On the concept of diathesis alternations as semantic oppositions 
Ana Fem~indez Ana Fernandez@uab es 
M a Ant6ma Martf amartl@hngua fil ub es 
Gl6na Vftzquez gvazquez@dal udl es 
Irene Castell6n castel@hngua fil ub es 
(U Aut6noma de Barcelona) 
(U de Barcelona) 
(U de Llelda) 
(U de Barcelona) 
Abstract 
In this article we present our conception of 
dtathesls alternations and how they 
intervene m the definmon of a model of 
lexlcal entnes We consider that dmthesls 
alternations are the syntactic realizations of 
opposmons of a more general semantic 
nature We will see how they interact with 
other components such as event structure 
and how different semannc classes of 
pre&cates at~se from that interaction 
1 Related Works 
The work of Levm (1993) presents a 
classification of the alternations in which the 
Enghsh verbs parttclpate This author presents 8 
groups in which she differentiates several 
subgroups The first three include the greater 
number of structures and seem to follow 
generahzatlon criteria The other alternations are 
classified in a more random fashion since either 
very specific groups of alternauons are proposed 
or non-semantically related alternations are 
gtouped together 
Other authols have made exphctt the 
subcategonzanon frames m which verbs can 
participate without using pmr assocmtmn In 
these cases a hst of the structures In which a 
verb pamclpates is presented Gtoss's (1975) 
and Samt-Dtzler's (1996) work Is an example of 
such a methodology applied to French In it, one 
of the structures in each class is granted priority 
From our point of v~ew, it ts of interest to 
dehm~t a class according to the partlclpatmn of 
~ts members m a given structure provided that 
this constructmn illustrates some semantic 
characteristics shared by the verbs 
W~th th~s ~dea m mind, we also point to the 
work of Wdlems (1981) This author considers 
that the members of a semannc class do not 
necessarily share the same syntactm 
charactenmcs and that factors such as the 
degree of concretmn of the verb and the 
morphosyntactlc composmon must be taken into 
account It ~s of considerable Interest to 
contemplate th~s type of phenomena m order to 
overcome the obstacle found with classifications 
based on the number and type of arguments 
The authors who focus on the syntacnc- 
semantic structures, such as Dev~s M~irquez 
(1993), consider that each one of the &fferent 
syntactic structures that a verb takes has a 
different meaning and that a semantic opposmon 
Is thereby estabhshed among the &ffeient 
frames 
• With regard to the representation of 
information, NLP oriented formalisms usually 
Include reformation about the subcategonzatmn 
reqmred by the verb It can be shown by 
declaring the list of structures m which the verb 
participates (Sager 1981), or these structures can 
also be generated from one frame as is the case 
of the LFG (Kaplan & Bresnan 1982), GPSG 
(Gazdar et al 1985), and HPSG (Pollard & Sag 
1987) 
We propose that given a type hierarchy of 
verb entries, the rules that account for syntactic 
behavior regarding alternations have to be 
associated to the representative type of a verb 
group m order to express interesting hngmst~c 
generahzatlons In these types, the participation 
of a set of verbs m a particular alternation and 
the mechanism required to express ~t must be 
spemhed Works carried out along these lines 
are those of Sanfillppo (1990) and Taul6 (1995) 
w~thm the Acqmlex project 
2 Model of lexicai entry 
Our initial hypothesis ~s that the syntax and 
the semantics of lexlcal ~tems are Interrelated 
(Levm & Pinker 1991, Levm 1993, Levm & 
Rappaport 1995) These authors consider that 
2:3 
verbs can be semantically classified based on the 
meaning they share The hypothesis is that the 
verbs of a semant\[c class will share the same 
syntactic behavior Hence, each semantic class is 
assocmted w~th the constructions m which the 
verbs of that group participate 
In our approach, we consider that relevant 
semantic information can be deduced from the 
syntactic behavior Thus, our semantic analysis 
includes a syntactic study of the 
subcategonzation frames m which different 
verbs can be found For this reason, m contrast 
to the above mentioned authors, we do not Infer 
syntactic behavior from the semantic 
characterization but rather it is syntax that helps 
us to complete this semantic description We 
also conslder that this relation can be 
formahzed, and that it is essential for the 
charactenzatlon of the entries 
The three elements around which the 
information that makes up the verbal lex~cal 
entry is organized are meaning components, 
event structure and diathesis alternations W~th 
regard to meanmg components, we draw 
principally from Talmy (1985) Accordmg to 
this author, these components play a central role 
in defining verbal semantic classes 
In our approach we have defined a small 
group of semantic components that we have 
organized into several levels The levels range 
from the mole general to the more specific The 
hrst level is common to all the predicates and 
accounts for Entity The second level serves us 
to distinguish events from states 0nly the first 
ones can have an Imttator At a third level we 
specify those components relevant for grouping 
verbs into semantic classes Change, Attitude, 
Tranaference, etc The components considered at 
a more specific level allow us to characterize 
pledicates, but not to define new classes 
hzstrument, etc From the standpoint of 
leahzatlon, they can be expressed ,,a the le-_cal 
item' (e g La pared se desplom6 / The wall 
crashed) or else syntagmattcally (eg El 
plfnclpe se transform6 en rana / The prince 
turned mto a frog) 
As concerns event structure, we follow 
Parsons' (1990) and Pustelovsky's (1995) 
\] By IlledOS of \[e'~tcdhz,ltlol't Int.orporatloN or e\[qe It t.:ln be 
undel stood 
works These authors consider that an event can 
be decomposed into a subatomic structure m 
whxch the temporal relatJons established 
between the subevents and the partlclpants are 
descnbed In this sense, several patterns of 
eventual behavior have been established They 
mteract with the other elements that constitute 
our model 
Finally, our model ptesents mfotmat~on 
about diathesis alternations This is the ~ublect 
of this paper and acts as the basis for the vetb 
classlficat~on we present here We have 
formalized this reformation in the Plrdptdes 
Lextcal Knowledge Base (PLKB) m the form ot 
modules In It, the types corresponding to each 
one of the modules that form the entry have been 
made explicit Also, the dmthesls alternations 
have been dealt with as lex~cal rules An extra 
module, FORLOG, has been defined to account 
for the relation between semantics and syntax, 
connecting all the mformatton to be found 
3 Concept of Diathesis Alternations 
Our starting point for the study of 
alternations is Levm's (1993) work Unlike this 
author, we think that it Is important to take mto 
account only those very general alternations 
(middle, causative-mchoatlve ), that explain 
relevant syntactic behavior and that leally 
highlight the relation between syntax and 
semantxcs Those constructxons that are very 
specific and m which veiy few velbs pattlclpate, 
such as Obhgarotw Adverb (8 5), have been left 
aside 
In our proposal we understand dtatheses as 
one of the syntagmatlc expressions of a semantic 
opposmon Dlathesis alternations are thus pairs 
of structures (or diatheses) related to each other 
by one of these oppositions 
With this concept m mind we have 
considered the existence of three possible 
oppositions depending on whether there is a 
change of focus m the participants (Change of 
focus and Underspecificatlon) or there Is a 
change m the event structure (Aspectual 
Opposition) For example, the sentences 
(1) a Elena cerr6 la puerta (Elena closed the dool ) 
b La puerta se cerr6 (The door closed) 
are related by a change of focus opposition 
whereas in (a) the cause that provokes the event 
23 
is expressed, m (b) the change undertaken by the 
entity is focahzed On the other hand, sentences 
such as 
(2) a Juan coml6 pescado (Juan ate fish) 
b Juan coral6 (Juan ate) 
are related by means of an underspeclficat~on 
opposition of (b) with respect to (a) Lastly, the 
aspectual opposmon ~s illustrated in the 
following examples in which an event (a) is 
related w~th a state (b) 
(3) a Sara pinta un retrato (S as painting a portraat) 
b Sara pinta muy bten (Sara paints very well) 
We start from the hypothes~s that these 
oppositions are general and mterhngulstlc in 
nature and that, therefore, the corresponding 
syntagmatlc realizations m each language have 
to be defined It ~s thus possible to estabhsh 
translation relations between the languages at a 
semantic level and for each meaning opposmon 
it wdl thereby be feasible to predict the syntactic 
structures that can express it (Fermindez and 
Martf 98) 
For example, Basque incorporates the cause 
by means of a morphological process as can be 
seen m the examples below 
(4) a Kanpalak.lo du (The bell rang) 
b Apalzak kanpalarl joarazl &o (The priest 
tang the bell) 
Whereas in other languages, such as 
Spamsh and Catalan, we need a causative 
auxdmry verb to express the same mformat~on 
(5) a La campana son6 (Sp) / 
La campana va sonar (Cat) 
b El cura htzo sonar la campana (Sp) / 
El capellh vafet sonar la campana (Cat) 
Conversely, m Enghsh nelther the 
morphological process nor the auxiliary verb ale 
needed 
Dmthes\]s alternations m our approach can 
be expressed m three ways in the same lex~cal 
atem (cut, break), by means of morphological 
processes (aburru/aburttr~e) Ol by composition 
(bore/get bored) In the following section, we 
present our typology of opposmons m greater 
detad They are grouped mto two classes In the 
first one, subsectlon 4 1, the opposition ~s 
between two event~ve structures whde m the 
second, subsection 42, the opposition is 
between an event and a state 
4 Typology of Diathesis Alternations 
4.1 Opposition event-event 
In this opposition, there ~s an alternation 
between two eventlve constructions We have 
defined two main types change of focus and 
underspeclficatlon In the former, we include 
those alternations m which a ddfetent 
participant becomes the topic of the sentence, 
thus occupying the subject position In the latter, 
we group those constructaons an which an 
element ~s omitted but the topic of the sentence 
remains the same 
4 1.1 Change of focus 
This opposmon lmphes a change of 
perspective m the subcategonzed elements -' The 
first subtype (a) is an alternation between the 
expression or the non-expression of the cause 
The second (b) is an opposmon m which there Is 
an interchange of arguments at a syntactic level 
without any loss of semantic reformation 
Finally, in the third subgroup (c) there is a 
change in the focalization of the components of 
a single argument which is semantically 
complex 
a) Cause one of the alternating structures ~s 
causative, \[e the cause ~s expressed m the 
constituent that occupies the subject posmon In 
the other structure, the ant~causat~ve, the change 
undertaken by the entity ~s focahzed and zt 
moves to the subject position, the cause ~ 
usually left unexpressed In th~s type of 
opposmon, a s~mdar change occurs m the 
focahzat~on of the event head m the causatwe 
structure the process ~s focahzed and ~n the 
ant~causatwe it ~s the resulting state 
As we have saxd, we consider that thas 
mfolmatlon can be reahzed m several ways In 
the following sentences we exemplify some of 
the possible reahzatlons 
NPI V NP2 
The heat has melted the ice 
NP2 V (PP\[prep+NP1\]) 
The ice has melted 
2 We =nc.lude the subjcc.t m the ~ub(.ategonzat\[on hst 
24 
NPI aux V NP2 
The news made John happy 
NP2 aux V (PP\[prep+NP1 \]) 
John was happy (with the news) 
b) Inverse these are s~mple event structures 
that express the same event from two different 
perspecuves implying an argument switch They 
revolve two participants that become the focus 
of attention in each alternating structure ~ 
these arguments ts possible when the 
mformanon is not considered relevant for 
communlcanve purposes or It ts provided by 
other elements of the discourse that are of a 
pragmauc or spatto-temporal nature 
NPI V PPI 
The prisoner escaped from prison 
NP1 V 
The prisoner escaped 
NPI V NP2 
The sun radmtes heat 
NP2 V (PP \[prep+NPl\]) 
Heat radiates from the sun 
c) Holistic: with the term hohsuc we refer 
to an opposition between a structure that 
presents an entity as a whole and another 
construction in which th e emphasis Is placed on 
one of its consututwe parts We understand this 
type of metonymlcal relauon In a broad sense, 
including as parts the instruments, the means, 
the properties, the contents, etc 
In this case, the possible combinations of 
the syntactic alternatmns ~s considerable The 
switch can be within the verb phrase or within 
the subject and complement positions There 
might be either a loss of mformauon or only a 
change in the dlstrlbutlon of the participants 
Next we observe some examples 
NPI V (NP2) prep+NP3 
Mary criticizes Lola because of her selfishness 
NP1 V \[NP3 PP\[prep+NP2\]\] 
Mary crmcizes the selfishness of Lola 
NPI pl V pl 
Ana and Ester met 
NPI' V PP\[prep+NP 1"\] 
Ana met w~th Ester 
4 1.2 Underspectficatton 
Cases of argument ehston without any 
switch in the argument position are mcluded 
here (see examples below) The omission of 
"t Parallehsms can be observed with verbs ot the type 
c omplat/vendet (buy/sell) or dalhectbtt (give/receive) m which a 
third pamc~pant is m~phed In such cases a different lex~cal Item ts 
required lor the mve~smn 
The four altemanons seen so far (4 1 la-b-c, 
4 1 2) respond to a different vision of the action 
In the next table, we show the different aspects 
of the sentence that can be altered dlffeient 
focus, different number of arguments 
subcategonzed (SUB), and Information loss In 
the first place, we can modify the information 
involving the topic of the sentence when we 
apply one of the first three opposmons 
Secondly, there can also exist a change m the 
number of elements contained in the 
subcategonzatlon frame And finally, and 
somehow related to the previous point, m some 
opposmons some reformation (meaning 
components) can be lost 
Cause Inverse Hohsttc ,Unders 
Dlf focus + + + 
Dlf SUB +/- - + + 
Inf loss +/- - + 
4.2 Aspectual Opposition 
This opposmon lmphes the alternation of an 
eventive predicate with a statwe one An 
eventlve predicate can become statwe when the 
tense is not maiked, m which case Kt IS also 
usUally accompanied by a modiflel In th~s type 
of alternation the switch of arguments ~s 
possible In the following examples (6) thele is 
no switch and we are predicating about a 
property of the first argument 
(6) a Maria danced the tango (event) 
b Marfa dances the tango (very well) (state) 
It seems-that all the events can be 
transformed into a state when the conditions 
previously mentioned are fulfilled 
The constluction known as middle is 
included in this group too Th~s structure Is 
pamc~pated by those tiansmve predicates that 
25 
allow argument sw~tch and that reqmre an 
adverbial modifier to express the manner m 
which a property of the enmy IS being affected 
(7) a My mother cut the meat (event) 
b This meat cuts easdy (state) 
Lastly, the construction known as adjecttval 
passive estar+partlc~ple combines with a 
transitive construction to express the same sort 
of opposition 
(8) a El ratio ha roto el juguete / The child has 
broken the toy (event) 
b El juguete est~i roto / The toy is broken 
(state) 
In short, our proposal for dlathesls 
alternations is based on estabhshmg 
generahzat~ons that allow us to characterize 
groups of syntactic constructions that provide 
semantic criteria for verb analysis We have 
presented m this section three semantic 
alternations that we have considered so far 
change of focus (cause, mverse and hohstlc), 
underspecificat~on and aspectual opposition 
This proposal has allowed us to classify verbs 
into semantic classes according to the semantic 
opposition m which they pamclpate Also, they 
can be further subclasslfied according to criteria 
regarding event structure and subcategorlzatlon 
In the next section we present the result of such 
grouping 
5 Resulting classes 
We have grouped about 4 000 velbs (1 350 
in each language approximately) 4 This grouping 
has been made according to the shared 
characteristics which define the behavior of the 
predicates The classes defined so far are 
Predicates of Change (800 predicates), 
Predicates of Attitude (200 predicates) and 
Predicates of Transference (350 predicates) Not 
all the verbs behave homogenously within each 
group This Is the reason why we have further 
subdivided them according to protobehavlors 
(see appendix) 
Thin approach shows a more general and 
consistent classification More general because 
4 We have available an LDB containing the translation 
equivalences m Catalan Spanish and Engh,,h for all these veibs 
hnkmg them to Levm s semantic classes and to the new classes we 
propose in this paper 
we have grouped 1 350 verbs belonging to 20 
Levm classes into just three classes, more 
consistent m the sense that we have apphed the. 
same criteria in defining all of them As can be 
seen in table 1 (see Appendix), classes share the 
same meaning components, two of which, the 
lmtlator and the entity, are also shared by the 
members of the three groups since all of them 
denote an event Furthermore, each class can 
present the same event structure and the same 
basic dmthesas altemauons as well 
As we have mentioned, these groups are 
subdwlded according to more specific syntactic 
and/or semantic criteria In the tables m 
Appendix we prowde for each class the 
distractive features, the subclasses with their 
defining elements, and the actual predicates to 
dlustrate them 
In the case of the verbs of change, two basic 
types are observed (change 1 and change 2) 
depending on whether the expression ot the 
component change ts made lexlcally or it ~s 
expressed as a different constituent (see table 2) 
In the case of verbs that express an atmude 
or feeling (see table 3), two groups are also 
observed (amtudel-2 and attitude 3) if we take 
into account the subcategorlzatlon and the kind 
of reahzatlon of the meanmg component specific 
to the class If we consider event structure the 
first group sphts into two (attitude 1 and attitude 
2) 
The type of event expressed by the verbs 
that we ale dealing with can also be simple or 
complex The verbs of atmude that have a 
complex event structure express two processes 
We thus differentiate between verbs that express 
a simple mental process (de~ear-dea,re) flora 
those that also denote another action (complex 
events) Thls other action can be of a 
commumcatwe type (c~tttcar-crtttctze) or of a 
different kind (reir de-laugh at) 
Finally, when dealing with vetbs of 
transference we find that their subcategonzat~on 
presents a wide range of poss~bdltltes It 
depends on whether the verb can express 
syntactically the points of the trajectory (tr de 
tin sttto a otro-go from one place to another) or 
only one of them (ponet una cosa en un ~ttto 
*(deade otro) - put somethmg m a place (flora 
another)), or ~f it can express a transference 
carried out autonomously by the entity (correr- 
26 
run) or it cannot (dar-gtve) The combmanon of 
these factors g~ve rise to four subgroups that we 
state m table 4 
6 Implementation 
The generalizations obtained at a theoretlcal 
level, both for the classes as well as for the 
subgroups defined according to the observed 
behavmrs, have been reflected m an LKB To 
that end, we have designed a type h~erarchy to 
take into account the umverse that we want to 
describe 
Each verb is ascribed to a type that contains 
mformatmn shared by the group This allows the 
task of entering data to be economized w~thout 
any loss of mformanon and takes into account 
the characteristics shared by sets of verbs 
The hierarchy contains two basic types 
uttles and entry Uttles subsumes all the objects 
that are used for the descnptlon of the basic 
elements m the entry meaning components, 
dmthesxs alternatmns, and event structure In 
each case, the basic elements are declared the 
three types of dmthests altematmns, the hst of 
components and the types of events (the 
temporal relation estabhshed between them and 
the head) 
Entry includes two subtypes to account for 
the simple and complex structures declared and, 
for each case, subtypes are defined for the verbal 
classes (entry-type-change 1, entry-type- 
change2 ) In these types, the syntact~c- 
semantic mformauon shared is declared Thus, 
for instance, the veib sorprender (to surprise) 
that belongs to the class of verbs of change 1 is 
assigned the type entry-t3,pe-changel 
sorprender 
entry-type-change 1 
<morf>="sorprend-/M3" 
<syntax subcat compl semref>=all 
<syntax subcat restcomp compl semref>=amm 
Ftgure 1 
As can be seen in figure 1, at the lexlcal 
level only idiosyncratic mformatmn is specified, 
l e morphology and selectlonal restrictions, 
whereas the data about meaning components, 
event structure and alternations ~s obtained 
through mechanisms of inheritance from the 
assigned type 
In the specification of the alternations (see 
figure 2), an attribute is included for each one 
(DIATVAL) It serves to indicate whether the 
verb pamc~pates in the constructmn and when ~t 
does it expresses the procedure reqmred by the 
verb to convey the meaning (lex~cal, syntacuc or 
morphological) From this reformation a lexlcal 
rule can be apphed and a new entry generated 
We can dlustrate this with the anttcausattve rule 
(change of focus) for the verbs of change such as 
romper (break) 
DIAT 
FOC-CHANGE 
ANTICAUSATIVE 
DIATVAL dlatrue 
D-LEX string 
D-SINT string 
D-MORF true 
Ftgure 2 
The activated rule generates a new entry, 
romperse, modlfymg some data of the input 
entry The following table examphfies the 
information that is altered 
Entry 
Orthography 
MC Realization 
INPUT OUTPUT 
complex 
simple 
2 syntag, 
1 mcolp 
2 
complex 
complex 
I syntag, 
lmc, lund 
1 
yes 
SUB 
Event Str head e I head e2 
Acc Altern rio 
The inclusion of all this mformatlon in LKB 
has permitted us to check the adequacy of the 
theoretical framework and to evaluate the 
possibllmes of formalizing it 
Conclusions 
The model of lexlcal entry provides exphclt 
criteria for the analysts of the predicates The 
appllcatmn of these criteria allows us to make 
manifest relevant generalizations about verbal 
behavior regardmg event structure, meaning 
components, and d~athesls alternatmns 
In this paper we have proposed a typology 
of alternations accordmg to the semannc 
27 
opposmon they denote Th~s approach allows us 
to deal with the multdmgual transfer problem 
and to account for a series of verb centered 
mismatches Spamsh, Catalan and Enghsh Show 
different mechamsms for the expression of the 
meaning conveyed by the alternations 
The study of alternations, together w~th the 
meaning components and event structure, has 
permitted us to group the predicates analyzed up 
to the present m three classes Change, Attitude 
and Transference Each one of them ~s 
exemphfied and they present dwerse degrees of 
homogeneity Th~s fact has taken us to the 
definmon of subclasses according to their 
behavior w~th respect to one or more of the 
before mentioned elements In the PLKB, 
mechamsms have been apphed for the 
inheritance of these behaviors m a general and 
economical manner 
Currently, we are working on transferring 
the data contained m the LKB into an LDB smce 
~t allows eas~er and faster access to the 
mformat~on The final a~m ts the integration of 
several resources available onto a platform that 
incorporates and relates different components 
for the analys~s of textual sources 
28 
Appendix 
Table I General chatactertsncs of the predicates 
CLASS CHANGE ATTITUDE TRANSFERENCE 
Meanmg components Ent~ty/Inmator 
Amtude Transterence 
Event Sttucture 
Semantic opposition 
Number 
Change 
e I event 
i e2 resulting state 
iel*<e2* 
Antic (oh focus) 
Aspectual opposmon 1 
8OO 
e I event 
Hohstm (ch focus) 
Passive (ch locus) 
!2oo 
e 1 event 
e2 state 
el* REL e2* 
Undeispectficatmn 
350 
Table 2 Predicates of change 
Subclms 
Subcate g o rtzatton 
Change 1 
NP V NP 
Change 2 
NP V NP PP 
MC Reahzatton lexlcal/mcorp syntactxc 
Example romper/to break convertlr/to convert 
Nvmber 796 4 
Table 3 Predicates of Att, tude 
Subclass Attitude 1 Attitude 2 Attitude 3 
Subcategortzatton NP V NP NP V NP AP 
MC Reahzatton lexlcal/mcorp syntactic 
ES el event el event el event 
e2 event 
e I *=e2 * 
Example adm~rat/to admne ct recto/to cHt~clze constdetm/to cons~de~ 
Number 146 79 16 
Table 4 Predicates of tJamference 
Subclass Transference 1 TI ansterence 2 
Subcat NP V PPI PP2 NPI V NP2 PPI PP2 
MC Reahzatton syntactlc(lncorporatlon) 
Number 60 40 
Example it/to go transportar/to transport 
S, tbclas s TI ans~erence 3 Trans fel ence 4 
Subcat NP V PP NPI V NP2 PP 
MC Reahzatmn syntactic 
Number 35 212 
Example llegar/to arrive poner/to put 

References 
Devls M~ilquez, P (1993) Esquemas ~mtdcttco- 
semdnttcos, Pubhcaclones de la Umversldad de 
C~idlz 

Fernandez Montraveta, A and M A Marti Antonfn 
(1998) "Hacm una propuesta de desajustes 
verbales", AESLA 1998, La Rloja 

Gazdal , G, E Klein, G Pullum, I Sag (1985) 
Genemhzed Phrase-Structure Grammar, 
Blackwell, Oxford 

Gross, M (1975) Mdthodea en Syntaxe, Hermann, 
Pans 

Kaplan, R& J Bresnan (1982)Letteal-Functtonal 
Grammar A Formal System for Grammatical 
Representation, m "The Mental Representation ot 
Grammatical Relations", The MIT Press, 
Carnbndge 

Levm, B (1993) Enghsh Verb Classes and 
Alternations, University of Chicago Press 

Levm, B I S Pinker (eds) (1991) Le;~tcal and 
Conceptual Semantics, Blackwell Publish, 
Cambridge 

Levm, B t M Rappaport (1995) Unnacusamw~ at 
the S)ntax-Lexlcal Semantics Level, MIT Pless 
Cambridge 

Parsons, T (1990) Events m the Semantics of Enghah, 
MIT Press 

Pollard, C & I Sag (1987) Head-Driven Phla~e 
Structure Grammar, Umversxty of Chicago Press 

Pustejovsky, J (1995) The Generative Lexicon, MIT 
Press . 

Sanfihppo (1990) Grammatical relations, Thematic 
Roles and Verb Semanttca, Ph Dissertation, 
Umverslty ot Edlmburgh 

Sant-Dizler, P (1996) Constnlctmg Verb Semantic 
Classe~ for French Methods and Evaluation, 
COLING'96 

Sager, N (1981) Natural Language Ptoce~.~mg, 
Addison-Wesley, Reading Mass 

Talmy, L (1985) Lexlcahzatlon Patterm, m Shopen 
(ed), "Language Typology and Syntactic 
Descnptzon", Cambridge Umverslty Press 

Tauld, M (1995) Representactdn verbal en una Ba~e 
de Conocmuento Lgxtco, Ph D, Barcelona 

Wdlems, D (1981) S~nta~e, lertque et s~matmque, 
Gent Uxtgeveng Umversa 
