the DRS for the first disjunct as a condition to the DRS for the second diSjunct Then the DRS associated with (14) would indeed be (r s), but the DRS for ( 12) would be (17) instead of(r6), 1.e. we get a double negation where we want to neganon at all.8 (r7) X �
X �
bathroom x in this house x
�
bathroom x in this house x
v it 's 1n a funny place
Note the structural similarity between the problem how to get from ( I 7) to ( r6) and our previous problem how to obtain (7) from (6). In both cases we should like to be able to erase the double negation. An explicit rule to this effect would be very much ad hoc, however, and would be quite unlike all other DRT construction rules. It would have the useful property of being able to make certain referents accessible to certain pronouns (e.g. the referent x is accessible from it in (r6) but not in ( 1 7)) but this very property would also make it theoretically suspicious for not being meaning preserving. If meanings determine context change potentials, as the dynamic perspective has it, then a rule to erase double negations that would change {6) into (7) (and ( 1 7) into ( 16)) cannot be meaning preserving since (6) gives a context which does nor allow reference to y while (7) gives one which does.9
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It's in a funny place
362 Neganon and Disjuncnon in Dtscourse Representation Theory
There is another difficulty with Kamp & Reyle's proposed solution to the problem of 'bathroom' sentences: (I6) simply does not have the truth conditions that (12) seems to have. Suppose there are in fact two toilets in the house, one of which is, and one of which is not in a strange place; then ( 1 2) is false according to our intuitions, but ( I 6) is true since its second disjunct can be verified.10 We therefore turn to an earlier proposal from Roberts ( I989), who renders ( 12) as (I 8).1 1 The idea here is that the material under the negation in the first disjunct is accommodated to provide an antecedent to the second disjunct. Since the first disjunct gives a negative answer to the question whether there is a toilet in the house, it is natural to interpret the second dtsjunct as pertaining to the possibility that there is one. X �
X
bathroom x in this house x
v
bathroom x in this house x
=>
y funny place y y-x
From a formal point of view 1t should be observed that, in a sense which will be made precise in the following section. ( 1 8) is equivalent to (19), the second disjunct of tts only condition. And indeed, we feel that this is correct, since intuitively ( 1 2) is equivalent to ( 20).
(I9)
X
bathroom x in this house x
(20)
=>
y funny place y y ax
If there's a bathroom in this house it's in a funny place
How can we revise the DRT construction algorithm so that it gives ( I 9) instead of (I 3 ) as an output for (I 2)? Here again we see that if we could but solve the double negation problem we would have a solution ro the disjunction problem as well. For suppose that we would revise the construction algorithm so that at any time that a sentence disjunction A or B is encountered a condition of the form (2 I ) (instead of the equivalent [4] V [[] ) would be added to the current DRS,12 then (22) would be the output for {I4), but for ( 1 2) DRS (23) would be obtained. The first of these is indeed correct in the sense that the anaphoric link is predicted to be impossible, but in the second we have a double negation again where no negation is wanted. The problem how to get from (23) to ( I 9) is formally similar to the problem how to get from ( I 7) to ( I 6) or indeed to the question how to get (7) from (6).u In this sense it can be said that the disjunction problem reduces to the double negation problem.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(I 8)
Ernie! Krahmer and Remhard Muskens
36 3
X
x .. jon�s
�
y car y x owns y
�
�
z => z htdes z- x
it
(23) => tl 's
in a funny place
It thus seems that if we can reviSe the DRT language by adding a new negation which obeys the law of double negations (i.e. whtch allows for cancelling double negations) we may not only solve the problems that we have encountered with Karttunen's 'umbrella' sentences, but we may also be able to deal with 'bathroom' sentences. An attempt to carry out such a revision will be made in section 3 below, but first let us look mto the syntax and semantics of the standard DRS language in some detail. 2 STANDARD DRT: T H E F O RMAL I T I E S
The Double Negation DRT of the next section will be a generalization of standard DRT and for the sake of easy comparison we shall give concise versions of the most important DRT definitions in this one. In fact, we shall extend the standard syntax slightly and add a sequencing operator ';' which takes two DRSs and gives a complex DRS.14 This addition, whtch seems natural in itself, in fact takes us already halfway from the standard set-up to the logic that is discussed in the next section, but nothing here hinges on the addition and a formalization of the core part ofDRT can be obtained by simply omitting all reference to ';'. One of the virtues of the DRS language is tha t DRSs are visually appealing. A disadvantage is that they take up a lot ofspace. This is especially annoying if one wants to talk about the DRS language and DRSs need to enter formal expressions in the metalanguage. For this reason we shall switch to a linear notation in this and in the next section, but in section 4, where we shall have occasion to discuss applications, we shall switch back to the easily readable DRS format again.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
X
bathroom x in thishous� x
364 Neganon and DLSjuncnon m DIScourse Represencanon Theory
The basic ingredients of the DRS language are familiar from ordinary predicate logic; we have terms (constants or variables, the vanables are also called discourse referents) and at least unary and binary predicate symbols. We use t to range over terms, P to range over unary predicate symbols, and R to range over bmary ones. With the help of these ingredients we build up conditions (cp) and DRSs (K) by the followtng rules, which are presented m Backus Naur Form. cp ::- Pt I t ,Rt2 1 t , - t2 1 -.K I K , v K2 1 K, K �- [x1 x,. i cp1 , ., cp,.] I K 1 ; K2 •
.
•
•
=> K2
•
•
•
•
•
•
.
•
•
•
•
•
(24) [x l x -john, -.[ 1 -.[y I umbrella y, x broughtyJJJ (2S) [ I [xl bathroom X, in-this-house xJ => [r lfunny-place y. r - x] ] Next, for each condition cp occurring in some DRS K' we are interested in the set ACC( cp) of discourse referents that are accessible from cp (m K') and it will be expedient to define ACC(K) for each K that is a subDRS of K' as well. Our definition will be (intensionally) different from the standard one, but (extensionally) equivalent. Setting ACC(K') - 0, we define ADR(K), the set of active discourse referents of any DRS K, by lernng ADR([x1 x,. I cp1 , , cp,.J) {x1, , x,.) and ADR(K1 ; K2) - ADR(K1) v ADR(K2). The discourse referents accessible from any subDRS or condition in K' can be computed in a top-down way by the following rules. 1 3 •
•
•
•
•
•
•
•
•
(i) If ACC(--.K) - X then ACC(K) - X (u) If ACC(K1 V K.z) - X then ACC(K1) - X and ACC(K2) - X (iit) If ACC(K1 => K2) - X then ACC(K1) - X and ACC(K2) - Xu ADR(K1) (1v) If ACC([x1 x" I cp1, ., cp.., J) - X then ACC( cp1) - Xu {x1, ., x,. ) ( 1 � i � m) (v) If ACC(K1 ; K2) - X then ACC(K1) - X and ACC(K2) - X u ADR(K1) •
•
•
•
•
•
•
In order to illustrate the procedure we compute the discourse referents accessible from r - X in (2s): since ACC((2S)) - 0 by definition. we find with rule (iv) that the set of referents accessible from (2s)'s only condition is 0; rule (iii) tells us that, since x is the only active discourse referent of this condition's antecedent, ACC([y I funny-piau y, y - x]) - 0 v {x} - {x} and a second application of rule (iv) shows that ACC(y - x) - (x, y). If x occurs in some atomic condition (i.e. condition of the form Pt or t1Rt2 or t1 - t.z) of K from
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In the second clause it is to be understood that n or m may be equal to zero. The set of discourse referents {x1, , xn) 1s called the universe of [x1 x,. I cp1, ., cp..,] and the conduions cp1 , , cp,. are the conditions of this DRS. As an example of a formula in this revised DRS language a linear alternative for (6) is given in (24) and an alternative for ( 1 9) is given in (25).
Enuel Krahmer and Re1nhard Muskens 365
•
•
•
•
•
•
DEFINmON (DRT semantics)
- {f l l t � f is defined & ll tlf e I(P)} �t IRtzl - {fl �t1�f and lt�]f are defined & (lt 1� f, ltzll.f) e I(R)} - {fl lt 1 � / and �tJf are defined & ll t 1 ll f - � t � .l) � tl - t� - {fl �g(J, g) E V<1J} rKI I\K1 V K� - lf1 3g((f, g) E I!K t � V (J, g) E ��} - {fl 'Vg((f, g) E ���� - 3h(g, h) E iJ<.zll} �� => Kzl] x l x . 9?t• · · ·• 9?mU - {(f. g) 1./lx 1 , , x.}g & g E l cp1l n . . . n � cp'"� H t··· K K - {(J, g) l 3h((J, h) e IK t ll & ( h , g) e V<"�D} I! t ; zll
IPt�
•
•
•
A proper DRS K is true iff the empty map 0 is an element of the domain ofil
K2) is equivalent with K1 => K2, and hence that Roberts' (1 8), discussed in the previous section, is equivalent with the simpler ( 1 9). The reader may also note that K V [x 1 x. l cp1, ., cp,.) and K V (x1 x. I -.K, cp1, ., cp,) are equivalent, provided that x 1, ., x. do not occur in K. This means that the revised consrrucnon rule whtch led to the construction of ( 1 7) in the previous section would give an output that is not semantically different from the output we get from the standard DRT construction rules (for example, ( 1 7) is in fact equivalent with ( 1 3)). Similarly, since [ I -.K1) => K2 is equivalent with K1 V K2, adopting the rule that led to the construction of(22) and (23) would have no semantic effects either ((2 3) is also equivalent with ( 1 3)). There is a semantic difference between •
•
•
•
•
•
•
•
•
•
•
•
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
which x is not accessible, we say that x is free in K. If K does not contain any free discourse referents K is called a proper DRS. Our definition of the semantics of the DRT language may at first blush seem dtfferent from the one given m Kamp ( 1 9 8 1 ), or Kamp & Reyle ( 1 993), although in fact (modulo our addition of the sequencing operator, which will 1 have relational composition as its semantics) 6 it will be equivalent. For reasons of conciseness and easy generalizanon we shall give a definition inspired by the one given in Groenendijk & Stokhof ( 1 99 1 )Y Let M - (D, I) be a first-order model with domain D and interpretation function I and let J, g , and h range over finite assignments, i.e. finite partial maps from the set of discourse referents into D. Define lt � f to be j{x) if t is the discourse referent x and x e dom(j), and define �t�f to be I(c) if t is the constant c. If.f{x) is undefined llx !f will also remain undefined. WnteJlx 1, , x.}g ifJ� g and dom(g) - dom(j) u {x 1, , x"}. We define the extension � cp� of a condition cp to be a set of assignments and the extension iK1J of a DRS K to be a binary relation between 1 assignments by means of the following inductioiL 8
366 Negation and Dtsjuncoon in Discourse Representation Theory
(6) and (7) though. and since we want a version of DRT in which double negations can be cancelled we shall define a new negation in the next section.
3
D O UBLE N E GAT I O N D RT
({J
K
:
t1 Rt2 1 tl - t2 1 K l v K2 1 Kl � K2 [x 1 . . . xR I cp1 , , cp,.] I K1 ; K2 1 -K
-
Pt I
:-
•
.
.
We interpret this language by borrowing a technique from partial logic. Conditions will as before have an txtension which consists of a set of partial assignments, but with each DRS K two relations between assignments will be associated, its extension IKf and its anti-extension vq-. In the definition below we give the semantics ofDouble Negation DRT. The idea is that all conditions, except those of the form K 1 V K2, have a semantics that does not differ from the one given in the previous set-up and that the semantics of K 1 V K2 is no different from that of -K 1 � K2• 1 9 The extension of a non-negated DRS K is as before, but its anti-extension is defined to be equal to the extension of [ 1 --.K] in the previous set-up. Negation is now indeed a flip-flop operator and switches berween extensions and anti-extensiOns. DEFINITION (Double Negation DRT semantics) IPtl
-
lt 1 RtJ
-
��I - �� I!K 1 V K�
� KJ �� Hx 1 • • x. I CfJ1 • • • ., cp,. ] f Hx 1 • • x. I f/1, . . ., cp. ] IJ �� ; KJ+ K �� ; � •
•
lfl �� � � is defined & ltlf E l(P)) lfl lt 1 � f and ltJf are defined & �t 1 1f, lt�f) e I(R)) {fl � � � � � and �tJf are defined & l t 1l f - ��zl � lfl Vg((f, g) E I!K 1 1- - 3h(g, h) E �J+)) lfl vg((f. g> e vc�r - 3h(g, h> e �J+)J ! l --.3g(f{x I• • • ., x.lg & g e �({J� � �cp,DJ !(f. g> l 3h((f. h> e ��r & < h . g> e IK¥)J !(f.f> l --.3g3h((f, g) E ��r & (g, h) E niC_zf)j ·
r�
r�
r� . . . r�
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The basic problem with negation in standard DRT is that it is not a flip-flop operation like its cousin in ordinary logic. Even the very syntax of negation discourages flip-flop behaviour: if K is a DRS, -.K is a condition and there is no comparable operator which takes us from conditions to DRSs again. In our variant ofDRT-Double Negation DRT-we remedy this and let the negation -K of a DRS K itselfbe a DRS. This is our only addition and we have removed the original negation, so that the syntax of Double Negation DRT looks as follows.
Enuel Krahmer and Retnhard Muskens 367
two conditions are said to be equivalent iff their extensions coincide. DRSs K1 and K2 are equivalmt iff JK1r - IJKzlr and IJK1 t - IK�· It is immediate that - -K is equivalent with K, whence the name 'Double Nega tion DRT. In the definition of accessibility a little care must be taken for the following reasotL Clearly, in [xl man x] ; [y I umbrella y, x owns y] the first occurrence of x should be accessible to the condition x oums y. (Note that DRS is equivalent to [x y I man x, umbrdlay, xownsy].) But in -[xl man x] ; [y I umbrella y, x oumsy] this should not be the case, while in - -[x I man x] ; [y I umbrella y, x owns y] the accesstbility should be restored agaitL To get this right we not only define the set of active discourse referents of a given DRS this rime, we also define its set of passive discourse refermts. The following clauses do the job. As before,
PDR( [x 1 x. I IP1, ., 91,.]) - 0 ADR(K1 ; K2) - ADR(K1) v ADR(K2) PDR(K1 ; K2) - 0 ADR(-K) - PDR(K) PDR(-K) - ADR(K) •
•
•
•
•
Accessibility in K can now be defined in a straightforward way by setting ACC(K) - 0 and computing the accessible discourse referents of subDRSs and subconditions with the help of the following rules. (i) If ACC(K1 V K� - X then ACC(K1) - X and ACC(K� - X v PDR(K1) (ii) If ACC(K1 => K2) - X then ACC(K1) - X and ACC(K0 - X v ADR(K1) (iii) If ACC([x1 x. I 9J1, , 91,.] ) - X then ACC(911) - X v {x 1 , , x. } (1 � i � m) (iv) If ACC(K1 ; K0 - X then ACC(K1) - X and ACC(K0 - X v ADR(K1 (v) If ACC(-K) - X then ACC(K) - X •
•
•
•
.
•
•
•
•
Again, an occurrence ofx in an atomic condition IP in K is said to beJru in K iff x � ACC( IP). An occurrence of x in a condition tp isJr« in tp iff it is free in [ I tp ]. A DRS K is proper iff no occurrence of a discourse referent in K is free in K. A proper DRS is trut if 0 is an element of the domain of its extensionJalst if 0 is an element of the domain of its anti-extensiotL The following lemma is of practical irnponance.
MERGING LEMMA x. I IPJ, . . ., IP.] ; [yt · · y,. I t/Jt, · · ., tiJJ] [x t is equivalent with •
•
•
·
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
ADR([xt · · · x., I (/Jt , · · ., 91.,.]) - {x t , · · ., x.}
Negaoon and DISJ Uncnon in Discourse Represemanon Theory
368
., (/J,., tJit, ., 1/j] , [x t XsYt Ylr I f/Jt, provided no referent m y1, , y, is free in any o f cp1 , · ·
·
·
•
·
·
·
·
•
•
4
·
•
•
•
•
, cp,..
APPLICATIONS
CONSTRUCTION ALGORITHM Input: For i =
a dtscourse Sl ··· · .Sn the empty DRS K0 =
1 to n do:
.�
D
(t)
Let K1*
(ii)
Keep on applytng construction rules to each rcductble condttton of K1* unttl a DRS K1 ts obta.tned that only contains t rreductble condtttons
=
K1_1
20 Go to
(11)
Applying one step of trus algorithm to ( s ) , reprinted as {26) below, gives (27) as an output. {26) It is not true that John didn't bring an umbrella. It was purple and 1t stood in the hallway. {27)
D·
It ts not true that John dtdn 't bnng an umbrella
(27) we encounter a negation and a proper name. For these we have construction rules that are slightly different from their standard variants. They are formulated as follows.
In
NEGATION RULE. Upon encountering any form of linguistic negation, prefix the DRS that the condition containing the negation belongs to with - and remove the linguistic negation.
PROPER NAME RULE. Upon encountering a proper name
a,
replace
a
Wlth a new discourse referent x and prefix the entire DRS under
construction with [x l a - x] .
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since we want to show in this section how our new versiOn of DRT deals with the kind of sentences that we have encountered in the first section, we must make clear how its construction algonthm works. Fortunately we can borrow many rules from the standard approach. The bas1c set-up is as follows (compare the following rule for the global structure of DRS construction With that of Kamp & Reyle 1 993: 86).
Enuel
Krahmer and Reinhard Muskens 369
This exhausts our changes to the construction algorithm. An application of the negation rule to (27) gives (28) and a subsequent application of the proper name rule (29). In the latter we may (if we wish) merge [x I john - x] and the empty DRS [ I ] to [x l john - x], accord1ng to the merging lemma of the previous section. This gives (3o) and with a second application of the negation rule we obtain (3 1). (28)
(3 1)
I; I;
ohn -
x
ohn -
x
1;0;I· -
m_ d_ n_ n_u_ b_ re_ l l_ b_ a _, n· n t d_ _ ._x 't _ _g_a_
x didn 't bn ng
� �·
--
an umbrella
x brought an umbrella
At this cruaal point we may cancel the double negation, with (32) as a result, and an application of the standard rule for indefinites brings us to (3 3). Now the Merging Lemma can be applied, so that we get {34) (32)
I;
ohn -
(3 3)
x
I·
I; .J .. .
(34)
x brought an umbrella
y umbrella y x brought y
X
y john - x umbrella y x brought y
Since there are no more reduable conditions now, the construcnon algorithm prescribes attaching a new DRS with the second sentence of our discourse as its only condition. The result is given in (3 s). Clearly, since y is accessible from this new condition, both occurrences of it can be resolved as y. (3 s)
X
y john - x umbrella y x brought y
' It was purple and 1t stood 10 the hallway
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
{3o)
D . - I John d t d n 't bring an umbrella
370 Neganon and Dtsjuncnon in Discourse Representation Theory This shows that our version of DRT treats double negations as holes for anaphora.21 That it treats single negations as plugs can be illustrated from the
treatment of (36)P Since the only difference between the first sentence of (26) and that of (36) is that the latter lacks a negation, it is obvious that the construction algorithm outputs (37) instead of(3 3) for this sentence. This DRS can no further be reduced and if the second sentence of (36) is added, as in (38), we find that the two occurrences of it cannot be resolved as y since the latter referent is not accessible.
[d (b]
(36) John didn't bring an umbrella. *It was purple and it stood m the hallway. (37)
y ; - umbrella y john - x x brought y
it
was
purple and I t
st ood in the hall�ay
This brings us to the treatment of 'bathroom' sentences. Supposing that the construction algorithm assigns (4o) to ( 1 2) (here reprinted as (39)), we see that these sentences no longer form a problem. Since x IS an active discourse referent of [x I bathroom x, in this housex), it is a passive discourse referent of its negation. This means that it will be accessible from the second disjunct, so that we can resolve it as x. The result is shown in (41). Note that this last DRS is equivalent to (19) (reprinted as (42)), so that (39) is predicted to be equivalent with (43). (39) Either there's no bathroom in this house, or it's in a funny place (40)
-
-
X
v 11 's In bathroom x in this house x
a funny place
X
y v funny place y bathroom x in this house x y-x
X
bathroom x in this house x
�
y funny place y y-x
(43) If there's a bathroom in this house it's in a funny place
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(38)
y ; - umbrella y john - x x brought y
Enuel Krahrner and Reinhard Musl-ens
371
Our semantics for disjunction does not treat both disjuncts on a par. It predicts that anaphoric links with an antecedent in the second disjunct and a dependent element in the first disjunct are out, but that links with an antecedent in the first disj unct and an anaphoric pronoun in the second are acceptable in certain
circumstances. In particular, sentences like (44) are predicted to be unacceptable if it is to be dependent on a bathroom only and not on previous or accom modated context (44) Either it's on the first floor or there is no bathroom in this house This is correct in our opinion. as (44) is very strange if no previous mention of a bathroom has occurred. Note, however, that a symmemc treatment of both tion.23
�I v K� - lfl Vg((f, g) E V<"��- 3h(g, h) E IKJ+) & vg(<J. g> E liKzt 3h(g. h> E IK�r)J ....
....
A final word on representations. In this paper we have used a representation language that extends the familiar DRT language and for some discourses the DRS that we obtain after applying the construction algorithm will not be equivalent to a DRS of the old language. Thus while the DRS for the first sentence of(26) turned out to be pan of the old language, the DRS in (37) could not be so reduced. Theoretically there is no problem here, but since discourses with an alternation of negated and non-negated sentences tend to get rather long DRSs and also for the sake of comparison with the standard DRT set-up, we may nevertheless want to use the old forms. To this end we may reintroduce the 'old' DRT negation into the new language, simply by defining -.K to be an abbreviation of semantics.
K
�
- [ I ] ,24
and by noting that this leads to the following
lrKD - lfl -.3t.J. g> E IK1rJ We now have the following useful lemma which has a simple proof SINGLE NEGATION LEMMA
K � - [x I · . . x. I IPt • . . ., cp,. ] - [xl . · . x. l cpl• · ., cp.,.] � K K ; - [x l . . . x. l cpl, . . ., cp.,] - [xl . · . x. I IPt · · · ., cp.,.] ; K ·
is equivalent with is equivalentwith is equivalent with is equivalent with
K� [ 1 -. [x 1 x. I IPt . . ., cp,. ]] [ l -.[x1 x. I IPt . . . ., cp,. ]] � K K ; [ 1 -.[x 1 x. l cp1, . . ., cp,. ]] [ 1-. [x 1 x. l cp1 , . . ., cp,. ]] ; K •
•
•
•
.
•
•
•
.
•
•
•
•
Since we can cancel double negations, and since we can trade disj unctions for
implications via the equivalence between K1 V K2 and -K1 � K2, and in virtue of the properties of the construction algorithm, we can now reduce our new DRSs to the old ones. The procedure is illustrated for (37) below. To this DRS
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
disjuncts can be obtained using the following alternative semantics for disjunc
372 Neganon and Disjunction in Dtscoursc Represcntanon Theory
the Single Negation Lemma applies. and we get (45). A last applicanon of the Merging Lemma results in {46), the form that we are used to associating with the first sentence of {36).
(•s) I : I · ohn - x
{46)
y
umbrdla y x brought y L.....=:=====....l �
X
john - x y
umbr�lla y x brought y
Acknowledgements TillS paper was presented to the SALT IV conference at the Uruversiry of Rochester m May
1994- We would like to thank Nick Asher, DaVId Beaver, Greg Carlson, Robm Cooper, Paul Dekker, Klaus von Heusmger, Makoto Kanazawa, Wilham Ladusaw, Luuk Lagerwerf, Mieke Rats, Cmge Roberts, Leonoor Oversteegen, Stanley Peters, Carel van Wijk and rwo anonymous referees for comments and crinctSms.
EMIEL K.RAHMER and REINHARD
MUSKENS Trlburg Univ=ity li!SlrtuUfor l.Anguaf! T«hnology and Artrfu:ral Inullrf!nu Warandelaan 1., Trlburg
Received: 2 1.07.94 Revised vers10n recetved: I o o 3·9S
P.O. &x 9 0 1 53 5000 LE
Tilburg
Th NtthN'Iamb
N O TES 1 Examples (1}-{4) are taken from Kart tunen's ongmal paper (Kamunen 1976� 2 E.g. Cluerchia (1 992� Groenendyk & Stokhof(1991� Kamp & Reyle (1 993). 3 We assume that negative verbs such as fori and Jorgtt are analysed WJth the help of neganons (= e.g. Kamunen & Peters 1 979 for such an analysis� 4 Double negations m standard EngliSh are one of our rwo main concerns 10 this paper. Neganve vrbs allow for easy
consttucnon of natural examples of double neganons, bur some also mrro duce problems that are orthogonal to our present interests (forgtt and dmy, for example, are verbs of proposmonal arnrude as well, of course) Therefore, m the rest of tlus paper we shall stick to srraighrforward examples of double negations, as the one m (S� See also note 21. S Roberts (1 989) amibutes tlus sentence to
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
�
Emiel Knhmer and Remhard Muskens Barbara Partee. In Evans
6 7
(1 977) we find EitkrJohn d� not own a donkry, or h� k«ps it wry qur�t.
For a dLSCUSSi on of the issue of acces sibduy m diSjuncnons see secnon 2.3. 1 (pp 1 8 5 - 1 90) of Kamp & Reyle (1993� Ro�ns (1989) gtves the followmg example (annbuted to Berman), however:
Eith" th"�s a bathroom on tkfirstfloor, or rt's on tk s�ondfloor We believe that m this eXAmple the
the speaker IS committed to the eXIStence of a bathroom. Note that the indefimte allows for subsequent anaphonc refer
ence: we can connnue w1th I k«pJorgnting
aactly
whm
it
is,
but it's
easy
to find,
another SJgn that the indefirute has wtde scope here.
8 In fact, m a sense that wtll be made preCISe in section 2, ( 1 7) and ( 1 3) are eqmvalem m me2nmg and the new rule that m construcnng the DRS for a dJSjuncnon we should add the neganon of the DRS for the first disjunct as a condmon to the DRS for the second does not gtve us any output that IS semantically d1fferent from the output of the ongmal construcnon rule.
9
Loss ofexplanatory power would also be a
consequence of a rule to this effect. Although the entatlment John didn'tJail to find an a�!Johnjound an a� would come out vahd under such a rule, thiS would ortly be a consequence of what we feel would be sleight of hand: the represenranon of the premise would be
rurned mto the representanon of the conclusion by mere snpulanon. I O An anonymous referee infonru us that he would call (1 2) netther true nor false in these Circumstances. 11us means that although his mruinons do not sq�Ure With Kamp & Reyle's solunon of render
mg ( 1 2) as (16), they neither square wuh Robens' ( 1 989) proposal (d1 scussed below) to treat (12) as ( 1 8), or wtth our
rendering of ( 1 2) as the eqmvalent (19� Perhaps a uruqueness presupposinon IS mvolved here. Robens uses a modal box mstead of an
1mphcation. to be quite prease, bur this IS for our present purposes.
unnu tenal
12
In secnon 4 below we shall gtve a slightly d1fferent analySIS of diSjuncnons. We shall not change the DRS construcnon rule for d1sjuncnons, but the sernanncs for the symbol V wtll be altered m such a way that A or B wtll be semanncally eqmvalent to if not A tkn B. In an earlier
versiOn of rhiS paper our analys1s of 'bathroo m ' sentences was based on Kamp
& Reyle's analysis plus our solunon to the double neganon problem. We WISh to thank Paul Dekker for InsiSnllB that the eqmvalence between A or B and if not A tkn B should be retamed. 1 3 Vanous people, mcludmg Werner Saurer and one anonymous referee, have sug gested that the relanon between (23) and (19), (17) and (16), and (6) and (7) should be one of inf"mu ln the1r proposals the DRT construcnon algonrhm JS ennched With an mferenCillS mechaniSm, so that drawtng conclusiOns Js an admissible processi ng rule. (7), for example, may he constructed from (6) m such approaches, smce It Will follow from (6) in some smrable DRS mferenang system (the one m Saurer 1993, for example) The pnnc lple challenge for theories along these hnes, a challenge also noted by Saurer, IS that of restricnng overgeneration. While It IS not d1ffirult to let (1 9) follow from
(23) and (7) from (6), 1t IS dJfll rult to do so
and not have many unwanted inferences m the barga1n. Consider Partee's 'marble' eXAmples: (a) I dropped ten marbles and found ortly rune of them. *It's probably under the sofa. (b) I dropped ten marbles and found all of them except one. It's probably under the sofa. On a narural account of inference the first sentence of (b) Will follow from the first
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
mdefirute noun phrase a bathroo m gets a wtde scope speafic readmg. Inrumvely
I r
373
374 Neganon and Disjuncnon
Discourse Repr�ntation Theory
a theory which allows
plugs for presupposition projecnon may
mference as an acceptable processm g rule
actually have more m common than J USt
Will not be able to explam the difference
theu mmes. Analogtes between amphora
sentence of (a).
m
I4
m
So
and presupposmons have been noted by
acceptability between (a) and (b).
Tius is the same as the conjunction m GroenendiJk & Stokhof( I 99 I �
I S The following analogy seems relevanL Karrtunen ( I 974) g�ves a set of rul� wluch allow us to compute when a
vanow authors (see espeaally van der Sandt I 992� I6 llus
1s
the
standard
I 7 See defirunon 26 m GroenendiJk
context (set of sentences) � sansfies the
Stokhof's ( I 99 I ).
presuppositions of a sentence So. The 1dea
I 8 The defirunon that
IS
to associ ate a local conto:t C With each of
semantics
of
sequenang-see Pratt ( I 976�
IS
close to GroenendiJk
&
g�ven here 1s very
& Stokhof's ( I 99 1 )
defirunon 2 6 , b u t we follow the standard DRT set-up by
LqSo) - � and proceeding to compute
(1)
the local contexts of proper subclauses by means of the followtng rules
(1)
(u)
IfLqnot S) - C then LqS) - C
(i1) IfLqs or S') - C then LqS) - C
(m)
argued
IfLq,f S then S'} - C then LqS)
and LqS') - C u (S) case
-C
m
IS
[ I grty x].
the
calculus
for
q
Fernando
[x J grty x] the
uruverse of the
[x J donkty x] Q
In the GroenendiJk
set-up it Will be equivalent to
& Srokhof
[x I donkty x] q [r l grryy]. Of course, condmons hke [x l donkty x] -> [x J grty x] Will not be generated by the standard DRT construe
&
non algorithm. I9 Note
that
the
folloWing
alternative
defirunon for the value of a d1sjuncnon
that the accesstbihty calculus for DRT
and
[x J donkty x] x m the
condition IS equivalent to
projection, for which the reader may
hterarure. 11te pomt that IS relevant here
by
second box has no effect at all and that the
well-known theory of presupposmon
Peters ( I 979), and the vast subsequent
m
occurrence of
Karmmen's
consult Karrtunen ( I 974), Karrtunen
extensively
of Jr 4g has the consequence that for
the local context of each
ThiS results
for
example
subclause of So entails all presupposmons that are mggered at the level of that
subclause.
rwsign
( I 992� As for (u), note that our defirunon
� now sansfies the presuppositions of So JUSt in
d!Sallowing what are called
mrots.
The use of firute assignments has been
and LqS') - C u (not S)
C and LqS') - C u {S) (1v) IfLqs and S') - C then LqS)
finiu (and hence p:�mal) assign
usm g
ments Instead of total assignments,
would embody a vernon of the Kamp
presuppostnon
&
Reyle theory of'bathroom' sentences.
projecnon m Karnunen's theory share
Important formal characrerisncs. Also
compare the accessibility calculus for 'Double Negation DRT which will be given m section 3 below. At several pomts in the paper we have and
hole
used the terms plug
Note also that we are not committed to an asymrnemc rreartnent of disj unction.
from Karnunen ( I 973) as an
informal way to refer
to
operators which
do and operators which do not block the
20
See the diSCUSSion at the end of this paper.
A more precise account would have the
posstbihty of amphonc reference. The
synttutrc analysiS of S1 as the contents of the
present observation suggests that our
new box. Compare Kamp & Reyle ( I 993). 2 I This means that, given our acceptance of
plugs for access ib1hty and Karrtunen's
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the subdauses of So. Local contexts can be computed in a top-down way by semng
Emiel Kr:ilimer and Reinhard Muskens 375 the Karnunen & Peters' anal}'SlS offowt as not mnnn Mt-, we have reduced an explananon of the acceptability of (a) to an explananon of the acceptabthty of (b� (a) John d1dn't forget to bring an umbrella. It's in the hallway (b) John remembered to bring an umbrella. It's m the hallway. A5
(a) It 1s not true that John is happy 1f he owns a donkey •1r ts grey. (b) It IS not true that John doesn't own a donkey. It ts grey If the imphcanon
m
the first sentence of
(a) were of the form -(K1 ; -K2), however,
the first sentence Itself would be equi valent to K1 ; -K2 and the discourse referent for a donluy would be accessible.
RE FERENCES Chierclua, G. (1 992). 'Anaphora and dynamic l<>g�c', LinguiStics and Philosophy, IS, 1 1 1 83. lkrg. M van den (1 993). 'Full dynamic predicate l<>g�c', 10 K.. BimbO & A. Mate (eds). Promdings oJth( Fourth Symposrum on Logic and lAngua�. EOtves University, Budapest Dekker, P. (I 993). 'Transsentential medica nons: ups and downs in dynamic seman ncs', Ph.D. dissertanon, ILLC dtSsertaoon senes, no. I, Amsterdam.
Evans, G. (1977), 'Pronouns, quantifiers, and relanve clauses ( I & 2)', Canadianjournal of Phrlosophy, 7, 467-5 36, 777-<)7. Fenundo, T. (1 992). 'Transinon systems and dynamic semantics', Logics in AI, LNC:S 633. Spnnger-Verlag, Berlin. GroenendiJk. J. & Stokhof, M (1 990). 'Dynamic Montague Grammar', m L Kalman & L Palos (eds), Papm from t� &cond Symposium on LogiC and l..Angua�. Akademiat Kiado, Budapest. 3-48. Gromendijk, ]. & Stokho( M (1991).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
one anonymous referee points out, there lS snll a problem wtth (b). as Jt needs to be explamed how a dtSCourse entity that i5 utttoduced wtclun the comple ment of :m attitude verb can become accessible for subsequent anaphonc reference. However, thts lS a general problem for DRT, for standard DRT as well as for our version. While we do see ways for deahng With the quesnon, we feel that the maner falls outside the scope of thts paper. See also note 422 In thts respect our neganon ts different from the dynamiC neganons considered 10 GroenendiJ k & Stokhof (1990), van den lkrg ( 1 993). and Dekker (1993) Wlule these negations correctly predict that a double neganon does not block anaphora, they also wrongly predict that a s1ngle neganon does not 23 Note the analogy berween the present dlSCUSSion and the old dJSCUSSlOn whether dtsjuncnon should be symmemc with
respect to presupposition projecnon (see �. Karnunen 1973� 24 The negated box - [ I J denotes the empty relanon and may thus stand proxy for the falsum .l in our theory. �K 1s thus defined as K q J. , as it lS m some vemons of the standard set-up of DRT. Note that in the set-up of section 2 an 1mphcanon K1 � K2 may also be defined as --.{K1 ; [ I � K2J) (see GroenendiJk & Stokhof 1991). so that negation i5 not only definable from unphcanon (and falsum), but unphcanon IS also definable from neganon (and sequena� This mterdefinab1hty 1s lost in our set-up, as K1 � K2 ts not eqUivalent to -(K1 ; -K2). For good reasons, as dtSCourse referents that were created 10 the antecedent of an 1mphcanon Will not become accessible 1f that imphcanon is negated, wlule discourse referents created under a neganon Will become accessible 1f delivered by a second neganon.
376
Neganon and Dlsjuncnon m Discourse Representation Theory
'Dynamic predtcate logtc',
Phrlosophy,
14. 39-IOO.
Lrngurstro and
Uruverstry
of
Massachusetts,
Amherst Heim, I. ( 1 9 8 3), 'On rhe proJecnon problem for presupposmons', in M Barlow rt a/. (eds), Promdrngs of thr &cond Wl'St Coast
Ccnfrrrnu on Formal Lrnguistia, Sranford Universtry, Sranford, I 1 4-25. K.admon, N. (1 987). 'On uruque and non
rt al (eds), Formal Mrthods rn thr Study of Languagr 1 , Machemansch Cenrrum,
K.amp, H. & Reyle, U. ( 1 993). From Discourst to Logrc, Kluwer, Dordrecht. l{;jrrrunen, L ( 1 973). 'Presuppostnons of compound sentences', Lrnguistic Inqu�ry, 4, 167-93· K.artrunen, L ( 1 974), 'Presuppostnon and
l{;jrrrunen, L ( I 976). 'Dlscourse referents', m
J. McCawley (ed.), SyntJU and Smrantia 7, No� from thr Lingurstrc Undrrground, Academtc Press, New York, 363-85. K.arttunen, L. & Peters, S. (I 979). 'Con vennonal rmphcarure', m C Oh & D. Dmneen (eels.). SyntJU and &mantia, Vol 1 1 Pmupposrtron, Acaderruc Press, New
York, I -56. Prarc, V. ( 1 976). 'Semannc constderanons on Floyd-Hoare logtc', m Prourdings of thr
1 7th IEEE Symposium on Foundations of OJmputrrScrrocr, 1 09-2 1 Roberts, C. (1 989). 'Modal subordmanon and pronorrunal anaphora m dtscourse', Lrn gurstm and Phrlosophy. u, 68 3-72 1
Sandt, R. van der ( 1 992), 'Presupposinon proJecnon as
anaphora resolunon',journal ofSemantia, 9, 3 3 3-77-
Saurer, W. ( 1 993), 'A narural deduction system for dtscourse representanon theory', journal of Phrlosophrcal Logic, :u, 249-302
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
uruque reference and asymmemc quann ficanon', Ph.D. disserunon, Uruverstry of Massachusetts, Amherst l{;jmp, H. ( 1 9 8 1 ), 'A rheory of truth and sernannc representanon', in J. Groenendtjk
Amsterdam, 277-322
Thro('(tical Lrnguistia, 1 ,
I 8 H )4-
Henn, I (1 982), 'The semanncs of defirure and mdefimte noun phrases', Ph.D. d=r
canon,
linguisnc context',
C iliford Uruvernry Press 1995
0- and D-Predicates: A Semantic Approach to the Unaccusative-Unergative Distinction I N GRID KAUFMANN
Univa-sity ofD�Idoif
Abstract
encode
o pnonal
d efinmg
propemes, and 0-predtc:ates, whtch
propemes. Wlule the dtsnncnon between the propemes
IS
conceptually
monv:ated, the two correspondmg predtate classes :are defined wuh reference to a logtcally
based nonon of'(predtate) field'. It IS argued that on the basts of thts dtsnncoon the dtfferences m the grammanal behavtour of un:accusaave :and unerganve verbs m German can be denved
:and do not requtre any form of synracnc represenranon bec:ause the propemes that are usually
uken to be :an mdtanon of the class membershtp of mrrans1ve verbs follow from the stants of
the propernes encoded by the two types of sem:annc predtcates
1
I NTRO D U C T I O N
One of the main topics that crop up in the diScussion of lextcal representatioru is the extent to which the syntactic and morphological behaviour of lextcal iterru is determined by their semantic representation. With respect to verbs, this question involves discussioru about the syntactic realization of arguments and about the spectrum of argument changing processes a verb may undergo. A theory in which these grammatical properties are derived from semantics has the advantage that less (probably redundant) informacion must be exphcicly encoded. The fact that argument structure changing processes such as passivizacion or resultacive formanon are often subject to subtle semantic corutramts provides further motivation for a semantically based approach. However, any theory in which lexical semantics plays a determirung role must specify primitive semantic elements that determine grammatical behaviour. In some approaches, thematic roles have been assumed to be the primitives which determine both the syntactic realization of the arguments, as well as whether a verb can undergo processes like passivizacion, resultacive construction, or middle formation. However, since it is difficult to find adequate characteriza tions for individual thematic roles, and since no criteria seem to exist to detenrune how many of them have to be assumed, these approaches lose much of their attractiveness as soon as subtle phenomena are taken into account.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
In dus paper :a conceptually b:ased dtsnncoon between two classes of semannc predtcates ts proposed· D-predt:ates, wluch encode object
378 0- and D-Pred1cates In an alternative approach, the relevant semantic properties can be derived by a decomposed srructure. Depending on the semantic predicates (and relations)
property change of a causer. Thus, assuming that changes of properties are encoded by means of BECOME one can conclude that BECOME has selec tional restrictions excluding predicates which encode concepts involving causation. Similarly, as Levin & Rappaport Hovav ( 1 995) themselves point out, mtraruicive verbs encoding an mternally caused process do not pamopate in the causative alternation. Again. one can conclude that the semantic predicate encoding direct causation may not embed predicates encoding internally caused processes. Thus, on the assumption that verbs are to be represented by semantic predicates, there seems to be evidence for the eXIstence of subclasses of semantic predicates that differ With respect to the concepts they encode and with respect to the semantic representations in which they may occur. As Levin & Rappaport Hovav (1 99 5) show, factors such as 'externally/ internally caused' as well as 'directional change' play a role in the diStinction of the predicates. As long as this difference is referred to only by linking rules, though, what seems to be a genuine semantic phenomenon is removed &om semantics. However, if the source of this difference could be integrated mto the semantic representation, the relevant properties could be used to formulate restrictions on semantic representanons directly, and explicit linking rules might rum out to be superfluous. In this paper an attempt is made to locate the difference between the predicates within semantic representation. I propose adiscinction between the two rypes of object properties that provides the conceptual basis for rwo classes
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
wluch predicate of them, the arguments of verbs have certain charactensncs, such as affectedness or agentiviry, which are relevant for argument linking. Levin & Rappaport Hovav ( I 995) present a recent approach to unaccusacivity mEnglish along these hnes in which an argument undergoing a directed change IS linked to direct object posinon. and an argument that denotes the immediate cause of the eventuality is linked to subject posicion. Linking rules hke these account for the unergative/unaccusacive diSnnccion as well as for the fact that it is only the direct object oftransitive verbs, and not the subject, that may undergo a change. More over, they account for the fact that in non-derived verbs only the subject can denote a causer. Argument changing processes, then, can be formulated as applymg to verbs with specific argument structures. Returrung to semannc representanons, this kind of analysis implies that there are predicates m the semantic representanons of verbs that encode concepts which involve causation or agentivity, and other predicates which encode concepts involving changes of objeets. What remains to be explained, however, 1S why one-place predicates encodmg these concepts always seem to be distinct. For example, there seem to be no (simplex) inchoacive verbs (and, therefore, presumably no semantic predicates) encoding concepts involving a
Ingnd
KAufmann
3 79
2
0-
A N D D - PRE D I CATES A N D THE I R C O N CE P T UAL B A SE
As has been mentioned in section 1 , the concepts of causation/agency and change obviously play a role in determining the differences between semantic predicates. Since predicates encode properties of their arguments, a classifica tion of predicates must follow from differences m object properties. I assume that the two types of object properties that should be distinguished are 'object defining properties' and 'optional properties'. (Please note that in the following the term 'object' refers to ontological entities and not grammatical objects, the latter being referred to with the terms 'direct object' or 'indirect object' to avoid misunderstandings.) In the next two sections, I will describe these property types and the predicates that encode them.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of semantic predicates. The assumption that the properties of the verb's argument may be derived from semantic predicates is not new. For example, van Valin (1 990) propsed a 'Role and Reference Grammar' account of split intransitivity based on Dowty's (1 979) theory of verb classification. In van Valin's theory, the differences between the intransitive verb classes also result from the predicates wluch form the semantic representation. In this paper, I will go a step further, attempnng to provide a conceptual motivation for these classes by a more thorough charactenzanon of the properties which the relevant predicates encode. The theoretical model I adopt 1S the two-level approach of meaning developed in Bierwisch (1 983) and Bierwisch & Lang (1 989), m which it is assumed that meaning is determined partly by the grammar and partly by general conceptual knowledge. The semantic form (SF) of a lexical item represents that part of meaning that is grammatically relevant, and it functions as an interface between (grammatically determined) linguistic knowledge and (conceptually determmed) encyclopaedic knowledge. The information encoded in SF can be regarded as the core information present in all the variants of a lexical item. The semantic components constituting SF are interpreted further in Conceptual Structure (CS), where meaning variants are derived wah the help of contextual knowledge. Thus, the differences between the properties that lead to the distinction of two semantic classes of predicates are determined m CS. Taking the unergative/unaccusanve diStinction as an example, I Will argue that it is the characteristics of the propernes encoded by these semantic predicates that show up in the properties of the1r arguments and that motivate restrictions on the applicability of argument changing processes.
380 0- and D-Pred1cates 2.1
Object defining properties and D-predicates
2. 1 . 1 Obj ect defining properties Object defining properties are those properties that are necessary for the conceptualization of an object Since they are object defirung, an object always has to display them in some way. As the term itself suggests, the classification of a property as object defining depends on the specific character of the object involved. Object defining properties ofconcrete objects are, for example, spacial properties Qocalization as well as shape), visual properties (such as colour or transparency), and material properties (such as substance and aggregate state).
lead to another quahty (or 'instantiation') of this property (as long as the object is not destroyed by the change). For illustration, consider the property of having colour. An object which undergoes a change in colour necessarily obtains another colour as a result. This fact can be accounted for if it is assumed that object defining properties are the dimensions that define the concept of an object, such that an object wruch displays the relevant type of property cannot 'leave' the relevant dimension bur must always be located at a specific co
ordinate (i.e. displays a specific instancianng property). 1 Figure
1
informally
Illustrates the dimensions of colour, aggregate state and shape, and relevant instantiating properties.
colour dimension: aggregate dimension: shape dimension:
blue, red, yellow, black, brown . . . gas-fluid-solid round, rectangular, . . . , OSI, os2, OSJ, . . . 'cup shaped', 'chair-shaped' . . .
Figure
1 DimensiOns and instannanng object definmg propemes
The shape dimension in Figure 1 shows that the instannating propernes of a dimension need not always correspond to lexical items. In this particular case one can imagine many possible instantiating propernes, whlle linguistic expressions exist only for a few, very simple shapes. OS 1 , OS2, OS3 refer to the
'Obj ect Schemata' ofLang (1 989), which are used as a means of representing the gestalt properties of objects. Object schemata are assumed to form part of our
conceptual knowledge of objects and may therefore be integrated into the shape dimensioiL 'Round' and 'rectangular' are properties for which there are lexicahzed expressions that encode (part of) the information of certain object schemata. The set of object defining properties which an object displays can be viewed as the space of properties in which the object is located and in which it can
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
The fact that these properties are object defining has an important consequence: any change affecting a specific object defining property may only
Ingrid Kaufmann 3 8 1 move. This space, therefore, makes u p the 'universe' o f the object. The specific instantiations of the object defining properties (for example, bemg blue, solid, and round) are the co-ordmates of this space, which define the consntution of the object at a specific pomt m time. Note that there may be differences with respect to wluch part of a dimension (i.e. which subset of instannating propernes) is relevant for the conceptual space of an object. For example, smce apples may not be blue or violet, these colours are not part of the colour dimensiOn of their conceptual umverse. Moreover, we find cases in wluch the instantiating propemes themselves are object defining, as in the case of being sohd, which is object defirung with regard to iCe. Figure 2 gives a rough sketch of the object defining space of a ball. For balls, particular instantiations of the
shape = round
blue red
material
plastic
aggregate = solid
�
yellow
r7
rg ·· · localisation
Figure 2 Concept of a ball aggregate state dimension and the shape dimension are object definmg, namely sohd and three-dimens10nal/round, respectively. Thus, the 'space' mcludes only these instantiations of the properties of the two dimensions. Since balls may change location, the whole locanon dimension is part of their conceptual space, the indices serving to differentiate between regions m which the obJeCt may be located. The intersection of all dimensions characterize the actual state of the individual objects. These co-ordmates, therefore, charactenze spee1fic individuals of a particular (natural) kmd, while the entire space characterizes the properties of the kind irsel£ Thus, the particular ball in Figure 2 iS round, solid, yellow, made of plastic, and localized in region r4 at a certain specific pomt m nme. Note that a more elaborate representation of an object defining space must therefore include temporal information so that the co-ordinates that an individual occupies in the course of events can be disnnguished by temporal indexation. A concept must also include information about which of the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
colour
382 0- and 0-Predicates
mstantiating properties that are themselves object defining, e.g. solid for ball, fluid for water; instantiating properties that cannot change for an individual but can be different for different indtviduals, e.g. the matenal of concrete objects; instantiating properties that can only change through external force, e.g. colour of concrete objects, localization of concrete, inanimate objects; - instanoating properties that might change due to internal force, e.g. localization for animate objects; - Instantiating properties that are bound to change during the life span of an mdividual, e.g. age in general, or siZe for animates. The status of the instanoating properties must be encoded on each dimension of the concedptual space. Note that the temporal mdexation of the individuals making up the universe of a concept is also necessary tfonly one instantiation of a dimension is object defining. If, for example, ice melts, it loses the property of being solid and cannot be referred to as Eis 'tee' any more, as the inacceptability of *Jliilliges Eis 'flmd 1ce' Illustrates. The resulong substance can, however, be referred to as something that was ice at an earlier stage if Eis 1s modified by the past participle of the verb that encodes the process of the relevant change: geschmolzenes Eis 'melted ice'.
2
2.1.2 D-predicates
Object defining properties and their instantiations are encoded by semantic predicates which I will call 0-predicates. What is rmportant for the organiza tion of semantic predicates is the fact that not every property is encoded by a predicate. Constder, for example, the property of having shape: while the existence of adjectives like round, square, and rectangular indicates the existence of semantic predicates encoding these instantiating properties, there is no reason to assume that, for example, predicates exist that encode the shapes of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
dimensions involved an individual can move in. For instance, although the ball concept allows a choice from among different sorts of material, such that a number of instantiations are part of the material dtmension, every individual ball must consist of a specific kind of material, and is therefore restricted to one instantiation of the dimension. While as far as the material of a ball 1s concerned, no change takes place, things are different with respect to the colour dimension. The colour of balls in general is not fixed, and therefore the whole colour field makes up one dimension of the space. Since change of colour can be brought about by specific types of external manipulation, an individual ball need not necessarily keep its colour throughout its whole 'life span'. In order to cope wtth the different status that the instantiating properties of a conceptual space might have for the individual objects, five cases must be distinguished:
Ingrid Kaufmann 383 desk lamps. Thus, there i s n o one-to-one relanon between propemes and predicates. I assume that predicates encoding instannations of object defimng properties are organized in predicate fields, which are the linguistic correlatives of the dimensions of object defirung propemes. In order to charactenze the predicates, I draw upon considerations concerning the logical structure of lexical fields presented in Beeh. Brosch & Schulz I 994), who suggest that logical dependence and logical independence characterize the relations between leXIcal items. If expressiOns depend on each
(
other, they belong to the same field, but if they are independent, they do not. Independence, therefore, can be used as a criterion to distingmsh fields, while dependence is a critenon that determines the internal structure of a field. Beeh
(
( I ) A formula A includes a formula B iff every (vanable) evaluation that satisfies A also satisfies B. In symbols: (i) A => B (2) A formula A excludes a formula B iff (ti) A => -.B ( 3) A formula B is dependent on a formula A iff (i) A => B or (ii) A => -.B or (iii) -.A => B (4) A formula B is independent on a formula A iff neither (i) A => B (ii) A => -.B nor (iii) -.A => B nor With respect to lexical fields, exclusiOn is relevant for the distinction of the members in a field. Thus, for example, the colour adjectives blue, red, green , and
black
exclude each other and are rhus distinct. InclusiOn, on the other hand, is
required to capture the relation between hyperonyms and hyponyms such as
red and bordeaux. Us1ng the relations introduced in ( 1 ) to (4), Beeh et al. discuss a
number of alternative definitions of leXIcal fields and they come to the
conclusion that logically based definitions of fields are either too narrow or too wide to capture, for example adjective fields. Simplifying the problem somewhat, thts inadequacy follows from the fact that, while the notion of a field based only on the relation of exclusion cannot capture hyponyrns and hyperonyrns, the notion of a field based on exclusion and inclusion is useless since through inclusion the field is extended by all the formulas that are always true. The sort of problems Beeh
et al.
encounter, however, arise in connection
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
et al. define dependence and independence of formulas by means of the notions
of exclusion and inclusioiL The English translation of their definitions is gtven in I )-{4).
384
0-
and D-Prediotes
with leXIcal items, and not with regard to pnmitive semantic predicates, since, to avoid redundancy, the predicates, being priminves of semannc representa tion, should exclude each other. Inclusion, thus, does not play a role at the level of the predicate field rtself (For reasons of terminological simplicity, I will talk of inclusion and exclusion of predicates when referring to the inclusiOn/exclu sion properties of the formulas formed by a predrcate and an argument.) I assume, nevertheless, that the definition in ( 5 ), which rs too narrow to capture the whole range of adjectives in a field, is sufficient for the required defininon of fields of D-predicates.
(s) A field
According to the definition in ( 5) , a predicate field consists of all the semantic predicates that encode the properties of one dimension. Since in each mterpretation L the formulas of a field must be evaluated with respect to a certam universe U, the notion of a field in ( 5 ) can be further restricted to those subsets of F that involve predicates that predicate of objects belonging to a speafic natural kind. This can be done by restricnng the uruverse U, with respect to which each formula has to be evaluated, to objects that are part of one conceptual space in each mterpretanon l From this restriction of the universe, lt follows that only predicates which encode those Instantiations of a dimension that are in fact relevant for the kmd are part of the field. Thus, while the colour field includes all colour predicates, rf the uruverse IS restncted to apples, the restricted colour field would mclude only predrcates such as RED, YELLOW, and GREEN , but not BLUE or BLACK. The dimensron itself is encoded by a predicate that rs included by all predicates of the field and, therefore, IS not itself part of the field. Thus, for example, HAVE_COLOUR(x) is included by RED(x), BLUE(x), BROWN(x), and so on. Note that predicates encodmg the dimensions always hold for all the objects of the relevant uruverse. This fact has an mteresting consequence. If one takes mclusion not to play a role within a field, but only at the level of the predicates encoding the obj ect defining properties themselves, i.e. the dimensions, then all object defining propemes relevant for a (natural) kind can be derived by inclusion. This follows from the fact that it is these properties that always hold, since, if a formula P(x) is included by any other formula because P always predicates of x, P must encode an object defining property of x. To summarize, predicates encodrng instantianons of object defining propemes exclude each other and include the predicate encoding the dimension, i.e. the object defining property itself
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
is a maxirnaP set F of formulas having the property that each evaluation sansfies at most one formula of F. (See Beeh et al. 1 99+ 1 7)
Ingnd Kaufmann 3 8 5
2. 1 . 3
Dynamic D-predicates
With respect to a certain object at a cenain time, instantiations of object defining properties can either simply be present or they can undergo some change. Correspondingly, two types of semantic 0-predicates have to be distinguished: static 0-predicates, which encode the presence of an object defining property, and dynamic 0-predicates, which encode the change from one property of a dimension to another. Static 0-predicates are usually lexicalized by adjectives, such as
rot
'red',
grofi
'big',
rund
0-predicates are lexicalized, for example, by verbs such as
'round'. Dynamic
schmelzen
'melt' or
explicitly speafied, or derived by negating the predicate that specifies the post state. For instance,
schmelzen
'melt' and
(sich) verjllissigen
'liquefy' characterize
the same post-state, 1.e. FLUID(x), but have differently speafied pre-states, namely, SOLIO(x) for
schmelzen ,
NON-FLUIO(x) (t.e. either SOLID(x) or
GASEOUS(x)) for verjlussigen. This dtfTerence between schmelzen and ver Jliissigen shows that, for a verb to detennine i ts specific pre-state, i t must encode information about the relevant structure of the field. I will represent dynamic 0-predicates which encode a change from one instantiation of a dimension to another not as semanoc pnmitives but by using the predtcate BECOME I assume a defirution ofBECOME which is sinnlar to a phase quannfier of type
I
m the sense of LObner
(I990). LOhner defines phase
quannfiers of type I as three-place predicates PQ I {<, x, P) with '<' being a 0exically fixed) scale defimng an order on the domain of P, 'x' being the argument ofP, and 'P' the predicate which is to be evaluated and which holds in the second phase but does not hold in the first phase of the scale. Since dynannc 0-predicates encode situations involVIng changes, an additional argument must be assumed for BECOME, which is to be realized as the referential argument ofverbs such as schmelzen . This argument has to refer to a situation s involving the property change, and whose event time is the minimal interval [t1 t2], such that --.P(x) holds at t1 and P(x) holds at t2• Moreover, since P applies to mdividuals, the relevant scale < consiSts of ordered temporal Instances of the individual undergoing the change, i.e. x. Since the predicate which holds for the argument in the pre-state can be leXIcally specified, it must be encoded by a funher predicate constant. Dynamic D-predicates that encode changes between two Instantiating properties of a field are therefore represented by means of the s-place predicate BECOME, which takes a scale < , a situation s, an individual x, and the predicates encoding the property of x in the pre- and the post-state as its arguments. This representation is illustrated with
schmelzen
'melt' as an example in (6a). For
notational simplicity, I abbreviate this representation as in (6b).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
wachsen 'grow'. These verbs generally specify the property that results from the change and that holds m the post-state, wlule the pre-state may either be
386
(6)
0- :and
a. b.
D-Pred1c:ates
schmdzen :
l x J. s BECOME (< , s, x, J.y SOLID(y), J.y FLUID(y)) J. x J. s BECOME (soLID <, s, FLUID(x))
Changes encoded by dynamic D-predicates need not always take place between two specified instantiating properties. The motion predicate MOVE is an example of a predicate that encodes a continuous change from one regton to another. On the assumption that 'being-located in region r1' is an instantiation of the localization field, MOVE involves an unspecified number of changes from one instantiation of the field to another. (For a definition of MOVE see Kaufmann 1 995.) Note that this is an effect of the lmguistically determined strucrure of the localization field which, for reasons of economy, does not regions. Instead, local verbs are generally underspecified with respect to the regions involved. Information about the relevant regions is proVIded by the local argument of the verb.• A different interpretation seems to be relevant for verbs such as
waclzsen 'grow' that, like MOVE, encode continuous changes, but
whose post-state is nevertheless speCified. I assume that the dynamic D-predicates encoding these verbs are similar to MOVE in encodmg continu ous changes between grades on the relevant dtmenslOIL They differ from MOVE, though. in that they specify the direction of change by means of predi cates that correspond to static expressions such as weit 'wide' and grofi 'big', as the interpretation of deadjectival verbs weiten 'Widen' or vergrOfiern 'enlarge'
demonstrate. However, since the properties encoded by static D-predicates like BIG or WIDE are themselves vague, a measuring umt is necessary to determine the post-state with respect to the pre-state. I Will not go mto any funher details of how dynamic D-predicates can be represented here, though. and turn instead to optional properties and 0-predicates.
2.2
Optional properties and 0-predicates
2.2.1 Optional properties In contrast to D-predicates, which encode object defining propemes, 0-predicates encode optional properties. These properties are opnonal because, if an individual object lacks them. no consequences follow for its specific objecthood, even though it may potennally display them. Examples of optional properties for humans are the properties encoded by verbs like arbeiten 'work', lachen 'laugh', and reden 'talk'. For a dog, the property of barkmg IS also optional since a dumb dog or a dog that deliberately never barks, for example, does not lose its doghood but remains a dog. Like instantiations of object defin ing properties, instantiations of optional properties may be arranged in dimen sions but, since they are optional, an object does not always have to display one
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
consiSt of predicates encoding the property of being localized in specific
Ingnd
Kaufmann 387
of the properties of the relevant dimension. For illustration, consider different forms of speech production lexicalized by agentive verbs such as schrriro'cry', Jlustern 'whisper', and rufen 'shout', which, as is required for a field, exclude each other. Though people can talk, whisper, cry, and so on, in order to be human beings, they do not have to perform any one of these actiVIties all the time. They may as well remam silent for a while, or laugh instead. This means that objects do not have ro remain permanently within the dimensions ofoptional propert ies bur may always 'leave' them. Although humans do not obligatorily have to display a property from the dimension of speech production, their
ability to display such properties must be
classified as object definmg.5 This means that the conceptual space of humans properties, however, human individuals need not be located on a co-ordmate of this dimens10n at every instance of their life span. Optionality of properties, therefore, is defined with respect to the individual objects of a kind, not for the kind as a whole, smce dimens10ns of optional properties are also part of the conceptual space of the kind.
2.2.2 0-predicates Smce the properties encoded by 0-predicates are optional, m the case of 'optional fields', there must be predicates which, although excluded by the predicates of a field, do not themselves belong to the field. This means that an optional field is not 'maximal' in the sense that it contains all predicates excluding each other. Take, for example, the field of speech production mentioned above. Since different verbs of speech production exist, there must also be different semantic predicates, such as WHISPER, CRY, and SAY, that represent them. These predicates, however, exclude each other and, therefore, the existence of a field of speech producnon is justified. Now, BE_SILENT and SING are examples of predicates that are also excluded by all predicates of this field bur do not belong to the field. Similarly, predicates encoding specific manners of motion such as WALK, RUN, SWIM, and FLY must exist in order to represent the verbs of motion. These predicates, however, exclude predicates encoding positions such as LIE and SIT. Since LIE and SIT exclude
MANNER_OF_MOTION, however, they cannot belong to the field. Thus, on the assumption that predicate fields are correlatives of property dimensions, fields of 0-predicates do not comprise all mutually exclusive predicates. Only a subset of the maximal field makes up a field of 0-predicates, namely, the set of those predicates that encodes instantiations of the dimension (the 'P1's in the following) and which, therefore, includes the predicate that encodes the dimension itself(the 'PoPTDIM'). The Po PTDIM is therefore included
by a subset of predicates of the maximal field, while it is excluded by the rest of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
mcludes tlus dimens10n. In contrast with dimensions of obj ect defining
388
0- and
D-Predu:ates
the maximal field, i.e. by the predicates that do not encode instanoations of the dimension (the 'P�1's). Thus, the maximal field can be described as consisting of two subfields which can be characterized by the relation that holds between their members and the predicate POYIDIM: while the P1s include the predicate,
the P�1s exclude it Note that, according to the definition in (s) above, the maximal set of predicates is not a field, because one of the predicates of the set is
included by a subset of predicates. This is a consequence of the fact that the POPIDIM would have to belong to the field, since it is excluded by the P�1s, but
may not belong to the field since it is mcluded by the P1s.
However, a subset (of the maximal set of predicates) that forms a field can be derived if those predicates that cause the violation of the definition in (s) are �
ives a non-maximal set of predicates excluding each other provided that none of the remaining predicates encode any dimensions. Note, however, that an important difference results with respect to the status of the fields: when the P1s are removed, the resulting field is 'object defirung' (i.e. one of the predicates always holds for each member of the universe), bur when the P�Is are removed, the resulting field is 'optional'. In
(7) the set of predicates of speech production and
the two derived fields
are illustrated. Bold letters symbolize the status of a predicate encoding a dimension, Le. a predicate that is included by other predicates of the set Predicates excluding each other are typed in normal letters.
(7)
a.
b. 'object defining' field
maximal set
BE_SILENT
BE_SILENT
PRODUCE_SPEECH WHISPER
c.
'optional' field: PRODUCE-SPEECH
Whether a subset of the predicate in a maximal set forms a field of 0-predicates, therefore, depends on whether there is a predicate which encodes
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
removed from the set This can either be done by removing the P1s or by removing the P Is. What remains in the second case is a non-maximal but consistent set of mutually exclusive predicates. The first strategy sinularly der
Ingnd
Kaufmann 389
a property instantiated by properties that are encoded by other predicates of the sec. On the assumption that the relation between the lexical items of a semantic class is defined by the existence of a predicate that is included by all members of the class, it would follow that it
is
the lexicalization of a number of verbs
encoding the opnonal properties of one dimension that is responsible for the existence of a predicate encoding the dimension. Thus, the inventory oflexical items detennines the complexity of a predicate field and the status of its subsets.
An
empirical difference between verbs represented by D-predicates and
verbs represented by 0-predicates is that, while there are representations with dynamic D-predicates, there seem to be no representations with dynamic 0-predicates (i.e. predicates encoding changes between two instantianng
of object defining properties every change necessarily leads to another property of the same dimenston. in the case of optional properties most changes that occur do not take place within a dimension but between some property of the dimension and a property outside of it. In the case of speech production. for instance, a change that often takes place is that between betng silent (not in the dimension) and talking (one of the properties of the dimension). Much less often. however, one can find changes between two properties of the dimension (for example, between whispering and screaming). While our experience tells us that object defining properties undergo systematic changes (e.g. solid substances melt when the temperature nses, objects change place or shape when they are pushed, individuals grow old as time passes), no such systematic changes are found with respect to optional properties. If changes between optional properties do not occur regularly, there is no sense in lexicalizing a dynamic 0-predicate that encodes such changes. Thus, there is a difference between verbs involving D-predicates and verbs involving 0-predicates. Note that this constraint on the lexicalization of dynamic 0-predicates is motivated by considerations of economy. This means that in prinaple it should be possible for specific changes between 0-predicates belonging to one field to be lex.icalized ifthere is some (for example, cultural) motivation. Although such a possibility exists, however, cases like this are expected to be very rare. There are, however, strategies which can be used to form complex expressions encodtng changes within fields of optional properties. In German. these strategies often involve unaccusarive verbs or their causative counterparts, e.g.
die Stimmt senken 'to lower one's voice' (speech fallen 'fall into a trot' (manner of motion field).
production field) and
in Trab
Though optional properties do not change within a dimension in a systematic way, there is another type of systematic change involving optional properti �, namely, the change by which an object enters or leaves a field. One
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
properties of a dimension).6 This can be explamed as an indirect (conceptually based) consequence of the fact that objects may leave a dimenston of optional properties but not a dimens10n of object defining propemes. While in the case
390
0- and
D-Predtcates
would therefore assume that dynamic 0-predicates may encode a change from a state outside a field into a state where a property of the field holds. This kind of change
is, in fact, lexicalized
in (complex) verbs with the particle
los, as in
lossingen 'start to sing'. lnteresnngly, though, these verbs are not interpreted as denoting the change of a property of their argument but rather as denoting the beginning of an actioiL Note that this means that only (real world) changes that take place within a dimension of properties are conceptualized as changes that an object undergoes and is affected by, while (real world) changes which go beyond the dimension are not evaluated as changes in one of the object's properties. This
can be explained by the fact that, by definition, BECOME is a
a property that holds in the pre-state which is excluded by the property that holds in the post-state, both encoded by predicates that are arguments of BECOME. Since for derived verbs like lossingen the property of the pre-state is not lexically determined, it can only be derived as a complement of the property encoded by the base verb. It is reasonable to assume that an even�ijty
is only interpreted as a change of a property of an individual if the property that is changed can be identified. In the case of a dynamic 0-predicate, the complement .....,p of a predicate P covers all predicates of the field, and since all
of these predicates include the predicate encoding the dimension of the object defining property, this property can be interpreted as the one undergoing a change. However, since in the case of 0-predicates not all predicates that exclude each other belong to the same field, there is no property that is both included by the base verb, and by all predicates excluded by it. Thus, there is no property that could be interpreted as the property of the individual undergong the change, so that the verb
can only be interpreted as encoding the beginning
of an independent eventuality. The question whether the nature of tlus d1fference is semantically encoded,
or whether it should be derived conceptually, will be left open here. If semanncally encoded, the difference would be taken as an indication that 0-predicates do not occur in verbs represented by means of BECOME, and that another semantic operator must be assumed in the representation of the particle
los. If the difference is derived
conceptually, it would be taken as an
interpretational consequence of the fact that no homogeneous property can be identified in the pre-state.
cases
can
Besides like lossingen, one find a number ofverbs that can be analysed as encoding a change in a field aiD-predicates derived by exclud�ttg the P1s from a
maxunal field, hke the one shown in (7b) For example, the deadjectival verb vers .
'become dumb/silent' encodes a change ftom any kind of speech pro ducnon to silence, so that the dimension PRODUCE_SPEECH has to be tummen
interpreted as an instantiating property ofa larger field here. The identification of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
predicate that encodes a change between two properties arranged on a scale. The representanon of a change by means ofBECOME presupposes that there is
Ingnd K:mfrrumn
39 1
PRODUCE_SPEECH as the complement encoding the pre-state is possible here since the adjective stumm itselfis represented by NOT-PRODUCE_SPEECH, so chat it can be derived by negation. From the discussion of 0-predicates it follows that the status of semantic predicates is determined by the inventory of predicates that a specific language makes use of in its lexical representation. Thus, for example, for a language chat does not lexicalize different verbs of manner of motion, there is no optional field chat must be removed from the maximal set Moreover, examples like
vers
tummen
show chat there is a certain degree of flexibility in the classification of predicate fields within a language, smce, by ignoring the fine-grained structure
guage specific. Since they encode object properties chat are conceptuahzations of oncological propemes, however, languages are still expected to entau similar predicates, at least in the core cases. Before turning to the consequences of the distinction between 0- and D-predicates for the properties of unergative and unaccusative verbs in the next section, a few remarks about alternative proposals for the classification of properties are in order. Smce the perception of objects and therr properties is an important question m phllosophy as well as psychology, several accounts of object propemes exist in both. I will not delve into an exploration of the vanous problems and ideas discussed in these areas, since I think that any proposal made ainlS at provtding a solution for an aspect of the problem which 1S relevant within a specific area of research. Therefore, the cnteria which are taken to be relevant for a classification of properties if one is interested in perception differ from the critena which are taken to be relevant if one is interested in their lmguistic categorization. Thus, although I claim chat there is a conceptual basis for the distinction between 0- and D-predicates, the resulting classes are filtered by constraints which are posed by specifically linguistic economy principles. Even with respect to different phenomena in language, different ways of classtfying properties might be relevant For example, as has been argued in a number of papers, the distinction between stage-level and individual-level predicates introduced by Carlson ( 1 977) plays a role for phenomena such as the grammaticality of there-insertion or the interpretation of bare plurals and absolute constructions (see e.g. Kratzer 1 988). Though the distinction of stage level and mdividual-level predicates also reflects the status of properties with respect to an individual, it differs from the distinction between 0- and D-predicates. The main difference lies in the fact that individual-level predicates, in contrast to D-predicates, encode properties chat individual
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of one of the predicates, optional fields can be reanalysed as instantiations of object defining fields in such cases. If this 1S correct, semantic predicates should not be analysed as uruversal or even innate primitives but rather as being lan
392
0- and D-Predu:ares
objects cannot get rid o£ Properties encoded by D-predicates, however, can vary since the individuals may shift from one instantiation to another. Thus, in contrast to ¥ak Frmch (at the moment), which is both an 0-predtcate and a stage-level predicate, know French , for example, is an individual-level predicate
but not a D-predicate. Due to their definition as predicates encoding optional
properties, 0-predicates should overlap with stage-level predicates, which may only hold for a while. Since the main criterion for 0- and D-predicates is not whether they hold forever or not, but whether the properties they encode belong to a field which the argument's referent can leave or not, D-predicates may be classified as either stage-level or individual-level.
U NE R G A T I VE A N D U N A C C U S AT I VE V E R B S
I t has often been observed that two classes o f mtransitive verbs can be dtstinguished by their grammatical behaviour, namely, unergatives and unaccusatives. These classes differ from each other with respect to a number of properties. Thus, for example, unergative verbs in German allow impersonal passive formation and undergo -er-nominalization, while unaccusative verbs
(
do neither see the more detailed discussion in section 3.2 below). In syntactic accounts lt 1S argued that the arguments of unergative verbs behave hke the subj�cts of transinve verbs, while the arguments of unaccusatives behave like direct objects. The difference between the two classes is therefore derived by assuming that the argument of unergative verbs 1s external, while that of unaccusative verbs is internal and must be promoted to subject (see e.g. Perlmutter I 978; Burzio I 986). If this analysis is correct, a clear-cut distinction between the two verb classes is to be expected, i.e. the verbs in each class should behave homogeneously with respect to the unaccusative diagnostics. However, the main problem for the analysis of unaccusativity is that in both classes verbs can be found that deviate or even vary in their behavtour with respect to some of the tests. Since this deviation often seems to be semantically motivated, proposals have been made arguing for a semantic basis for the distinction (e.g. van Valin I 99Q). In more recent syntactic accounts (e.g. Bresnan & Zaenen 1 990; Levin & Rappaporr Hovav I 99S; Pustejovsky I 99 I ; Zaenen
I 993), the differences in the syntactic status of the arguments are derived from semantic structure by linking rules which refer to semantic mformation. Levin & Rappaporr Hovav { I 99 S) present a detailed semantically based analysis of unaccusativity in English which takes the same factors to be relevant as the analysis presented here. In fact, the analysis which will be described in this paper has drawn inspiration to a large extent from earlier papers by these authors.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3
Ingnd Kaufmann
393
There is one important difference, however, between the proposal made here and that of Levin & Rappaport Hovav: although their approach is based on semantic factors, they still take the unergative/unaccusative distinction in English to be represented syntactically (i.e. in terms of argument structure). This position is motivated by what they call the 'deep unaccusative diagnostics', that is, resultative formation and causative alternation in English. Deep unaccusative diagnostics are processes that refer to the lexical properties of the verb and its arguments, while surface unaccusative diagnostics refer to surface syntactic posioons. Levin & Rappaport Hovav class1fy such well-known tests as ne-cliticizaoon in Italian. as well as there-insertion and locative alternation in English, as surface unaccusative diagnostics, which they doubt to be reliable
classified as deep unaccusative diagnostics, in contrast, may not be obscured by discourse functions. Resultative formation. for example, depends m Levin & Rappaport Hovav's analysis on the verb's ability to assign accusative case, which they assume to be a lexically determined property of unergative verbs. Levin & Rappaport Hovav pomt out that other deep unaccusative diagnostics such as impersonal passive formation (agentive) -er- nominalization. and perfect auxiliary choice are often obscured by semantic influences. They therefore regard passive formation. for example, as a necessary but not sufficient condition for unergative verbs. What remains unclear, however, is what kind of relation holds between the semantic effects which obscure these tests and the semantic properties that form the basis of the distinction. In order to explain the behaviour of the verb classes with respect to the diagnostics, in a syntactic approach an explanation must be giVen for the fact that the same semantic factors that determine argument linking also influence syntactic processes, even though these factors are not encoded syntactically in these cases. As far as I can see, however, such an explanation has not been given up to now. In the following I will argue that the semantic factors relevant for the difference in the behaviOur of intransitive verbs can be derived from the more fundamental distinction between the predicate types introduced m section 2 above. I will claim that the behaviour of the verbs is determined by the conceptual interpretation of the predicates that constitute their semantic form. The role of agentivity and telicity, which are semantic properties that are often referred to by proposals with explicit
linking rules, follow as a side-effect of the
core properties of the two predicate classes, just like the syntactic differences. For example, the data in which telicity is argued to play a role always involve
verbs encoded by dynamic D-predicates. I will argue, therefore, that the properties that are usually taken to reflect the internal status of the argument in syntactic approaches actually follow from the fact that the verb encodes a change of an object defining property.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
tests since they are obviously deternuned by discourse funcnons and therefore do not properly reflect the unergative/unaccusative distinction. The processes
394
0- and D-Pred!(:;�tes
representative choice of tests usually considered to be unaccusative diagnostics. I will leave out, however, the syntactic tests presented in Grewendorf ( 1 989), since I share Levin & Rappaport Hovav's scepticism with regard to the vahdity of surface unaccusative diagnostics (see e.g. Levin & Rappaport Hovav 1 995: 289). For a n alternative, non-syntactical analysis of most o f the data presented by Grewendorf see Kaufmann ( 1 995). Before I turn to the discussion of the unaccusative diagnosncs in German, I will briefly introduce the general framework of my approach in the next secnon. In section 3.2 the rradinonal tests for German unaccusanves are exammed: -er-norrunalization, impersonal pas1ve formation, the attributive use of the past participle, and choice of the perfect auxiliary. In section J.2. 5 the perfect auxiliary choice of the Dutch verbs of motion is discussed, since it is not lexicalized, as in German, and therefore it provides evidence for the conceptual basis of auxiliary choice. Finally, in section 4 a semantic analysis of two argument changing processes is proposed, namely, resulrative formation and causative alternation. While Levin & Rappaport Hovav ( 1 995) argue that these processes provide evidence for the syntactic status of unaccusativity in English, it can be shown that this does not hold for German.
3.1
Theframework
The Semantic Form of a lexical item is represented as a decomposed structure consisting of semantic predicates and their arguments. Crucially, SF is assumed to have a srriccly binary structure. Logical relations like conjunction are non commutative, so that one of their two arguments is lower (more deeply embedded) than the other. This hierarchical structuring also has implications for the interpretation of semantic information since, by determining the order
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Although I claim that the so-called unaccusative verbs can be characterized by their semantic representation, I do not assume that there is a semantic class corresponding to the so-called unergative verbs. Rather, 'non-unaccusatives' can be of two types. They can be represented either by static D-predicates or by 0-predicates, and they form a class only in contrast to verbs represented by dynamic 0-predicates, since non-accusatives do not encode a property change. Thus, they do not behave homogeneously with respect to those tests that involve agentiviry or control. Even verbs represented by 0-predicates are not homogeneous in this respect, since only 0-predicates encodmg properties of animates might involve internal control at all. Thus, for example, animates can control their localization properties, while inanimates cannot. Tests that are sensitive to control are therefore expected to lead to different results m these cases. In order to show the relevance of the predicate distinction, I will discuss a
Ingnd
Kaufmann
395
{8)
'Az /,y A.x 3
2
x
CAUSE
e >
(BECOME (y POSS
z))
The syntactic realization of the individual arguments is determined by a linking mechanism by means of which the highest argument (in a nominative accusative language) is assigned nominative case, the lowest argument obtains accusative case, and the middle argument gets dative case. The linking mechanism is a modified version of that suggested by IGparsky {1 989, 1 992), which. however, I will not discuss here (for details see Joppen & Wunderlich 1 994). The elegance of this approach lies in the fact that the connection
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
of the arguments in semantic relations, it renders the information in SF asymmetrical. In the case of conjunction this asymmetry has important consequences for the combination of semannc predicates in SF. If one assumes that more deeply embedded conjuncts must specify the information provided by the higher conjuncts, an interesting constraint is imposed on semantic structures, since then verb meanings may not be sets of arbitrary semantic predicates (see also section 4 below). A consequence of the strict hierarchical ordering of elements in SF is that they can be unambtguously ranked according to their depth of embeddedness. The ranking of the arguments denoting the participants can then be used to derive argument structure, so that no extra linkmg rules need to be assumed. It follows that arguments that are more deeply embedded in SF are linked to deeper positions in argument structure. The binary structure of SF is illustrated for geben 'give' in {8). The representation is formulated in terms of categorial grammar with argument structure represented as a sequence of lambda abstractors. CAUSE, BECOME, and POSS are primitives of SF having type (t, (e, t)), (t, t) and (e, (e, t)), respectively. x, y, z are variables of type e, that is, mdividual variables. Argument saturation is conceived of as functional applicatioiL For reasons of lucidity, the first argument of a relation is placed in front of its functor. Note that the numbers below the A.-abstractors are simply an informal means of showing the ranking of the variables m the 8-grid.
396
0- and
0-Prechcates
between the two dtfferent levels of representation is accounted for with only a few general principles. Moreover, there is no need to assume explicit linkmg rules referring to specific semantic mformanon. As a consequence, systematic (i.e. non-lexicalized) differences between verb classes must be motivated by dif ferences in semannc structure. 3.2
Traditional criteriafor classifying intransitive verbs
(9) test 1. 2.
3·
4-
-er-nominalization past partiople as attribute perfect auxiliary 1mpersonal passive
unerganves
unaccusatives
possible impossible haben 'have' possible
imposstble possible sein 'be' 1mpossible
Examples of two verbs that are classified as unaccusative and unergative accordance with these tests are given in (I o) and ( I I). (w) unergative verbs: lachen
(I I) unaccusative verbs: fallen
gelacht it was laughed 'People laughed' b. der Lacher 'the laughter' c. Der Mann hat gelacht the man has laughed 'The man laughed' d. •der gelachte Mann the laughed man a.
m
Es wurde
a.
*Es wurde gefallen it was fallen
b. •der Faller the faller c. Der Mann ist gefallen the man is fallen 'The man fell' d. der gefallene Mann the fallen man
The verb ausreifien 'flee/run away' in (I 2) shows that these diagnostics, however, do not classify all intransinve verbs as belonging exclusively to one of the two classes.
( 1 2)
Aus diesem Heim wird haufig ausgerissen 'People often run away from this home' b. der AusreiBer the run away-er c. Das Kind ist ausgerissen 'The child ran away' a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
For German (and Dutch) the following tests are generally assumed to be relevant for the classification of unerganve and unaccusative verbs (c£ e.g. Hoekstra I 988; Wunderlich I 98 s):
Ingnd Kaufmann
397
d. das ausgerissene Kind the run-away child In the followmg I will discuss these diagnostics in some detail. For each process, I will propose a lexical analysis first and then discuss the semantic or pragmatic restrictions that play a role and that pose problems for syntactic approaches. In a last step, I will provide monvation for why the properties encoded by the predicate types offer an explanation for the difference in the behaviour of the verbs with respect to the diagnostics. I will argue that, in fact, different kinds of restrictions play a role for the diagnosncs in
(9),
which are
semantic or pragmatic effects of the processes taken as tests.
BieiWisch
(1989) showed that, with respect to event nominalizatons, deverbal
nouns in German display a certain degree of idiOsyncrasy. The same holds for - er-nominalizanons, which are not lexicalized systematically, somenmes due to the existence of lexicalized forms derived by other derivational processes, sometimes without any obvious reason. Since productive derivation is restricted by the blocking effects of eXIsting nouns, this has consequences for the acceptability of spontaneously derived nouns. Therefore, any attempt to provide a semantic (or even syntactic) characterization of - er-nominalization can only reveal the general interpretation strategy, without explaining every single lexicalized form or blocking. This has to be kept in mind when in the following a semantic explanation for the fact that only so-called unergative verbs undergo - er-nominalization is suggested. Examples which provide evidence for the claim that only unergatives, but not unaccusatives, undergo - er-nominalization are given in (1 3) and ( 1 4) respectively.
(I 3) (14)
Arbeiter 'worker', Laufer 'runner', Anrufer 'caller', Blinker 'beamer' *Ankommer 'arriver', *Ertrinker 'drowner', *Schmelzer 'melter', *Wachser 'grower'
It is often assumed that nouns formed by {productive) - er-nominalization in German, as well as in English. refer either to the agent or to the instrument of the base verb.7 Thus, only verbs that assign agent or instrument roles should undergo - a-nominalization. Levin & Rappaport (1 988) argue that it is not the thematic role but rather the syntactic starus of the argument that determines whether - er-nominalization is possible. They claim that - er- nominalization is restricted to verbs with an external argument that becomes the referential argument of the derived nominal. If this is correct, the two types of verbs must differ with respect to whether their argument is realized outside or inside the VP (or, if one assumes that all arguments are generated VP-internally, whether
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
3 .2.1 -er- nominalization
398
0- and
0-Predicates
the argument is realized in SPEC, VP, or in a lower position). However, evidence against generally restricting - er- nominalization to verbs with an external argument comes from the fact that not every verb with an external argument can be thus nominalized, as can be seen from the transitive verbs in ( I sa),8 as well as from the intransitives usually classified as unergatives in ( I sb). ( I s) a. *Nehmer 'taker', •sager 'sayer', *Haber 'haver' b. *Klatscher 'clapper', *Weiner 'cryer', *Scheiner 'shiner' In order to exclude these nominalizations, further restrictions are necessary.
if
-
nominalization cannot be expressed in terms of thematic roles.,..I do not think that this means that it must be accounted for in syntal.-ue- terms. If the characterization of the relevant verb class 1s not restricted to properties of argument strucrure alone, an alternative formulation could take into account the semantic funcnon of the suffix - er. The lexical functor -er which takes a verb as lts argument and derives a noun that charactenzes the external argument of the base verb. In terms of argument structure, - er- nominalization has the effect that the external (i.e. the highest syntactically realized) argument of the base verb becomes the external argument of the derived noun. Since in the case of nouns the external and the referential argument coincide (see Bierwisch 1 989), the external argument of the base verb, in this case, also becomes the referential argument of the noun. While one of the internal (i.e. non-higher) arguments of the base verb may be realized as a genitive complement, the referential argument, 1.e. the situation variable, must be appropriately bound. In contrast to nominalizations involving present or past participles, such as die Fallende 'the one who is falhng', die Gefollene 'the one who has fallen', m the case of - er- nominalization the base verb does not provide any information by means of which the situation could be 'temporally anchored'. I want to propose that 1t is the fact that the situation vanable is not temporally anchored that leads to the resmction on - er nominalization. If no information by which the situation can be temporally anchored is given, the situation variable must be bound by contextual knowledge. The claim is, then, that verbs that are represented by dynamic D-predicates, i.e. the so-called unaccusatives, refer to situations for which, for some reason, contextual binding is restricted. The same should hold for 'unergatives' which may not undergo - er-nominalization either. Since the descriptions that - tr- nouns provide of their referents make reference to the situation denoted by the base verb, the referent is characterized only by its participation in the event. Participation in an event is a suitable characterization of an object only if one of the following two conditions holds:
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
However, semantic restricuons on - er- nominalization must be assumed anyway, 1t can be argued that they are, in fact, the only relevant restrictions. Although I agree with Levin & Rappaport { I 9 8 8} that the resmction on er
Ingnd Kaufmann
399
Typical examples of referents involved in events which can be interpreted habitually are given by nouns denoting occupations (see ( 1 7)) or instruments (see (I 8)), both ofwhich are often formed by - er-norninalizatioiL (I 7) Schreiber 'writer', Arbeiter 'worker', Maler 'painter', Briefrrager 'postman', Backer 'baker', Tellerwascher 'dish-washer', SrraBenfeger 'street-cleaner' (1 8) Mixer 'mixer', Roller 'scooter', Handfeger 'hand-broom', Blinker 'indicator', Sraubsauger 'vacuum cleaner'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
either the referent displays the characteristic property in a specific situation that can be identified by the speaker and the audience, or the referent displays the characteristic property habitually, i.e. is regularly involved in such an event. I will call the first interpretanon the 'specific-situation reading', and the second one the 'habitual reading'. One possible way ofbmdmg the situation variable with respect to a specific situation reading is by having the actual situanon provide the relevant context. For example, imagine you want to refer to a person in an audience whose name you do not know. The easiest way to describe this person to someone else is to refer to the action the person is performing at the moment. This may be done by expressions like der Zwischenrufer da 'the interrupter over there' or der Frager 'the asker'. Since the speaker and the audience share the knowledge about the actions of the described individual, spontaneous descriptions can be easily interpreted if the situation variable is bound in this way. Therefore, - er- nouns referring to participants of an ongoing situation, can be productively denved. Whenever the referent of a - er-noun lS identified as a participant of a specific situation, descnptions of this kind make sense. If no specific situation is identifiable, however, an informative characterization in terms of the role that the intended referent plays in an event is only given if the referent can be identified because he plays this role in events in general, i.e. if he assumes this role habitually. Thus, for example, der Frager can be used to refer to a student on a course who is known to always ask questions. Identifying an individual through a habit, however, only makes sense if this habit is specific enough to distinguish him from other individuals. This is why - er-nouns derived from verbs encoding everyday situanons that anyone is involved in, such as those in (I s) above, are bad as long as they are not further specified (see (I6)). ( I 6) a. •Aufsteher Friihaufsteher nser early-nser b. *Umfaller Umfaller-Partd over-faller over-faller party c. *Sager Ja-Sager sayer yes-sayer d. *Geber Almosengeber g�ver alms-giver
400
0-
and D-Pred1cates
In general, the habitual reading of - er- nominalizatons seems to be more imponant chan the interpretation in which the referent 1S idennfied with respect to a specific situatioiL This may be explained pragmatically by the possibility to nominalize present and past parnciples, which also refer to speofic situations. Note that the interpretation of these nominalizations is less costly than that of the corresponding - er-nouns since the situation is already anchored. Since in the case of- er-nouns the specific-situation reading is not the only one, and other possible interpretations must also be taken into account, nominalizations of participles should be preferred if the specific-situation
Mter having outlined what I assume to be the interpretation of the core cases of - er-nominalization, it remains to be shown how, by means of chis interpretation, the behaviour of the s
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
reading is intended. Note, however, char the most natural way to describe someone by what he is doing is not by using a nominalization, bur by using a noun referring to the kind, such as der Mann 'the man', modified by an adjectival present participle, as in der schlafende Mann 'the sleeping man'. Although the specific-situanon reading is nor frequent wtth productively denved - er-nouns, it is more frequent among leXIcalized ones, as nouns like Anrufer 'caller', Empfiinger 'receiver', Zuhorer 'liStener', Sprecher 'speaker', Sieger 'wiriner' show. These nouns are derived from verbs encoding standard situations which occur regularly bur which m contrast to other standard situations mvolve some unique element (besides the specific participants, and time and region of occurrence). For example, for Sprecher and Sieger this unique element is the relevant message or the specific contest by which the situation can be identified.
Ingrid Kaufmann 40 1 The habirual interpretation, on the other hand, is only appropriate ifthe object can be discriminated from other objects because it regularly undergoes a change of this property. Since changes of object defining properties are generally not controlled, it is improbable in most cases that one mdividual of a kind can be distinguished from others by habirually undergoing a change that other indtviduals of the same kind do not undergo. A better means ofcharactenzing an individual is provided by changes which can be controlled, provided they are speafic enough to give a non-redundant characterization of the individual: FrUhauJs�her 'early riser', which was mentioned above, is one example. The fact that names ofinstruments, which make up a great pan ofthe - er- nouns refemng to t nanirnates, are never derived by unaccusatives follows since there are no To conclude this section, I would like to mention a number of - er- nouns with particle base verbs encoding a change, such as Aussiedler 'evacuator', Aus
reifier
'run away-er',
Ausbrecher
'break out-er' (prison breaker). These nouns
allow the specific-situation interpretation as long as the post-stage of the event snll holds. Examples like these show agam that the contexrual interpretation, and not a difference in an argument strucrure, is relevant for - er nominalization: an interpretation 1s only possible if the siruation variable can be conrexrually bound. Thus, the fact that - er- nominalizations of so-called
unaccusatives are restricted can be explamed by the properties they encode, since these properties do not provide an adequate description of an individual. Note, though. that it does not follow from this explanation of the restriction that all verbs encoding optional properties must undergo
- er- nominalization.
Such verbs are equally subject to the condition that the siruation variable must be bound, and in the habirual reading they must also provide a non-redundant description of their referent. But since among the optional properties of an individual the propemes with respect to which individuals can differ are more numerous, it is more likely for an individual to be non-redundantly described by the optional properties it displays. Thus, the restrictions on - er- nominalization can be explained by the assumption that the situation variable of the base verb must be contexrually bound, together with the pragmatic restriction that the derived description should be at least as informative as competing nouns that already exists, or can
be derived with less effort. J .2.2 Attributive usage of the past participle
( )
The examples in ( 1 9) and 20 illustrate the difference in the behaviour of intransitive verbs with respect to the attributive use of the past participle. The verbs in
(19)
unergative.
are traditionally classified as unaccusative, those in (zo) as
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
instruments whose function IS to change their object defining properties.
402 0-
and 0-Predu:ates
(20) a. *der gearbeitete Mann das geschmolzene Eis 'the melted ice' 'the worked man' b. *das gelachte Madchen b. der errotete Junge 'the blushed boy' 'the laughed girl' c. *die geschienene Sonne c. die vergroBene Wolke 'the enlarged cloud' 'the shone sun' In a syntactic approach these data can be explained by the assumption that the past paniciple is predicated of the direct internal argument Since the argument of unaccusatives is internaL their paniciple can be used attributively, while the participles of unergatives, which have an external argument, cannot A consequence of clus analysis is, however, that for verbs of motion two lexical entries must be assumed, because verbs of motion can be used attnbutively only if a directional PP is realized (see (2 1 )). Thus, their argument must be internal if a duectional PP is selected, and external if not (2 1) a. der *(in den See) gesprungene Mann 'the (into the lake) jumped man' b. das *(durch den Marsch) gelaufene Kind 'the (through the mud) run child' c. die *(m die Berge) gereisten Senioren 'the (into the mountains) travelled old people' ( 1 9)
a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I consider this behaviour of the verbs of motion to be an indication of the semannc character of the relevant constraint Following Wunderlich ( 1 987), I assume that the semantic contribunon of the past paniciple is to introduce perfective aspect LOhner (1988) encodes perfective aspect by means of an operator 'perf' which operates on an event and introduces the state that d1reccly follows the event (the post-state of the event). The parnciple, therefore, provides the information that the mdividuals denoted by the verb's arguments have the property of having panicipated in the event described by the verb at some time before reference time. On the assumption that modification is to be analysed as 8-identificanon (see Higginbotham 1 98 5), an attributive paniciple must provide information that can be predicated of the argument of the head noun. I assume that 'havmg panicipated in a cenain event some time ago' lS not an appropriate attribute. In order to be used attributively, the attributed property must be displayed by the referent of the argument at reference time. What has to be worked out, then, is what kind of verbs provide a post-state in which a panicipant of the event displays a property that has been brought about by the event Since verbs that encode a change of an object defining property of their argument always provide a post-state in which an mstantiation of the object defining property holds and, funhermore, since this instantiation can always be identified by the semantic representation of the base verb, one can predict that verbs represented by dynamic D-prediates proVIde
Ingnd Kaufmann
403
the relevant information about the argument of this prediate. In the case of verbs of motion in (2 1 ) , the verb itself does not identify a region that is the goal of movement Thus, although one can conclude that the moved object must be located somewhere, the specific localization property cannot be inferred. If a directional PP is realized, however, the location property of the individual in the post-state can be identified. The fact that the participle, in the case of transitive verbs, can only be predicated of the object is explained by a general constraint on SF, which. among other structures, excludes dynamic theme predicates from being followed by further predicates whose arguments could
see
be realized as direct objects ( section 4 and Kaufmann 1 99 5). Restrictions on the use of the past participles of transitive verbs show that have to be semantically encoded, but can also be pragmatically inferred. The differences in acceptability in the example in
(22) and (23) can only be derived
from conceptual information concerning differences in the affectedness of the nouns.
(22) a.
das gelesene Buch 'the read book'
b. das getragene Kleid 'the worn dress' c. das geschlagene Kind 'the beaten child'
(23)
a. *das gelesene Schild 'the read sign' b. *der gesehene Film 'the watched movie' c. *der geschlagene Punching-Ball 'the beaten punching-ball'
These examples suggest that what is relevant is that the argument of the participle should be affected by the event in some recoverable way.
An
explanation along these lines also seems to be more suitable for the examples in
(2 1 b), in which it is not so much the localiZation properties of the moved object that seem to be relevant but, rather, its cleanliness. Whatever the correct characterization of the relevant kind of affectedness is, those property changes that are semantically encoded (by dynamic D-predicates) definitely also fall Within the scope of any pragmatic constraint concerning affectedness. As in the case of - er-nominalization, one can argue that, if a semantic or pragmatic restriction is necessary anyway, there is no need for a less restrictive syntactic constraint In the approach suggested here, the difference in the behaviour of intransitive verbs follows from the fact that only dynamic D-predicates provide information about their argument in the post state, while 0-predicates and static D-prediates do not
J.2.J Impersonal passive Passive formation can be characterized as a lexical process by which the highest argument (the subject) is blocked for structural linking, thus allowing the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
what can be predicated of an argument in the post-state does not necessarily
404 0-
and D-Predtcates
lowest argument (the direct object) to become subject The former subject may be realized by a von 'by' PP, or it may be interpreted as being existentially bound. Since, in the case ofpassive formation in intransitive verbs, no argument is left that can be realized as subject, the resulting construction is impersonal. The possibility of impersonal passive formation is often related to the presence of an external argument Having an external argument, though, is not sufficient for a verb to allow passive formation, as the example in (24d) in contrast to (24b) shows. Obviously, in this case arurnacy also plays an important role. The relevance of animacy shows up as soon as the referent of the subject of the base verb is added in a von 'by' PP. Auf der Party tanzten viele Gaste at the party danced many gusts 'At the party many guests danced' b. Auf der Party wurde (von vielen Giisten) getanzt at the party was (by many guests) danced c. Vor dem Fenster tanzten Schneeflocken in front of the window danced snow-flakes 'In front of the window snow-flakes danced' d *Vor dem Fenster wurde (von Schneeflocken) getanzt in front of the window was (by snow-flakes) danced a.
(25) a. Bei Jazzkonzerten sitzen die Zuschauer meistens, aber auf Rock konzerten wird etgenclich immer gestanden in jazz concerts sit the audience usually, but in rock concerts is actually always stood 'In jazz concerts the audience usually sits, but in rock concerts they actually always stand. b. ?Auf der Weide liegen die Kiihe haufig, aber im Stall wird eigenthch immer gestanden in the meadow lie the cows often but in the stable is actually always stood c. •Im Lager liegen die Fahnenstangen normalerweise, aber auf Platzen wird immer gestanden in the depot lie the flag-poles usually but not in squares it is always stood As (25)
shows, even state verbs (with low agentivity) allows impersonal passive formation, provided the subject is human. The contrast between (2 5a), on the one hand, and (25b) and (25c), on the other, suggests that the relevant subject property is notjust animacy, but humanity. In any case, control seems to play an important role in passive formation, which is impossible for inanimate objects.10 The same effect is illustrated by the examples in (26).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(24)
Ingnd Kaufmann
40S
(26) Wegen der guten Thermik wird hier viel von Segelfliegern/??von Mowenr'von Zeirongen geflogen 'Because of the good thermal, there is a lot of flying by gliders/by sea gulls! by newspapers'
3 .2.4 The choice of the perfect auxiliary
Since all transitive verbs take haben 'have' in German (see (27)), it is clearly the default perfect auxiliary. Intransitive verbs traditionally classified as unergatives behave like ttansitives in this respect while unaccusatives take the auxiliary sein 'be' (see (28) and (29)).
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since even unaccusatives like sterben 'die' and follen 'fall' allow passive formation if the subject is understood as controlling the 'acnon', one can conclude that impersonal passive at least is restncted to 'instirotionalized con texts' involvmg animate or sennent individuals, which 1mply control. There fore, the restriction on passive formation cannot be accounted for by syntax alone. An explanation for the restrictions could be provided by the assumption that the demotion of the highest argument is motivated by the 'topicalization' of a lower argument, or, in the case of impersonal passive formation, topicalization of an adverbial that characterizes a secondary aspect of the siruanon encoded by the verb. Restrictions on passtvization would be then expected to be similar to restrictions on topicalization. The differences in acceptability observed in (24), (25), and (26), would chen follow from the fact that there is a correlation between the degree of animacy of the referent and the degree of optionality (i.e. the number of alternatives) of the property encoded by the verb. This hypothesis would explain that verbs that are encoded by dynamic D-predicates rarely undergo passive formation since, wtth respect to the change of object defining properties, there are not many alternatives which could justify this kind of topicalization. It would also explain why transitive verbs like ahneln 'resemble' do not undergo passive formation, and why the acceptability of tmpersonal passive formation is affected by the realization ofadverbials and the agentive von -phrase. A local adverbial, for example, allows the interpretation that a region (in contrast to other regions) can be described by the fact that a specific siroation takes place in it, while the exphcit realization of a vo n -phrase restncts the interpretation to alternatives involving the optional properties of its referent However, whether an analysis of the semantic effects of passiviza tion along these lines is adequate remains to be shown by further mvestigation. Whatever the reasons for the restrictions are, though, the data show that passivization must have access to semantic information, i.e. a syntactic analysis lS not sufficient
400
(27)
0- and D-Prechcates a.
Gestern hat Michael hier einen Vortrag gehalten 'Yesrerday Michael gave a talk here'
Since the auxiliary is the functor which takes the past participle as its argument, a criterion has to be found by which the class of verbs that take sein can be identified. In syntactic approaches this criterion is the internal status of the argument As far as the semantic motivation for the choice of the perfect auxiliary in German is concerned, two main positions can be found in the literature (see Brinkmann 1 992) for a discussion): a) The choice is determined by the thematic role of the verb's subject theme verbs take sein, agent verbs take haben. b) The choice is determined by the event structure of the verb: relic verbs take sein, atelic verbs take haben. Both positions have their shoncomings. The first approach cannot explain why intransitive agentive verbs like tanzen 'dance' take haben when realized without a directional PP (see (Joa)), while they take sein if they occur with a directional PP (see (3ob)). The aspectual criterion can deal with these cases but it cannot explain why a verb like bleibtn 'remain', for example, takes sein rather than habm (see (Joe)). Moreover, it does not explain why relioty is relevant only for intransitive verbs. (30)
Die Giiste haben den ganzen Abend getanzt the guests have the whole evening danced 'The guests danced the whole evening' b. Die Gaste sind in den Saal getanzt the guests are into the ball-room danced 'The guests danced into the ball-room' c. The Giste sind am Tisch sitzen geblieben the guests are at the cable sit remained 'The guests remained sitting at cable' a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
b. Daphne hat ein tiefes Loch gegraben 'Daphne has dug a deep hole' c. Jeanne und Marco haben ein Lied gesungen jeanne and Marco sang a song' (28) a. Gudrun hat gelacht 'Gudrun laughed' b. Kun hat den ganzen Sonntag gearbeitet 'Kun worked the whole Sunday' (29) a. Wahrend des Stromausfalls ist das ganze Eis geschmolzen 'During the power failure the whole ice-cream melted' b. Die Vase ist zerbrochen 'The vase broke'
Ingnd Kaufmann
407
1 Apparent exceptions:
aujhoren 'stop' and anfangen 'begin, start' are examples of verbs that rake ltaben
although they encode some kind of change. Both verbs rake an infinitive verb as their argument encoding either a (non-stative, non-punctual) optional property (see (3 I )) or the change of an object defining property (see (32)). On the assumption that these verbs encode a change that leads to a post-state, aujhoren 'stop' would express the change from the property encoded by the verb to its complement, while anfangen 'start' would encode a change from a complement property to the property encoded by the verb.
(3 I )
Ray hart auf!f:ingt an zu schreien/arbeiten!lachen 'Ray stops/starts shouting/working/laughing' b. Seit sie gealt wurde, hat die Tiir aufgehort!angefangen zu quietschen 'Since it was lubricated, the door has stopped/started squeaking' c. �schi hat aufgehort!angefangen zu sitzen/sich zu setzen 'Uschi stopped/started sitting (stativeYsitting down (punctual)' a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
I will not go into the details of the semantic representation of the perfect auxiliaries and whether auxiliary srin can be analysed as being identical with the copula sein. However, since perfectivity is assumed to be encoded by the past participle here, the perfect auxiliaries do not themselves provide temporal information, whatever their status may be. Therefore, it can be assumed that the perfect auxiliary sein is at least related to the copula, if not actually being the copula irsel£ I therefore assume that the perfect auxiliary sein has the same function as the copula, i.e. as a mediator between a predicate and its argument. Moreover, I assume that the semantics of the past participle imphes that the relevant condition for the choice of sein is a defined post-state that can be predicated of the verb's highest argument. The representation of the verb must therefore include a predicate that encodes a property of the highest argument's referent in the post-state. The choice of haben is the default case, that is, habm occurs if the highest argument is not predicated of in the post-state. 1 1 Note that this explams why bleiben 'stay' rakes sein although i t is not relic. Since bleiben implies that there is no change of state, the post-state must be identical with the pre-state and is therefore defined. It also provides motivation for why transitive verbs always take haben, since the higher argument of a transitive verb never undergoes a change and therefore no defined post-state may be predicated of it. A number of apparent and genuine exceptions to this characterization of the selectional properties of srin have to be considered, though. Genuine exceptions must be treated as lexicalizations and, smce haben is the default auxiliary, they are assumed to occur only in the case of sein. Before I turn to these lexicalizations, I will briefly discuss some verbs that I assume to be apparent exceptions.
408
(3 2)
0- and D-Pred1cates a.
Gestern hat der Schnee aufgehonlangefangen zu schmelzen 'Yesterday the snow stopped /starred melting'
b. *Gestern hat das Hemd aufgehon!angefangen schmutziglweiB zu sein 'Yesterday the shin stopped/starred to be dirty/white
On the basis of the rescricnon of dynamic 0-predicates discussed in section 2, it is easy to show that aujhoren cannot provide a defined post-state when its verbal argument encodes an optional property, since the complement of an
if
0-predicate does not encode a homogeneous property. Similarly, the infini tival argument encodes the change of an object defining property, no specific property can be identified since a predicate encoding the clumge of a property
duration of this post-state can be made explicit by the adverbial fur x Stunden/ Tage 'for x hours/days', as can be seen in (3 3).U This is not possible in the case of
anfongen, as the examples in (34) show. (3 3)
a.
(34)
a.
Richard ging fiir drei Stunden nach Hause
'Richard went home for three hours' b. Die 10 Meter dicke Eisschicht schmilzt im Sommer fur zwei Monate 'The ro-metre-thick ice-layer melts for two months in summer' *Ray 6.ng fur drei Stunden an zu schreien/arbeiten/lachen
'Ray starred shouting/working/laughing for three hours' b. *Die Tiir 6ng fur einen Monat an zu quietschen 'The door starred squeaking for a month' c. •oer Schnee fing fur drei Tage an zu schmelzen 'The snow starred melting for three days' I take the data in (34) as evidence that no measurable post-state is provided by the representation of anfongen and its verbal argument Thus, the choice of the perfect auxiliary
haben
is expected. Idiosyncratic exceptions among the
intransitive verbs that choose haben are the particle verbs zunehmen 'gain (weight, size, etc.)' and abnehmen 'lose (weight, size, etc.)'. The behaviour of these verbs can only
nehmm
be
derived from the transitive status of the base verb
'take', so that the measure phrase
direct object in these cases.
has
to be classified as an optional
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
has no complement In the case of anfongen, it is less evident that a post-state is not provided smce it can be argued that it is the property encoded by the verbal argument that holds m the post-state. However, there is empirical evidence suggesting that this is not the case, and that anfongen, therefore, must be represented in a way that differs from verbs that imply a post-state, i.e. that it should be represented not by means of BECOME. In general, if a verb provides a post-state, the
Ingnd
Kaufmann
409
(35)
Alben hat auf dem Boden gelegen Alben has on the(dacive) floor lain 'Alben was lying on the floor' b. Ingrid hat in der Ecke gestanden Ingrid has in the(dative) comer stood 'Ingrid was standing in the comer' (36) a. Da Albert isch uffm Bodda glagge the Albert is on the(dative) floor lain 'Alben was lying on the floor' b. D'Ingrid isch im Eck gschtande the Ingrid is in the(dative) comer stood 'Ingrid was standing in the comer a.
There is, however, a further difference between Standard German and the South German dialects. While in standard German the change of posicion is expressed by che reflexivized causative verbs of position, i.e. by a grammatically transitive verb (see (37)), in the South German dialects there is an intransitive
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
2. Genuine exceptions: Turning to the lexicalized verbs taking sein , the first to deserve mention is the copula itself While staove verbs in general provide no defined post-state (see the discussion of the verbs of posicion below) and therefore take the perfect auxiliary habm, the German copula takes the auxiliary sein. Besides that, two semantic classes have to be discussed: verbs of motion, which always take sein in German, and verbs of posicion, which take haben in standard German, but s�n in the dialects of South GermaiL Although the auxiliary choice of sein must be regarded as a lexicalized property, it will be shown that there is a conceptual motivation concerning the existence of a defined post-state that is over generalized to a whole semantic class. I will briefly discuss the situation in the South German dialects first and then have a closer look at the verbs of morioiL Verbs of posicion are represented by static predicates that must be classified as optional for movable objects. The participle of these verbs expresses the informacion that a phase in which these properties hold has occurred. However, as is the case with states in general, no information is provided about whether, in the post-state, the same property still holds, whether another of the propert ies of the dimension holds, or whether a change between two properties takes place. This follows from the fact that states are unbounded: the perfect tense can be evaluated with respect to a period during which the state lasted, but this does not necessarily mean that it ended after that period. Thus, the participle of state verbs does not provide a defined post-state and haben should be selected. This is in fact the case in Standard German (see (3 5)), but not in the southern dialects, as is demonstrated for Alernanic in (36).U
410
0- and D-Predtcates
.inchoative variant of the verbs of position (see (3 8)). Since in these cases a defined post-state that can be predicated of the highest argument exists, sein is expected to occur.
(37)
(38)
Albert hat sich auf den Boden gelegt Albert has himself no the(accusative) floor laid 'Albert lay down on the floor' b. Ingrid hat sich in die Ecke gestellt Ingrid has herself in the(accusative) comer stood (trans.) a. oa Albert isch uff da Bodda glagge the Albert is on the(accusative) floor laid 'Albert was lying down on the floor' b. D'Ingrid isch in da Eck gschtande the Ingrid is in the(accusative) corner stood 'Ingrid took her place in the corner'
a.
one could argue that there is only one leXIcal entry with a facultative component BECOME (see (39)). The choice of the perfect auxiliary is determined by the inchoaove variant. If a representation like this is assumed, however, a solution has to be found for the problem of how the situaoon variable of the stative verb can be integrated in the complex representatioiL
(3 9)
l x(l s BECOME(�UE<, s,) LIE(x))
Alternatively, if two lexical entries are assumed, the auxiliary choice must be lexically marked in the case of the stative variant, but can be motivated by an overgeneralization of the inchoative variant. With respect to the attributive use of the past participle, I have argued that verbs of motion such as rennen 'run', laufen 'walk', and schwimmen 'swim' do not provide information about their argument in the post-state as long as no directional PP is realized. Still, such verbs take the perfect auxiliary sein in German (see (40)�
(40)
Uschi ist den ganzen Nachmirrag gerannt/gelaufenlgeschwommen 'Uschi ran/walked/swam the whole afternoon'
As can be seen from the behaviour of some activity verbs such as tanzen 'dance', sein is triggered by the realization of a directional PP (4r a. b), which forces the motion reading and therefore classifies the verb as belonging to the class of verbs of motioiL The same holds for verbs encoding motion which are derived from verbs of stimulus emission (see (41c, d)).
(41)
a.
Die Kinder haben getanzt 'The children danced'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Since the stative and the inchoative variant share the same phonological form,
Ingnd Kaufnunn
41
1
b. Die Kinder sind durch den Saal getanzt 'The children danced through the hall' c.
Der Zug hat gerumpelt
'The train rattled' d. Der Zug ist in die Stadt gerumpelt 'The train rattled into the city' It must be assumed, therefore, that the perfect auxiliary sein is lexically specified to select every verb belonging to the class of verbs of motion, i.e. every verb which includes the predicate MOVE in its semantic representation. The expected difference between a verb complex that only encodes a continuous
German. It can, though, be demonstrated for the verbs of motion in Dutch, will show in the next section.
as
I
3.2.5 Perfect auxiliary choice in Dutch If it is rrue that the choice of the perfect auxiliary in Dutch reflects the semantic properties of the verb complex, the following behaviour of the verbs of motion is to be expected: •
If no PP is realized at all, no specified post-state is given and (the default) hebben 'have' is chosen.
•
Goal and source preposinons, which (directly or indirectly) introduce information about where the individual is localized in the post-state, provide a specified post-state and zijn 'be' is chosen.
If perfect auxiliary choice in Dutch is not only sensitive to a semantically represented post-state but rather to a conceptually derivable one, this should lead to certain differences if path PPs are realized. •
With respect to path prepositions such as door 'through' and over 'over, across', two cases must be distinguished (see Kaufmann 1 993a for a more extensive discussion). - When the traversed object is unbounded (e.g.
through snow and iu),
no
post-state is given, since no information is provided about whether the process (movement within the specified materiaVregion) still holds or not. Therefore
Mbben is expected.
- When the traversed object is bounded and likely to be completely traversed (e.g. owr tht bridgt), a specified post-state can be inferred. In this case,
the object is localized in the region where the path ends. Since
for example, encodes the
door,
crossing of an interior space, the post-state of
moving through something must be the localization behind the object of
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
change oflocation, and a verb complex that provides a defined post-state due to the semantic contribution of a directional PP, can thus not be demonstrated in
412
0- and D-Pred1cates reference. If the object of reference trulY plausibly be either traversed or stayed in (through the room,
through theforest), both readings are possible and
the context will trigger the interpretation. The expected behaviour of Dutch verbs of motion without a PP and with goal and source PPs is well attested in the literature; see e.g. van Hour, Randall & Weifknborn (1993), who give the examples in (42).
(42)
a.
De hal heefrl*is minurenlang (in her rand) gerald 'the ball has/*is minutes-long in the round rolled 'The ball has rolled around for minutes'
b. De hal *islheeft in 2 sekonden van de tafel af gerald
2 seconds from the table rolled 'The ball has rolled off the table in 2 seconds' The predictions about the prepositions
door and over have been verified by an
informal mqmry of eight native speakers of Dutch developed in collaboranon with Ursula Brinkmann. The test persons were presented with a sentence with a past tense verb and asked to transform it into the present perfect. They were told that the test is about regular and irregular participle fortru1tion, which unfortunately had the effect that two of them dropped the auxiliary in some cases. The corpus
of test sentences is given in (43); the two PPs given in each case
were presented to different test groups. A number of filler sentences involving non-motional verbs were added in the original test. As the examples show, each verb has been tested with PPs involving nouns that force the bounded interpretation as well as with PPs that allow the unbounded interpretation. Furthermore, the selection of verbs reflects the fact that the component which encodes the manner of motion may provide different amounts of information that can also be conceptualized independently of 'pure motion'. For example, while 'walk' is simply a way of moVlng by using one's own legs, 'hike' and )og' are also sporting activities.
As Brinkmann (1992) noted, the focus upon of the sporting activity aspect of these verbs also influences auxiliary choice. The selection of haben by the German verb tanzen shows that it is not classifie d as a verb of motion in its basic use. Thus, Dutch dansen is also a candidate for a genuine non-motional verb, which, therefore, could be expected to choose
hebben. In (43), the number of
tested persons that chose hebbm and zijn is given below the example. U and B indicate which reading is possible, the former standing for 'unbounded reading', the latter for 'bounded reading'.
(43)
a.
Het hondje sprang over her graslover her hekje 'The little dogjumped over the grass!over the hedge' /B : hebben: o zijn: 3 U: hebben: o njn: 4
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
the ball *is/has in
Ingrid K.:mfmann
41 3
h. De pilot vloog over de grens/over de woestij n
'The pilot flew over the border/over the desert' IU,B: hebben: o zijn: 3 B: hebben: o zijn: 4 1. The paar danste door de kamer/doer de deurporten 'The couple danced through the room/through the door posts' /B : hebben: 2 zijn: I U/B: hebben: 3 zijn: I J- De rnieren kroopen door het gatje in de mur/door de kamer 'The aunts creeped through the small hole in the wall/through the room' B: hebben: o zijn: 4 IU,B: hebben: I zijn: 2
(44)
frequency
general
bounded only
zijn
47
28
hebben
21
2
unbounded only
bounded or unbounded
4 (gesprongen)
15
4 (gewandelt)
15
Genera.yy, the answers show that when the 'bounded' interpretanon o fa PP was possible the auxiliary zijn was chosen in nearly all cases. This confirms the hypothesis that, if a specified post-state can be derived, zijn is possible. In cases
in which both the bounded and the unbounded interpretation were possible, both auxiliaries were chosen by different persons, though the choiced of zijn dominated in the case of 'pure' motion verbs. hebben was preferred only in the case of gewandeld 'hiked', gedanst 'danced', and gejogd jogged', which can be
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
b. Rita zwom door het meertje/over de lijn 'Rita swam through the lake/over the line' /B : hebben: o zijn: 2 U,B: hebben: I zijn: 3 c. Paul wandelde door de bergen/door de felden 'Paul hiked/wandered through the mountain/through the field' U: hebbe11:4 zijn: o IU,B: hebben: 3 zijn: o d. Marlies fietste door het stadjelover de finishlijn 'Marlies rode her bicycle through the town/over the finish line' U,B: hebben: o zijn: 4 /B: hebben: o zijn: 3 e. Ted liep door de deuropening/door het bas 'Ted walked through the door-frame/through the wood' B: hebben: o zijn: 4 IU,B: hebben: 3 zijn: o £ Het jongetje rente over de bruglover het voetbalveld' 'the boy ran over the bridge/over the football ground' B: hebben: o zijn: 4 IU,B: hebben: I z�n: I g. Peter jogde door de park/door de tunnel 'Peter jogged through the park/through the tunnel' 1• /B: hebben: o zijn: 3 U,B: hebben: 3 zijn: I
4 1 4 0-
and D-Pred1cates
interpreted as sporting activities. Since, in the case of owr and door, the post state can only be derived through conceptual knowledge, the semantic encoding of the post-state seems not to be necessary for perfect auxiliary choice in Dutch. The examples in (45) give further evidence for the assumption that the choice ofauxiliary is actually triggered by conceptual information, since the atooliary changes if the delimiting rime adverbial in 3 uuren 'in 3 hours' is added, as was confirmed by the testees.
(45)
Paul heeft door her bos gewandelt Paul has through the forest wandered 'Paul wandered through the forest' b. Paul is in 3 uren door her bos gewandelt Paul is in 3 hours through the forest wandered 'Paul wandered through the forest in 3 hours'
a.
describe possible ways to cover a distance, but acoviries for which reaching a goal is not primary. As long as the boundedness of the situation can only be conceptually derived, therefore, there are two options with these verbs: they can be classified as activities, focusing upon the 0-predicate and ignoring the fact that the change of locarion could lead to a specified post-state, orFey can be classified as verbs of motion, focusing upon the movement and the location that results from it In the first case, hebbm is chosen, in the second zijn. With 'pure' verbs of motion, the first reading is only triggered if the speafic manner of motion is contrasted with another one. Note that an explanation along these lines has already been given by Brinkmann (1 992) for the observation that with door and with the goal preposition naar 'to', which does not require the
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What the difference in the behaviour of'pure' verbs of motion and the sporting activity verbs illustrates is the influence of the second predicate in the represen tation of these verbs, which encodes the 'manner of motion'. While MOVE is a dynamic D-predicate since it encodes the change of the object defining property ofbeing located, predicates such as WALK, SWIM, and DANCE must be dasstfied as 0-predicates since they are only relevant if the referent of their argument is moving. If the individual does not move, none of the predicates of the field holds. Thus, verbs of motion include an 0- and a dynanuc D-predicate. Since the criterion that is relevant for the choice of zijn is always fulfilled if a post-state is semantically encoded, no conflict arises in cases where such a state is provided by means of goal-PPs. If the decisiOn whether there is a post-state or not, though, has to be made on the basis of conceptual information which leaves both options open, the character of the properties encoded by the predicates is drawn upoiL In the case of pure verbs of motion the acnvity encoded by the 0-predicate only specifies how the movement takes place (for example, step by step in the case of WALK). Verbs such as dance, jog, and hik�. m contrast, not only
Ingnd
Kaufmann
41 5
reaching of a goal either, both Iubben and zijn are possible with different mterpretacions. Auxiliary choice for the verbs of motion in Dutch may therefore be accounted for by the assumption that a specific post-state must be derivable. Note that, although the verbs of motion differ in German and Dutch as far as the choice of the perfect auxiliary is concerned, there is no need to conclude that their semantic representations also differ. This is supported by the fact that they behave in a similar maner with respect to the remaining tests, particularly as far as the attributive use of the past participle is concerned. Thus, it is reasonable to assume that the choice of the perfect auxiliary lS lexicalized m the case of the German verbs of motion but transparent in Dutch.
4 LE X I C AL D E R I V A T I O N : RE S U L T A T I VE C O N S T R U C T I O N S A N D T H E I N C H O A T I VE - C A U S A T I VE A L TE R N AT I O N Resultacive formation introduces a result predicate which predicates of a direct object that is not necessarily subcategorized by the base verb. Examples for resultacive constructions with tramicive base verbs are given in (46).
(46)
a.
Karl iBt seinen Teller leer 'Karl eats his plate empty' a'. *Karl iBt seinen Teller 'Karl eats his plate'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
So far the discussion has shown that each of the four traditional tests for unaccusacivity is based on semantic informacion, although different semantic aspects are relevant. The hypothesis that unerganve and unaccusacive verbs do not form grammatically relevant classes is confirmed by the fact that the differences in the behaviour of the verbs with respect to the tests can be explained by the property distinction introduced in section 2. Thus, I claim that intransitive verbs only have to be classified by their semantic representation, and not by the syntacnc status of their arguments. Since a semantic explanation does not involve the claim that each verb must behave homogeneously with respect to all the tests, it provides a more adequate analysis of the data. However, the tests discussed up to now do not belong to those that, as Levin & Rappaport Hovav (1 995) argue, provide clear evidence for the syntactic character of the distinction m English, namely, resultacive formation and inherent causicivizacioiL In the next section, I will discuss the corresponding German constructions, and show that in these cases, too, a syntactic analysis is not necessary and that, furthermore, the predicate distinction plays an important role.
416
0- and
D-Predicates
b. Marianne schreibt ihren FUller leer 'Marianne writes her pen empty' b'. *Marianne schreibt ihren Fuller 'Marianne writes her pen' Simpson ( 1 98 3) noted that unaccusative and unergarive verbs behave differendy with respect to resultative formation. While in the case of unaccusa tives a resultative predicate may be added without the introduction of a further argument, unergauves require an additional argument which can but does not have to be coreferent with the subject. This is never the case with the resulta tives of unaccusatives. a. She danced *(herself) tired b. The ice melted (*itself) liquid c. She danced her shoes to tatters d. *The ice melted the floor wet
most approaches dealing with resultative construcnons, this behavwur lS traced back to the fact that, if the verb itself has no direct object, resultative formation introduces a direct object as an argument of the resultative predicate (see e.g. Carrier & Randall in press; Simpson 1 98 3). If the verb itself already has a direct object, as in the case of unaccusatives, the result pred1cate predicates of this argument. In order to caprure this difference, these authors conclude that the semantically motivated distincuon of unergative and unaccusative verbs must be encoded syntactically. A look at the German resultative constructions, however, shows that there are reasons to believe that resultative formation is not possible with unaccusa tives at all. The only constructions that are possible involve a directional PP (i.e. a PP encoding a change) acung as a resultative predicate that only specifies the change the verb already encodes15 (see (48)). Unergatives, in contrast, allow either resultative adjectives or PPs, which may introduce information about a change not encoded by the base verb (see (49)). In
(48) a. *Das Eis schmilzt fli.issig The ice melts liquid' b. Das Eis schmilzt zu einer Pfiitze The ice melts to a pool' c. *Das Steak verbrennt schwarz The steak burns black' d. Das Steak verbrennt zu Holzkohle The steak burns to charcoal' e. *Das Steak verbrennt zu einer bitteren Mahlzeit The steak burns to a bitter meal'
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
(47)
Ingnd Kaufmann
(49)
a.
417
Er lauft den Rasen plan
'He runs the lawn flat' b. Er lauft den Rasen zu Marsch 'He runs the lawn to mud' c.
Sie arbeitet sich miide
'She works herself tired' d. Sie arbeitet s1ch in einen Rausch 'She works herself into a frenzy'
(so) a. �r schmilzt das Eis fliissig 'He melts the ice liquid' b. Er schmilzt das Eis zu einer Pfi.itze 'He melts the ice to a pool' c. 1:r verbrennt das Steak schwarz 'He burns the steak black' d. Er verbrennt das Steak du Holzkohle 'He burns the steak to charcoal' The observation that only unergatives undergo resultanve formation cannot be counted as evidence for a semantic restricnon, since it follows e.g. from the assumption that resultative formation is restricted to verbs that assign accusative case. What must be shown is how the restriction of resultative formation to unergative verbs can be explained in purely semantic terms. On the assumption that resultative formation is a lexical rule which extends the semantic representation of the base verb by introducing a result predicate (a predicate embedded by BECOME) and its argument, this restriction can in fact be explained by a general constraint on semantic representations which provides an interpretation of hierarchical ordering of the predicates in SF see section
3· I ).
(
The rule for resultative formation is shown in (s I ), and the
constraint is given in
(52).
(See Kaufmann
(1 995)
for a discussion of the
motivation of this constraint.) (s I ) Resultative formation:
VERB(. . .) - VERB(. . .) & BECOME(<, e, P(y))
(52) Principle for the interpretation of asynunetric conjunction (PIC)
In a decomposed strucrure, the predicate that is embedded in the highest
position in SF activates those sortal properties of its arguments that are selectionally relevant. Any more deeply embedded predicate must specify
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
A possible syntactic explanation o f the data i n (48) could b e that i n German resultative adjectives are not allowed to be predicated of subjects. However, evidence against this explananon is provided by the transitive counterparts m (so), which behave in exactly the same way even though m this case the argument of the resultanve predicate is the direct object.
4 1 8 0-
and D-Predtcates
these properties. Specification lS restricted to those properties also referred to by the immediately preceding predicate. As an example for sortal properties that are activated consider the representa tion for the verb gehen 'walk' in (S J ).
(53)
A. P ,l.x (WALK(x) & MOVE(x) & P(x))
.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
According to the PIC, it lS the first predicate WALK that activates the relevant properties. Assuming that the conceptual interpretation of WALK can be described as the alternating contact of the referent's feet with a supporting ground, such that each foot is located on the ground and then removed, what is activated (besides general properties such as animacy and physical properties) 1s the locational property of the individual as well as the informanon that different regions are involved. MOVE specifies the informanon about these regioru by introducing the sequence of regions (the path) in which the individual is located in each part of the situatioiL The predicate variable P, finally, can only be saturated by a predicate that specifies this path further, i.e. by a directional PP. The observed resmctions on resultative formanon With respect to so-called unaccusatives can now be explained straightforwardly 1f the fact that these verbs encode the change of an object defining property is taken into considera tioiL Since this change concerns no other propemes, any introduction of a component encoding some further change of another individual by the rule of resultative formation would violate the PIC. Note that, due to the formulanon of resultative formation, a new direct object must be introduced in any case Th11S, even in resultative constructions with a reflexive direct object, the co indexation of the arguments is not encoded in SF itsel£ This is why the resulta tive formation of verbs encoded by dynamic D-predicates is also impossible with reflexives, since their SF would violate the PIC as soon as the resultative component is introduced. The motivation for the assumption that, at least in German, modification by a specifying directional PP has not to be analysed as an instance of resultative formation is gJ.Ven by the fact that these PPs themselves encode a change. As they may specify the change of location in the case of the verbs of motion, these PPs may also specify other kinds of property change. Adjectives, though, do not themselves encode changes and can only receive a resultative interpretation if the verb provides information about the change, as e.g. in the case of werden 'become' or by resultative formatioiL Verbs represented by 0-predicates often encode concepts that are more complex than those represented by dynamic D-predicates, and therefore activate more properties. Since, for irlSrance, all optional properties involving agentiviry activate what one could call properties of the 'physical state' of an individual, result predicates like erschopfi 'exha11Sted' are always possible with
Ingrid K2ufmann
419
these verbs. This, however, is not a consequence of their being optional properties. There are also verbs encoding optional properties that do not allow resultative construction, simply because they do not involve anything that can be interpreted as having an effect on something. Examples are sagen 'say', hassen 'hate', and holfen 'hope'. Funher evidence for the relevance of the PIC is gtven by constraints on the result: resultative formation is possible only if the verb provides some informa tion which can be interpreted as the influence of the agent upon the new patient, and which possibly leads to the effect encoded by the result predicate. (54)
Karl iBt seinen Teller leer/*seinen Pullover schmutzig. 'Karl eats his plate empty/his pullover dirty.' b. Die Kunden stehen den Rasen platt!*die Wand dreckig. 'The customers stand the lawn flat!the wall dirty.'
a.
To conclude the discussion on unaccusativiry, I will show that the semantic restriction on the inchoative-causative alternation also follows from the predicate distinction together with the PIC. As the examples in (49) and (so) show, only so-called unaccusative verbs may occur in morphologically unmarked causative-inchoative pairs in German. (49)
Der Stoch brach 'The stick broke' b. Der Mann brach den Stock 'The man broke the stick' c. Die Butter schmolz 'The butter melted' a.
(so)
Das Baby lachte 'The baby laughed' b. •1ch lachte das Baby 'I laughed the baby' c. Der Mann arbeitete 'The man worked' a.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
While the container of the food in (54a) can be conceptually derived from the interpretion ofEAT, the clothes cannot Similarly, STAND provides informa tion about the suppomng ground, but nothing about the sides. Note, though, that since resultative constructions are not lexicahzed, they can draw on conceptual and encyclopaedic knowledge to some extent Still, resultative constructions involving result predtcates that cannot be motivated by the infor rnanon encoded by the predicate of the base verb are expected to be either idiO matic or less acceptable. According to the explanation given here to account for the restrictions on resultative formation, the behaviour of the so-called unaccusatives is derived in the same way as the semantic restrictions on the result predicates in the case of the resultatives of unergative or transitive verbs. Instead of assuming that the verb is able to assign accusative case, thus allowing the introduction of a new direct object, in this account it is only the semantic potential of the verb that is relevant Whether this explanation also holds for Enghsh remains to be examined.
420 0-
and D-Predtcates
d. Ich schmolz die Butter 'I melted the butter'
d. *Ich arbeitete den Mann 'I worked the man'
Similar inchoative-causative verbs can be derived by adjectives, in which case the inchoative variant
is
realized with a thematically empty reflexive which
often encodes middle functions in German, i.e. in example interpretation.
(5 I )
( 5 I b),
the neuter
a. Der Mann weitet die Hose
'The man widens the trousers'
b. Die Hasen weiten sich 'The trousers widen'
(
D-predicates. Levin & Rappaport Hovav I 995 ) argue that causative-mchoative
pairs are semantically transitive, and that, in their intransitive use, only the
mternal argument is realized. In the following, I will try to give an explanation of the observation that only those predicates that also represent so-called
unaccusative verbs may follow the semantic predicate encoding causation in the SF of inherent causative verbs. Tlus explanation again draws upon the restriction formulated in the PIC.
In contrast with verbs that require the penphrastic causative construction with lassen 'let!rnake', which encodes an indirect causation of the event encoded by the causativized verb, inherent causative verbs encode direct
causation. Thus, the situation encoded by the corresponding intransitive verb is
caused by direct physical manipulation of the causee by the causer. Through physical manipulation, only those physical properties of the patient that can be
manipulated by an external agent are acnvated and may be further specified by
a suitable predicate. Since it is mostly object defining properties that can be manipulated physically, it follows that predicates that may specify the activated properties are restricted to a subclass of those that encode ur).accusative verbs.
Thus, the claim
is
that, as in the case of the resultative tonstruction, the
restriction is conceptually rather than syntactically motivated. Inchoative-causative pairs like those m (49) and
( 5 1 ) are often represented as shown m (52), where the Semantic Form of the mchoative verb is embedded by
the operator CAUSE.
(52)
BECOME (soLID<. e, FLUID(y)) b. CAUSE(x, BECOME(roLID<, e, FLUID(y))) a.
However, since the same operator CAUSE is introduced both by morphologi cal and syntactic causativization (the latter being possible for unergative verbs),
no semantic restrictions can be motivated by this representation. I assume,
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
What has to be explained is why inherent causative verbs are limited to alternating with unaccusanve verbs, i.e. verbs encoded by dynamic
Ingrid K:mfmann
42 1
therefore, that the causative variant of inchoative-causative pairs should be represented by a relation that encodes an unspecified physical manipulation of the agent upon the patient, as can be seen in (S J).
(s J)
BECOME(s01.m<. e, FLUID(y)) b. MANIPULATE(x. y) & BECOME{wLID<, e, FLUID(y))
a.
s C O N C L U S I O N AND O UTLOOK In this paper, a conceptually motivated distinction of two types o f semantic predicates has been proposed. The source of the diSnnction is the concept associated with nouns encoding the objects which can function as referents of the arguments of verbs. On the assumption that a concept may be regarded as the set of all properties (dimensions) that the individual objects belonging to a kind might display, a distinction can be made between those properties that each individual object must always display in some way or other (i.e. the object defining properties) and those properties that an individual can, but need not, display (i.e. optional properties). This difference in the status of the properties that are relevant for the conceptualization of objects is reflecated by the organization of the semantic predicates that encode the instantiations of the properties. From the predicate field whose members encode the properties of one dimension, one predicate must always be predicated of each object if the field corresponds to an object defining dimension, but not if it corresponds to an optional dimensioiL Property changes are only encoded with respect to fields encoding object defining properties.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It follows from the PIC now that MANIPULATE (which encodes physical manipulation) can only be followed by predicates encoding properties or changes of propemes that result from physical manipulatioiL Thus, the fact that so-called unaccusatives differ from unergatives with respect to causativiza tion, in that only the former allow inherent causative alternants while the latter can only undergo syntactic causativization, follows from the fact that unaccusatives are represented by semantic predicates that encode a change of object definmg properties which can be understood as resulting from physical manipulatioiL Note that this explanation again makes weaker predictions than a syntactic account which associates the two types of causativization with the syntactic status of the argument of the causativized verb. It neither predicts that all verbs represented by 0-predicates must undergo periphrastic causativiza tion, nor that verbs represented by dynamic D-predicates may not. In every case it is the individual verb meaning that decides whether the encoded property may be brought about by direct or indrrect causation, or by neither.
422 0-
and D-Prechcates
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
It is important to note that the resmcnon of dynamic predicates to D-predicates is a genuine linguistic constraint, and that since the inventory of semantic predicates depends on the lexical material of a language, it may lead to language specific differences. This follows from the fact that the status of a field of 0-predicates arises from the complexity of the maximal field of predicates that exclude each other. Thus, even if the conceptualization of objects is the same for the speakers of two languages, assuming that these languages differ with respect to which of the instantiating properties of the dimensions are acrually encoded by semantic predicates, differences in the status of these predicates can result. From the language specific character of the status of the predicates, and from the fact that the statuS depends on the inventory oflexical items, it follows that in the course of language acquisition overgeneralizanons are to be expected with respect to operations which are sensitive to the status of the field (such as the derivation of inchoative verbs). This should be so until the child has acquired the relevant vocabulary that determines the status of the field. Since the more lexical items a child learns, the more complex the fields become, it is to be expected that the child may start with fields ofD-predicates and end up with fields of 0-predicates, but not the other way round. In the second part of the paper I have shown that the distinction between the predicate classes allows for a semantic analysis of the unaccusative/unergative distinction in German. I have argued that the only verb class that can be identified by its behaviour with respect to the unergative/unaccusative diagnostics is the class of verbs encoding changes of properties. The rest of the intransitive verbs, however, do not form a homogeneous class, neither in their semantic representation, nor in their behaviour with respect to the diagnostics. This interpretation ofthe facts in the case oflanguages like German, Dutch, and Italian is supported by the observation that the tests apply systematically to the semantic verb classes. This does not mean, however, that the phenomenon under discussion must be purely conceptually based in every language. For example, although most of the arguments presented here also hold for English, it remains to be investigated whether an explanation can be found for the syntactic data which shows that, as has been observed by Carrier & Randall (in press) and Levin & Rappaport Hovav ( 1 995), there is a difference in the status of the direct object in the resultative constructions of transitive verbs and of intransitive verbs in English. Otherwise, the distinction between the predicate types can be used for the formulation of linking rules like those proposed by Levin & Rappaport Hovav. As far as I can see, their linking rules can be captured by the generalization that arguments of D-predicates are linked to the direct internal argument (direct object), while arguments of 0-predicates are linked to the external argument Although a purely semantic analysis of unaccusativity is to be preferred, since it avoids redundant encoding, a grammaticalization of the relevant difference in certain languages would not be surprising.
lngnd Kaufmann 423 However, since grammaticalization generally obscures the basic semantic phenomena, the fact that the effects of the properties encoded by the semantic predicates show up so clearly in languages like German, Dutch, and Italian provides eVldence against grammaticalization. I assume that a grammaticaliza tion of the conceptual dtfference actually occurs in languages with active/inac tive case marking, such as Lakhota, Central Porno, Mohawk, Caddo and Guarani (see e.g. Mithun
1 99 1 ). G rammaticalization in these languages has the
effect that the distinction between the two predicates 1s encoded in a way that seems
to
be
semantically
transparent,
but
is
cross-linguistically less
homogeneous than the S
As
m
the case of one-place predicates, relations can be disnnguished with
respect to the propemes they encode. After saturation of the inner argument, 0-relations encode the opnonal properties of the object denoted by their higher argument, while D-relations encode its object defining properties. While most relations are optional, only a small number of object defining rel ations seem to exist, ·namely, different kinds of affiliation relations, such as
PART_OF or the localization relation, which has been treated together with the one-place D-predicates here. What seems to be characteristic of object defining relations is the fact that both arguments seem to be involved in the
relation in a similar manner. This is reflected by the existence of lexicalized instances of semantic converse relations like 'x located in y'-'y contains x' or 'x
possesses y'-'y belongs to x'. Optional relations, on the other hand, involve a conceptual asymmetry between their arguments that can often be interpreted as a directedness of the relation from the higher towards the lower argument.
As in the case of intransitive verbs, typical properties of arguments of transitive and ditransitive verbs can be derived from the character of the conceptual rel ations that are encoded by the semantic relations in their SF (see Kaufmann
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
verbs into two discrete classes, no room is le& for subtle differences in degree of
424
0- and D-Pred1<:ates
1994a, 1 994b, 1 995). It can thus be shown that the predicate distinction and its conceprual motivation also plays a role for other verb classes.
Acknowledgements T1us paper is a rnr�sed version of a chapter of Kaufmann (1 995� I would hke to thank Ursula Bnnknunn, Birgit Gerlach, Sandra Joppen. Claudta Matenbom, Ralf Naumann, Albert Ortmann, Barbara Snebels and Dteter Wunderltch for comments on earher versiOns and discusston during the dtfferent stages of development of thts paper, and especully Ray Fabn both for diSCUSSion and for helpmg me wtth my English. Furthermore, I thank Manfred BleiWISCh and rwo anonymous reviewers, whose cnocal rernarb forced me to clanfy the nooon of 0- and D-predicates and rethmk my argumennoon.
Received: o6.o2 95 Revised vernon received. 09.00.95
Hnnnch-Hrin�-Unr�tat �ldoif Univc:sitatsstr 1
4 0215 �/doif Gmnarry
t-TnQr/:kaufmann�sap rr.lrng unr-du=ldo ifd�
N O TE S 1
Note that the nonon 'dimension' does not tmply any ordenng of 1nstannanng propemes. 2 In Kaufmann (1 995), I used the labels 'agent predicates' and 'theme predicates' followtng the thematic role names of the protorypical arguments of these predt cates. However, smce the arguments of the predicates do not have to bear these roles, and since I do not take thematic roles to be semannc prirrutives, I wtll use the less suggestive labels 'D-' and '0predlcates' here. 3 The set should be 11UXlmal m the sense that there IS no super set of formulas that also fulfils the defirution. 4 The locahzaoon field therefore does not consist of one-place D-pred1cates, but of D-relations. See Kaufmann (1 995) for a discussion ofD-relations. 5 ThiS fact has been pomted out to me by art anonymous reviewer. 6 Note that the use of the nooon 'dynamic 0-predicate' to encode a change berween
7
predtcates should not suggest that the non-dynarruc O-pred1cates must be scanc. In conrrast to (non-dynam1c) D-predicates, which, as far as I can see, in fact always encode scanc properties, verbs represented by O-pred1cates often encode dynarruc siruanons. In pamcular O-pred1cates sornlly resmcted to ani mates or humans, such as WORK or TALK, involve dynamics. However, the dynamic character of these predicates dtffer from that of dynarruc D-pred1cates m that there IS no change of a property of the argument's referenL What IS relevant m the followtng discuss1on is this sense of the term 'dynamic' as a change of properry. In German, there is also a small number of lextcahzed - " nouns such as Lach" 'laughter', Briilkr 'shouter', Rulpur 'belcher' which refer to the 'enory' produced by the action. I wtll ignore these nouns here since they cannot be denved productively.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
INGRID KAUFMANN
&mrnarfu r Al/gnn nn� Sprachwissnrschafi
Ingnd K2ufmann 425 ru
1S a further resmroon on the adverbial, namely, that the post-state must be reversible. Furthermore, the Information how the end of post-state 15 brought about must be denvable e1ther from the context or from the verb mel£ However, smce the examples in (3 I) involve agentive verbs, the agent Itself has control over the duranon. Thus, this resmcoon CaiiDot be relevant for the macceptability. I 3 I owe the Alernaruc examples to Albert Ortmann, who also drew my artenoon to the exiStence of the mchoarive vanants diSCUSSed below 1 4 Interesongly, two of the t�t persons who chose zyn here had chosen hebbtn first and then corrected themselves later. I 5 See also Pustejovsky (I 99 I), who argues that formaoon of resultanve consrmc oons must mtroduce a change m event srmcrure; otherWISe the predicate spea6es ony the informaoon already giVen by the verb. Smce unaccusaoves already encode a change, they do not fulfil tlus cnterion.
I 2 There
RE F E R E N CE S Beeh, V� Brosch, � Schulz, Kl-D. (I994), 'Feldsrrukruren 1m Lex!kon', Arbtitspapim d($ SFB 282 'Thron� � Laicons', NR 52, H.-Heme-Uruversicir Diislse dor£ B1erwisch. M (I983), 'Semanrische und konzepruelle Repcisentaoon lex.Ikali5cher Emhe1ten', in R Ruzicka & W. Morsch (eds), Untmuchungm zur Stmantrk, Srudia grammanca XXII, Ahderrue, Berhn, 6I-
sional Adj«twrs: Grammatrcal Structur� and Conuptual lnUrprnatron, Spnnger Verlag,
B1erwisch, M (I989), 'Event nominalizarions: proposals and problems', m W. Morsch (ed.), Wortstrulaur und Satzstruktur (Lin gui.sti5che Reihe A I 94). Akaderrue der lSC aften der DDR, ZISW, Berlin, WLSSelh
Berlm-Heidelberg-New York. Bresnan, J� Zaenen, A (I990), 'Deep uwc cusariVIty m LFG', in K. Dziwirek, P. Farrel, & E. Metas-Blkad1 (eds), Grammati cal Rdations, CSLI, Stanford, 45-57· Brinkmann, U. (I992), 'Choice of auxiliary for intraOSirive verbs of motion: an analysis of an uwccusative d�agnosncs', MS, N1jmegen. Bul"Zlo, L (I986), Italian Syntax, Re1del, Dordrecht. Carlson, G. (I977), 'A uru6ed analysis of the English bare plural', Unguistics and Philo
I-73· Bierwisch. � Lang,
Carrier, J. & Randall, J.
99·
E.
(eds) (1989),
D1mm-
sophy. J, 41)-98.
(in
press), From
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
an anonymous rCVIewer noted, all of the verbs cued rn ( 1 0) have an obllgatory direct object. Thus, one could assume that these verbs may not be nommalued smce rheu obligatory argument would be rendered facultanve. However, there are also transitive verbs With facultative direct objects which do not allow -n noiillJlahzaoon, such as • vmuhu 'under stander' or •Schm«ku 'taster'. 9 This compound has been used to refer to a German pollocal party that broke its coalmon pronuses m order to remain m government. Io See Zaenen (I993) for a smular cla�m with respect to Impersonal pas51ve formaoon m Dutch. I I Smce reflexive verbs wluch fulfil the semannc cntenon proposed for snn (see below) also take the aux1hary habrn m German, it could be necessary to assume that even the default habrn has selecoonal properoes wluch mclude the subcategon zaoon of all syntarocally rransmve verbs, rhus includmg reflexives. Note that French reflexives, m contrast, follow the sernanoc cntenon for snn.
8
426 0- and D-Pred!cates BieiWISCh & E. Lang (eds� Dim�ISional Adjtctivt$; Grammatical Structurt and Con uptual lntnpraation, Spnnger-Verlag. Berlin-Heidelberg-New York, 26 3-4 1 7. Uvin, B. & Rappaport, M (1988), 'Non-event er-nominals: a probe into argument sttuc rure', Lrngurstics, �. 1007-983. LeVIn, B. & Rappaport Hovav, M (I 994), 'A prelirrunary analyru of causaove verbs m EngliSh', Lingua , 9:1, 3 5-77leVIn, B. & Rappaport Hovav, M (1995), UnaccusatJVIty: At � Syntax-Ltxlcal &man tics Intnjact, MIT Press, Cambndge, MA London. LOhner, S ( I988), 'Ansatze zu emer mtegralen semanrischen Theone von Tempus, Aspekt und Aknonsart,' m V. Ehnch & H. Vater (eds� Ttmporalstmantrk. &itrtigt zur Linguistik d� :U.trifmnz, Niemeyer, Tiibingen, I63-91. LOhner, S. (I990), Wahr ntbtn Falsch· Dual( Optratortn als di( Quantortn natiirlich� Sprac�. Niemeyer, Tiibmgen. Mtthun, M (1 991), 'Acnve/agenove case marking and Its moovarion', Langua�, 67, 5 I D-46. Ortmann, A. (I 994� 'Possesrso kongruenz: Eine Fallsrudie zum Verhilm1s von Semanok, Morphologie und Synrax', MA thesis, H.-Heme-Universicit Diisseldor£ Perlmutter, D. (1978� 'Impersonal pass1ves and the unaccusaove hypothesiS', m J. Jaeger tt a/. (eds� Proc. of tht Frrst Annual Mttting of tht &rktlq Linguistrc Socrtty, Berkeley, Uruvermy of Cahfonua, 1 5 7-89 Pusrejovsky, J. (199I� 'The synrax of event structure', Cognition, 4. 47-8 1. Rosen, C. (I984� 'The mterface between semantic roles and mmal grammaocal relaooru', m D. Perlmutter & C. Rosen (eds� Studia rn Rtlational Grammar, Chicago Univemry Press, Chicago, 38-77Stmpson,J. (I983), 'Resulracives', in L Levin, M Rappaport & A. Zaenen (eds), Papm in LLxical Functional Grammar, IULC, Bloom mgron, 143-57. Soebels, B. (1994 'lexlkalische Argumente und Adjunkte: Zum semannschen Beitrag von verhalen Prafixen und Parrikeln',
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
Conctptual Structurt to Syntax, Foru, Dordrecht. Dowty, D. (1979� Word MtaningandMontagu( Grammar, Rride� Dordrecht. Dowty, D. (199I), Thematic proto-roles and argument selecoon', Langua�. 67, 5476I9. Grewendorf, G. (1989� Ergativtty rn Gmnan, Foru, Dordrecht. Higginbotham,]. (1985� 'On Semantics', Lin guistic Inquiry, 16, 547-93· Hoekstra, T. (I988), 'Small clause results', Lingua , 74. 101-39. Hour, A. van, Randall, J� & Weillenborn, J. (I99J), 'Acquinng the unaccusative unergarive di.!rincnon', in M Verrips & F. W1jnen (eds� "I'M ACJ[ursition of Dutch, Amsterdam Senes of Child Language Development 1. Joppen, S. & Wunderhch, D (I994), 'Argu ment lmkmg m Basque' (to appear m Lingua). Kaufmann, I. (1993a� 'Semannc and con ceptual aspects of the prepostnon durch ', m C. Zdinksy-Wihhelt (ed.� Ik &mantics of Prqmitions: From Mtntal Proussing to Natural Langua� Procming, de Gruyter, Berlin. Kaufmann, I. (I99�). 'What IS an Impossible verb?' (to appear m Folia Linguistica). Kaufmann, I. (I994b� 'Agew- und Thema prad!kate und der Aufbau der Seman Form', d(S Arbdttn eischen Sondnforschungsbmuhs :zS:z, 'lbeone des Lexikons', Nr. 58. Kaufmann, L (I995� Korzuptudl( Grundlagtn smtantisckr Dtkompositionsstrukturtn: Di( Kombinatorik lokol� Vabtn und pnidikatiwr Kompkmmu, Niemeyer, Tiibmgen. K.iparsky, P. (1989), 'Agreement and hoking theory', MS., Stanford. Kiparsky, P. (I992� 'Sttucrural case', MS� Stanford. Kntzer, A. ( I 988� 'Stage-level and mdividual level predicates', m M Krifka (eel.� Gtnm dty in Natural Langua�, Proutdmgs of t� 1988 Tiibingtn Conf�t. Lang, E. (1989� 'lbe semantics of dimen Sional de:ngnarion of spacial objects', m M
Ingrid Kaufmann 427 Ph.D. dissertanon, H.-Heme-Univemcit Diisseldor£ Valin, R D. van (1990), 'Semantic parameters of split intranSIUVIty', Langugt, 66, 22 1-00. Wunderlich, D. (1985), '0ber die Argumente des Verbs', Lmgurst!SCht Bmchu, 91, 1 8 3227 Wunderlich, D. (1987), 'Parnzipien 1m
Deurschen', Lmguistisck &richu, II 1, 345--&i. Wunderhch, D. (1 994), 'CAUSE and the structUre of verbs', MS, Diisseldor£ Zaenen, A. (1993), 'Unaccusativiry m Dutch', m J. Pustejovsky (ed.), &mantic.s and tht Larkon, Kluwer, Dordrecht-Boston London.
Downloaded from jos.oxfordjournals.org by guest on January 1, 2011
