From: AAAI Technical Report SS-95-01. Compilation copyright © 1995, AAAI (www.aaai.org). All rights reserved.
"Polymorphic"Causatives:
ComplexPredicates in French
Annie Zaenen
RankXerox Research Centre, Meylan,France
annie,
zaenen@xerox,
fr
Mary DahTmple
Xerox PARC,Palo Alto, CAUSA
dalrymple@parc,
xerox,
com
January 17, 1995
1
Introduction
The French causative construction is a complexpredicate construction, in which multiple predicative elements combineto
forma single syntactic predicate. In example(1), the two verbs
faire "cause" and venir ’come’ combineto give rise to a syntactically monoclansalstructure with a complexsemantic represe.ntation:
(2) Pierre a laiss6 Paul venir
let
come
’Pierre let Paul come."
Frenchpermissives maybe syntactically biclausal, in contrast
with the normally monoclausalcausative construction. I The fstructure for (2) is given in (3); the complement
clause headed
by venir ’come’ bears the XCOMP
relation to the main clause
verblaisser ’let’:
(1) Pierre a fait venir Paul.
(3) ¯ PRED LAISSER
In this case, it is the intransitive verb venir ’come’whichcomSUBJ [PRED PIERRE]
bines with causative fromto form a new two-place syntactic
pre~cate. In fact, causa~ve faire can combinewith a verb of
OBJ [PRED PAUL] -----any arity to forma newsyntactic predicate; the resulting predPRED VENIR"
icate has as manyargumentsas the original verb, with the adXCOMP L SUBJ
dition of a new argumentfor the causer. Wepropose an approach to syntactically-derived complexpredicates whichproducesthe correct syntactic structure and correctly specifies the This stands in contrast to the situation with causatives, which
relation betweenthe syntax and the meaningwithout relying on form complex predicates and are monoclansal.
an ambiguity in the causative verb to permit combinationwith
verbs of varying arities.
2.2 Causatives: Complex predicates
I
The French causative construction is exemplified in (4), with
its f-su-ucturegivenin (5):
2 Thenature of complex predicates
Muchwork has been devoted to the analysis of complexpredicate constructions (Manning1992, Alsina 1993a, Butt 1993,
and works cited therein, amongmanyothers), and in particular to the syntactic features of these constructions. Theanalysis presented here for the Frenchcausative construction closely
resembles the analysis proposed for Urdu complexpredicates
by Dalrympleet al. (1993), employinga resource logic, linear logic, to guide the composition of meanings. The approach gives rise to a clean characterization of completeness
and coherence (Dalrymple, Lamping, and Saraswat, 1993),
and mesheswell with the treatment of other semantic phenomena, such as quantification, modification, and intensional verbs
(Dalrympleet al., 1994a; Dalrympleet al., 1994b).
2.1 Permimives: Syntactically
(4) Pierre a falt venir Paul
causc come
’Pierre caused Paul to come."
(5) "PRED FAIRE(VENIR)
SUBJ [ PRED PIERRE]
OSJ [PRED PAUL] Evidence for the differing status of these constructions
comes from critic climbing. A critic object appears adjacent
to its verb. For biclausal constructions, the clitic object of the
subordinate clause verb appears next to the subordinate verb:
x Dueto lackof space,wewill notdiscussthe monoclausal
permissive
constructionexemplified
in (a):
(a) Pierrea lais~ venirPaul
bidausal
The Frenchpermissive construction is exemplified in (2):
180
(6)Marieveut" [inmnnger]
wants it to eat
’Mariewantsto eat iL’
However,in the case of complexpredicates the clitic object appears adjacent to the complete verb complexof whichit is an
object:
(7) Pierre l’a
falt venir
him-has cause come
’Pierre has caused him to come."
(8) Marie la lui a falt manger
it him has cause eat
"Mariehas causedhim to eat it.’
Muchother evidence has been presented that differentiates
these two constructions and evidences the monoclansality of
the complexpredicate construction; see, for example, Abeille
and Godard(1994) and the references cited there.
3 Semantic composition as deduct/on in linear logic
Dalrymple, Launping, and Saraswat (1993) present an approach to semantic composition within the LFGframework
which makesuse of a fragment oflinearlogic 2 (Girard, 1987)
to state constraints on the compositionof meanings;see also
Dalrymple et al. (1993), Dalrymple et al. (1994a),
Dalrympleet al. (1994b). Webriefly review the assumptions
of this frameworkhere.
A key principle of the approach is to distinguish meanings
from insu’uctions about how to assemble meanings. The language of meaningcan be any suitable logical language for expressing truth-conditional meaningsof words and phrases; for
current purposes, first-order logic will suffice. Thelanguageof
assembling meaningsor glue language is a fragment of linear
logic (the tensor fragment) used to express constraints on the
compositionof meanings. That is, the glue language expresses
how the meanings of words and phrases can combine.
Wewill go through a simple examplebefore turning to the
more complicated case of complex predicates. The relevant
lexical entries for sentence(9) are given in (10):
(9) Marie a mang6 la pomme
ate the apple
(I0) Marie
NP (T PR]~) = MARIE
To ~* Marie
pomme N 0" PRIED) = POMME
To, "-* apple
manger V (T PR~’~)= MANGER
VX, Y. (T, AGENT)-,.. X ® (To THEME)~
--0 L, "* eat(X, Y)
Since we present this simple examplemerelyin order to iLlustrate our formal assumptions, we ignore the syntax and semantics of the other wordsin the sentence,as well as issues of tense,
aspect, definiteness, and so on.
Our analysis requires a means of specifying the meanings
of particular syntactic objects and the relations betweenthese
meanings.
Wetake advantage of the projection architecture of LFG(Kaplan, 1987; Halvorsen and Kaplan, 1988),
which enables a characterization of meaningsassociated with
f-structures. Notationally, the subscript o- represents the projectionfunction from f-structures to semantic representations.
The expression’We’represents the semanticprojection of the fstructure T, for example. Weassumethat semantic projections
are associated via the ~,~ relation with meanings,and so the expression "T¢~.~ Marie" indicates that the meaning Marie is
recorded as the meaningof T~,- Semanticprojections can also
be structured with attributes and values, just like f-structures,
and the values of these attributes can be a&u~atedwith meanings. Thus the expression (T~, AGENT~
~, X means that the
semantic projection T~, of the f-structure T has an attribute
AGENT,and that X is recorded as the meaning of the value
of(To AGENT~.
Thelexical entries in (I0) give rise to the f-structure and associated semantic informationin (I I):
PRED MANGER
1
(II)
f:
I
SUBJ g:[PRED MARIE]
OBJ h: [ PRED POMME
]
ga ~’* Marie
ha ~ apple
vx, Y. (f.AGZVr)
®(f~,THEME)~.+
--0 fo ,,.. eat(X, Y)
Lexical items like Marie and pommehave meanings like
Marie and apple. Verbs are more complicated. The verb
mangermakesthe following semantic contribution:
VX, Y. (/¢ AGENT)~
®(/~ THEME)~.
--0 fa ~* eat(X, Y)
This rule states that if the agent of mangerhas meaningX, and
(®) the theme of manger has meaning Y, then ( --0 ) those
meanings are consumed, and a meaning for the f-structure
for the sentence, f, is produced. For further details, see
Dalrymple, Lamping, and Saraswat (1993).
Note that the verb manger’eat’ does not subcategorize for
grammaticalfunctions such as SUBJand OBJ, as is assumedin
classical LFGanalyses. Instead, it specifies semantic/thematic
information about its arguments. The grammatical functions
that express these arguments are given by mapping rules,
which specify a relation betweenarrays of thematic roles and
the grammaticalfunctions that realize them. For the example
at hand, the followingmappingrule is relevant:
aSeeScedrov
(1993)for a tutorial inmxlu~on
to linear logic; seea/so (12) [(Vf, X, Y. ((f SUBJ). ,..* X) ® ((f OBJ).
Saraswat
(1993).
--o (f, AGENT)~.. X ® (fa THEME)~..
181
This mappingrule states a general relation betweensubjects (15)[(V.f,
X,Y.((f
OBI.~etnt)
X)
,-,~
and agents, and objects and themes: given a subject X and an
®((f
SUBJ).
-,~
object Y, an agent X and a theme Y can be posited.
-o(/.AOrta3
--X ® (/.THEME)
--.
Themappingrule in (l 2) is a general rule about possible thematic role/grmnmaticalfunction associations for f-structures in This rule relates themesto subjects, and requires the agent to
general. Wecan instantiate this mappingrule for the particular be realized as an oblique phrase. Again, a well-formedderivaf-structure that we are concernedwith, the f-structure labeled tion wouldnot be possible, since the rule allows the derivation
f:
to proceed only in the presence of an OSLagentphrase. In this
and in the abovecase, the interaction of the mappingrules with
(13)
vx, Y. (g.
...
x)® (h.
-~
completenessand coherence constraints ensures that no other
-.o(faAGENT).,~X ® (ft,THEME)
,,.*
derivation produces a well-formedoutput.
Having
donethis,we beginthesemantic
deduction
withthe It is importantto note that weintend no claimsabout the correctness of the specific details of these mappingrules; rather,
following
premises:
our claim is that mappingrules should be of the general form
Marie: go ",~Marie
we have illustrated, specifying possible relations betweenthematic roles and grammaticalfunctions. In particular, no theoapple : ha ",~ apple
retical significance should be attached to the choice of thematic
mappingX
: VX,Y. (g~, -,., X) ® (h,, -,.,
role labels used here; for the verb eat, for example,labels such
-o (f. AG~rI~~ X ® (f~,THEME),,-,
as ’eater’ and ’eaten-thing’ woulddo as well. Mappingrules
eat : YX, Y. (f. AGF3V~.,.+ X ® (f.THEME)Y
mayspecify thematic/semantic information as we have given
--o fa ~’* eat(X, Y)
it, as assertions about roles such as agent andtheme. AlternaFromthese premises, the following deduction is possible, giv- tively, such information can be specified in terms of structmes
of Conceptual
Semantics (Jackendoff, 1990), as in Butt’s aping the desired result:
proach (Butt, 1993), in terms of Proto-Roles (Dowry,1991),
Marie @apple @ mappingl ® eat
in Alsina’s approach (Alsina, 1993a), or by someother means.
Grammaticalfunction information can be specified in terms of
(f~ AGENT) ~ Marie
grammaticalfunction labels such as suBJ and OSJ, as is done
®(ft, THEME)~ apple
here; alternatively, such informationcan be given in terms of a
®eat
feature decompositioncross-classifying grammaticalfunctions
~" f~ ",~ eat(Marie, apple)
(Levin, 1986; Bresnan and Kanerva, 1989; Alsina, 1993a).
Ourclaims relate to the formal properties of mappingrules, and
The use of linear logic instead of classical logic as the thus our approach could be adopted as a part of any one of a
"glue language’ provides clear advantages which we have dis- numberof specific approaches to mappingtheory.
cussed in other work (Dalrymple, Lamping, and Saraswat,
1993;Dairympleet al., 1993): it allows us to capture the intuition that lexical items and phrases contribute uniquelyto the 4 Complex predicates, mapping rules, and deduction
meaningof a sentence, thus enabling a clean semantic definition of the LFGrequirements of completeness and coherence. Weare now ready to consider how our approach applies to
This requirement is what ensures the selection of the correct complexpredicates. Our task is to explain howcausativefaire
mappingrule. In a semantic derivation, the wrong mapping can combinewith a verb of any arity - here, with the intransirule might be incorrectly chosen, for exampleone whichrelates tive verb venir ’come’- to license a syntactically monoclausal
argumentstructure. Wealso want to ensure that the syntactic
experiencers to subjects, and themesto objects:
representationfor this sentenceis correctly related to its (complex)
meaning. Wewill consider the following example:
(14)!(Vf,X.Y.((f SLrBJ)t,
,,~ X)®((f OBJ)t,
-~ (f, ~XPOSaVCER)-~
®(f.
THEME)
.,.*
Y)
(16) Pierre a fait venir Paul
causecome
The use of this mappingrule in the derivation of the sentence
’Pierre caused Paul to come.’
above wouldnot allow a well-formed derivation. First, the
meaningfor a clause containing the verb manger’eat’ can only The first issue we must address is what the semantic structure
be obtained in the presence of an agent, and this mappingrule of this sentence should look like. Semantically, is the French
does not provide one. Second,this rule asserts the presence of causative verb two-place or three-place?
an experiencer which is not consumedduring the derivation;
the result is incoherence,since the premisesof the sentence do (17) a. cause(Pierre, come(Paul))
not lead to a single meaningassociated with the sentence, with
b. cause(Pierre, Paul, come(Paul))
no premises remaining.
Alternatively, a mappingrule such as the following might Our frameworkenables us to produce either meaning unprobhave been incorrectly selected:
lematically; for discussion of the linguistic issues involved in
182
this choice, see Alsina and Joshi (1991). Welack space to ad- This rule maps a SUBJX and an OeJ Y to the thematic roles
~, crucially, this mapping takes place
dress this issue in detail, and will makethe assumptionthat the CAUSER
and AGENT
first altenuttive is the correct one.
whetherthese roles are both associated with a single verb (the
Thef-structure and lexical entries for sentence(I 6) are:
usual case) or with more than one verb (the case of complex
predicates). In other words, we claim that in the case of complex predicates, the SUBJand OBJ(and other grammaticalfunc(18)
tious) of a single f-structure can be associated with thematic
roles of multiple verbs.
The semantic contributions from the lexical entries of the
words in the sentence and the instantiated mappingrule give
the following premises:
(19) Pierre N’P (T PI~T}) = PIERRE
To "~ Pierre
(21)
Pierre: (go"~Pierre)
Paul
venir
NP (T PRED) = PAUL
To "* Paul
V (T
Paul : (ho ",~ Paul)
mapping2 : (VX, 3,,, Z. (go ~ X) ® (ho
.-o (fo CAUSER)
-~ X®(f~ AC~Z~
-.~ Y)
PRED)=COME
vx. (to AC~rr)
-,* X--o to "-~cone(X)
faire
V
come: (YX. (fo AGENT).,-*X .--o f® "" cone(X))
muse : (VX, P.(f~, CAUSER)~ X @ f® ~
-.-o f® ~ cause(X, P))
vx,P. (to CAUSER)
-..
®(T.
"-*
P)
-~ To~ cauae(X,
P)
Fromthese premises, the deduction proceeds as follows:
The semantic contribution of causativefaire is one of the two
crucial ingredients in our analysis:
vx,P.(to CAUSER)
-.. X®(1". "-* P)
.-o to ~ cause(X,
P)
(22)
Paul ® Pierre
@mapping2 ® come ® cawe
b
(fa CAUSER).,., Pierre ® (fo AGENT).,.+
@come® cause
I-
(/,,
}-
fo ".* cause(Pierre,
CAUSER)’,.*Pierre
®fo ~ cone(Paul) ® cause
come(Paul))
Causafivefa/re requires an argumentwith the role of CAUSER,$ Semantic forms and resource-accounting
the role filled by Pierre in the sentence above. Wereiterate
that the exact choice of thematic role label is not the important Manysyntactic frameworksimplicitly or explicitly assumethat
issue here; rather, our approachrequires merelythat the the- the array of grammaticalfunctions licensed by a syntactic predmarie/semautic information referred to by mappingrules and icate is immutableand cannot be altered in the course of a
lexical entries be expressed clearly and unambiguously.
derivation. Complexpredicates seem to violate this assumpBesides a CAUSER,
causative faire also requires another ar- tion, however;,verbs such as causativefa/re seemto operate on
gument:a preliminary meaningfor the f-structure that it is as- syntactic predicates so as to increase their arity by one.
sociated with. This meaningwill be obtained when the verb
In the context of LFG,certain assumptionshave traditionally
venir is provided with its arguments. Causativefaire requires, been madeabout the nature and function of semantic forms, the
then, that the CAUSER
X and the preliminary meaning P be values of the attribute PRED.Kaplau and Bresnau (1982) asconsumed, and a new meaning produced: cause(X, P). Tiffs sumePREDvalues for verbs of the following form:
preliminary meaningmay be provided by a verb of any anty;
(XsuB (ToBJ)>’
thus, we need only one lexical enntry for causative fa/re, no (23)’eat<
agent
theme
matter what the arity of the verb with whichit combines.
It is here that one of the advantagesof the use of linear logic According to Kaplan and Bresnau (1982), these semantic
for guiding semanticcompositioncan be clearly seen: classical forms encode four types of information:
logics wouldnot allow a step such as this one, where two different assignments of meaningto To are madeon the left and
1. Specification of the semanticrelation
right handsides of the --o linear implication; the resourceori2. Mappingof grammatical functions to semantic roles
entation of linear logic makessuch an analysis possible.
Thesecondcrucial ingredient in our analysis is the following 3. Subcategorization information (the governed grammatimappingrule:
cal functions)
(2o)!(vf, x, Y,z.((fsuBJ).
-.. x) ®((f OSJ).
--o(foCAUSER)
-,.. X®(f. ACEtCr)
.,~
4. Instantiation to indicate semantic distinctness (predicate
uniqueness)
183
Encodingthese kindsof informationby meansof a single forWebelieve that the syntactic function of predicate uniquenull devicepermitsthe syntacticallyrelevant aspectsof mean- ness is an importantone, andso wecontinueto permitsemaning to be confinedto a single placein the f-structure. Butwe tic formsin the lexical entries of semantically
contentfullexical
haveseen that subcategorizafioninformationmustbe allowed itemsto specifythe valuesof PRED
attributes in the f-structure.
to changein the courseof a syntactic derivation, andso some However,wealso believe in a necessaryseparation between
changeis necessaryto allowfor the formationof complex
pred- syntaxand semantics,so that whichparticular semanticform
icates whilestill allowingfor the abovefour kindsof informa- appearsis irrelevant for syntacticpurposes.Thatis, the syntax is concernedonly with the presenceor absenceof a semantion to be properlyencoded.
tic form,andnot with its internal structure. Thus,if the only
Kaplan(personalcommunication)
observesthat the effect
is to ensurepredicate uniqueour approachand the approachtaken in muchother recent LFG remainingfunction of the PReD
literature (Alsina1993a,Butt 1993,among
others) is to u’eat ness, it woulddo as well to assumethat the PRED
value for a
the different kinds of informationencodedby semanticforms sentencewith a complexpredicate is contributedby the main
with separate and independentmechanisms.Specification of verb (for example,VF.NIR),andthat the function of FAIRE
the semanticrelation is accomplishedin the lexicon: a verb tO modifythe argumentstructure but not to contribute to or
like ’eat’, for example,specifies that its meaning
is eat(X, Y) changethe PRED
valueof the construction.It is for this reason
whengiven an agent X and a themeY. Weassumethat the that wedo not take a position on howa semanticformsuch as
mappingof grammaticalfunctions to semanticroles, as well ’FAIRE<VENIR>’
is formed,since webelieve that the par6cas the particular inventoryof governed~’ammadcal
functions, ular syntacticshapeof this predicateis not an issue.
are givenby the mappingrules: the mappingrules mightspecify that for a verblike "eat’, the subjectbearsthe thematicrole 6 Condnsion
of agentandthe object bearsthe role of theme.
Giventhese newmechanisms,complexpredi~t~ and sire- Theuse of linear logic enablesa treatment of complexpredpler constructionsare handledidentically. If a rule exists to icates whichproducesthe correct syntactic structure andcormapa set of grammatical
functionslicensedby a complex
pred- rectly specifies the relation between
the syntaxandthe meaning
icate to a set of thematicroles, then the result will be a mono- withoutrelying on an ambiguityin the causativeverb to perclausal syntacticstructure, whetherthis array of thematicroles mit combinationwith verbs of varying arities. Ourdeduction
arose in associationwith a single lexical item or with multi- fi’amework
enablesus to use linear logic to state suchoperaple predicativeelements.This aspect of the analysis of com- tions in a formallywell-definedmanner.
plex predicates is thus unproblematicfor any approachwhich
assumesthat verbsare lexically associatednot with an array 7 Acknowledgments
of grammatical
functions,but with an array of thematicroles;
mapping
rules will associatethe appropriatearray of grammati-Wewouldlike to thank John Lamping
and Vijay Saraswatfor
cal functionswiththese thematicroles in all cases, for complexvaluable input and comments.
predicatesas wellas for simplercases.
Onthese assumptions,only one function of the semantic
formremains:iustantiation to indicate semanticdistinctness. References
Onthe classical LFGview, different PRBD
values are incomAbeille, Anneand Danielle Godard.1994. Frenchcausatives
patible; this disallowsthe presenceof multiplefillers of a sinin I-IPSG.MS,Universityof Paris VII.
gle syntacticargument
slot. Forexample,critic doublingis disallowedin somelanguagesbecausethe clitic pronounand the Alsina, Alex and Smita Joshi. 1991. Parameters in
full nounphrasecontributeincompatible
PREDValUes,as in the
causative constructions. In Lise M.Dobrin,LynnNichols,
following example(Grimshaw,1982):
and RosaM. Rodriguez,editors, Papersfrom the TwentySeventhRegionalMeetingof the ChicagoLinguistic Society,
(24) a. Je cherche
Pierre
pages31-46, Universityof Chicagu.ChicagoLinguistic SoI amlooking for Pierre
ciety.
’I amlookingfor Pierre.’
Alsina, Alex. 1993a. Predicate Composition:A Theory of
b. Je le cbercbe
Syntactic FunctionAltemations.
Ph.D.thesis, StanfordUniI him amlooking for
versity.
’I amlookingfor him.’
Alsina, Alex. 1993b. A theory of complexpredicates: EvPierre
c.*Je le cherche
idence from causatives in Bantu and Romance.Paper preI him amlooking for Pierre
sented at the Complex
Predicates Workshop,
CSLI,Stanford
University.
Note, however,that in the case of complexpredicates the
samesituation seemsto arise. Eachof the verbsthat makeup Bresnan,Joan and Jonnl M. Kanerva.1989. Locative inverthe complexpredicate constructionwouldbe expectedto consion in Chicbev~a:Acase study of factorization in gramtribute a PRED
with a uniquevalue.
mar. Linguisticlnquiry, 20(1):1-50. Alsoin E. Wehrliand
184
T. Stow¢ll, eds., Syntax and Semantics 26: Syntax and the Levin, Lori S. 1986. Operations on lexical forms: Unaccusative rules in Germaniclanguages. Ph.D. thesis, M1T.
Lexicon. NewYork: AcademicPress.
Butt, Miriam. 1993. The Structure of ComplexPredicmes in
Urdu. Ph.D. thesis, Stanford University.
Manning, Christopher D. 1992. Romance is so complex.
Technical Report CSLI-92-168,Center for the Study of Law
guage and Information, Stanford University.
Dalrymple, Mary, John Lamping, and Vijay Saraswat. 1993.
LFGsemantics via constraints. In Proceedingsof the Sixth Saraswat, Vijay A. 1993. A brief introduction to finear concurMeetingof the EuropeanA CL, University of Utrecht, April.
rent constraint programming.Technical report, XeroxPalo
European Chapter oftheAssociation for ComputationalLinAlto Research Center, April.
gnistics.
Scedrov, Andre. 1993. A brief guide to linear logic. In
Dalrymple" Mary, Angie Hinrichs, John Lamping, and Vijay
G. Rozenberg and A. Salomaa, editors, Current Trends in
Smmwat.1993. The resource logic of complex predicate
Theoretical ComputerScience. WorldScientific Publishing
interpretation. In Proceedingsof the 1993Republic of China
Co. Revised and expandedversion of the article originally
ComputationalLinguistics Conference ( ROCLING
), Hsitou
appearing in Bulletin of the EuropeanAssoc. for Theoretical
National Park, Taiwan, September. Computational LinguisComputerScience 41, 1990.
tics Society of R.O.C.
Dalrymple" Mary, John Lamping, Fernando C. N. Pereira, and
Vijay Saraswat. 1994a. A deductive account of quantification in LFG.In MakotoKanazawa,Christopher J. Pifl6n,
and Hem’iettede Swat~,editors, Quantifiers, Deduction,and
Context. Center for the Study of Languageand Information,
Stanford, California. To appear.
Dalrymple, Mary, John Lamping, FemandoC. N. Pereira, and
Vijay Saraswat. 1994b. Intensional verbs without
typeraising or lexical ambiguity. In Conferenceon InformationOriented Appmachesto Logic, Language
andComputation,
Moraga,California. Saint Mary’sCollege.
Dowry, David R. 1991. Thematic proto roles and argument
selection. Language, 67(3):547-619.
Girard, J.-Y. 1987. Linear logic. Theoretical ComputerScience, 45:1-102.
Grimshaw, Jane. 1982. Subcategorization and grammatical relations. In Annie Zaenen, editor, Subjects and other
subjects. Indiana University Linguistics Club, Bloomington,
IN, pages 35-56.
Halvorsen, Per-Kristian and Ronald M. Kaplan. 1988. Projections and semantic description in Lexical-Functional Grammar. In Proceedings of the International Conference
on Fifth Generation ComputerSystems, pages 1116--1122,
Tokyo,1apart. Institute for NewGeneration Systems.
$ackendoff, Ray S. 1990. Semantic Structures.
Press, Cambridge, MA.
The MIT
Kaplan, Ronald M. 1987. Three seductions of computational
psycholinguistics. In Peter Whitelock, Harold Somers, Paul
Bennett, Rod Johnson, and Mary McGeeWood,editors, L/ngui,~ic Theory and ComputerApplications. AcademicPress,
London, pages 149-188.
Kaplan, Ronald M. and Joan Bresnan. 1982. LexicaiFunctional Grammar:.A formal system for grammatical representation. In Joan Bresnan, editor, The Mental Representation of Granunatical Relations. The MITPress, Cambridge, MA,pages 173-281.
185