Acquiring Predicate-Argument Mapping 
Information from Multilingual Texts 
Chinatsu Aone, Douglas McKee 
Systems Research and Applications (SRA) 
2000 15th Street North 
Arlington, VA 22201 
aonec@sra, corn, mckeed@sra.com 
Abstract 
This paper discusses automatic acquisition of predicate-argument mapping in- formation from multilingual texts. The lexicon of our NLP system abstracts the 
language-dependent portion of predicate-argument mapping information from the 
core meaning of verb senses (i.e. semantic concepts as defined in the knowledge base). We represent this mapping information in terms of cross-linguistically generalized 
mapping types called situation types and word sense-specific idiosyncrasies. This 
representation has enabled us to automatically acquire predicate-argument map- ping information, specifically situation types and idiosyncrasies, for verbs in English, 
Spanish, and Japanese texts. 
1 Introduction 
Lexicons for a natural language processing (NLP) system that perform syntactic and 
semantic analysis require more than purely syntactic (e.g. part-of-speech information) 
and semantic information (e.g. a concept hierarchy). Language understanding requires 
mapping from syntactic structures into conceptual representation (henceforth predicate- 
argument mapping), while language generation requires the inverse mapping. That is, 
grammatical functions in the syntactic structures (e.g. subject, object, etc.) should be 
mapped to thematic roles in the semantic structures (e.g. agent, theme, etc.). 
In this paper, we discuss how we acquire such predicate-argument mapping information 
from multilingual texts automatically (cf. Zernik and Jaeobs work on collecting thematic 
roles \[20\]). As discussed in Aone and Mckee \[1\], the lexicon of our NLP system abstracts 
the language-dependent portion of predicate-argument mapping information from the core 
meaning of verb senses (i.e. semantic concepts as defined in the knowledge base). We 
represent this mapping information in terms of cross-linguistically generalized mapping 
types called situation types and word sense-specific idiosyncrasies. This representation has 
enabled us to automatically acquire predicate-argument mapping information, specifically 
situation types and idiosyncrasies, for verbs in English, Spanish, an.d Japanese texts. 
In the following sections, we first describe how we represent the predicate-mapping 
information. Then, we discuss how we acquire situation type and idiosyncrasy information 
automatically from multilingual texts and show some results. 
2 Predicate-Argument Mapping Representation 
Each lexical sense of a verb in our lexicon encodes its default predicate-argument mapping 
type (i.e. situation type), any word-specific mapping exceptions (i.e. idiosyncrasies), and 
107 
# of required NP or S arguments 
CAUSED-PROCESS 2 
PROCESS-OR-STATE 1 
AGENTIVE-ACTION 1 
INVEKSE-STATE 2 
default thematic roles prohibited thematic roles 
Asent Theme 
Theme Agent 
Agent 
Goal Theme Agent 
Table 1: Definitions of Situation Types 
English Spanish Japanese 
CAUSED-PROCESS kill matar, mirar korosu, miru 
PROCESS-OH-STATE die morir shibousuru 
AGENTIVE-ACTION look bailar odoru 
INVERSE-STATE see vet mieru 
Table 2: Situation Types and Verbs in Three Languages 
its semantic meaning (i.e. semantic concept) in addition to its morphological and syntactic 
information. In the following, we discuss these three levels in detail. 
2.1 Situation Types 
Each of a verb's lexical senses is classified into one of the four default predicate-argument 
mapping types called situation types. As shown in Table 1, situation types of verbs are 
defined by two kinds of information: 1) the number of subcategorized NP or S arguments 
and 2) the types of thematic roles which these arguments should or should not map to. 
Since this kind of information is applicable to verbs of any language, situation types are 
language-independent predicate-argument mapping types. Thus, in any language, a verb 
of type CAUSED-PROCESS has two arguments which map to AGENT and THEME in 
the default case (e.g. "kill"). A verb of type PROCESS-OR-STATE has one argument 
whose thematic role is THEME, and it does not allow AGENT as one of its thematic roles 
(e.g. "die"). An AGENTIVE-ACTION verb also has one argument but the argument maps 
to AGENT (e.g. "look"). Finally, an INVERSE-STATE verb has two arguments which 
map to THEME and GOAL; it does not allow AGENT for its thematic role (e.g. "see"). 
Examples from three languages are shown in Table 2. 
Although verbs in different languages are classified into the same four situation types 
using the same definition, mapping rules which map grammatical functions (i.e. subject, 
object, etc.) in the syntactic structures 1 to thematic roles in the semantic structures may 
differ from one language to another. This is because languages do not necessarily express 
the same thematic roles with the same grammatical functions. This mapping information 
is language-specific (cf. Nirenburg and Levin \[16\]). 
The default mapping rules for the four situation types are shown in Table 3. They are 
nearly identical for the three languages (English, Spanish, and Japanese) we have analyzed 
so far. The only difference is that in Japanese the THEME of an INVERSE-STATE verb 
is expressed by marking the object NP with a particle "-ga", which is usually a subject 
1 We use structures similar to LFG's f-structures. 
108 
CAUSED-PROCESS AGENT 
THEME 
PROCESS-OR-STATE THEME 
AGENTIVE-ACTION AGENT 
INVERSE-STATE GOAL 
THEME 
English/Spanish Mapping Japanese Mapping 
(SURFACE SUBJECT) (SURFACE SUBJECT) (SURFACE 
OBJECT) (SURFACE OBJECT) ~SURFACE SUBJECT) ~SURFACE SUBJECT) 
(SURFACE SUBJECT) ~SURFACE SUBJECT) 
(SURFACE SUBJECT) (SURFACE SUBJECT) 
(SURFACE OBJECT) . (SURFACE OBJECT) (PARTICLE "GA") 
Table 3: Default Mapping Rules for Three Languages 
marker (cf. Kuno \[12\]). 2 3 So we add such information to the INVERSE-STATE mapping 
rule for Japanese. Generalization expressed in situation types has saved us from defining 
semantic mapping rules for each verb sense in each language, and also made it possible to 
acquire them from large corpora automatically. 
This classification system has been partially derived from Vendler and Dowty's as- 
pectual classifications \[19, 9\] and Talmy's lexicalization patterns \[18\]. For example, all 
AGENTIVE-ACTION verbs are so-called activity verbs, and so-called stative verbs fall 
under either INVERSE-STATE (if transitive) or PROCESS-OR-STATE (if intransitive). 
However, the situation types are not for specifying the semantics of aspect, which is ac- 
tually a property of the whole sentence rather than a verb itself (cf. Krifka \[11\], Dorr \[8\], 
Mocns and Steedman \[15\]). For instance, as shown below, the same verb can be classified 
into two different aspectual classes (i.e. activity and accomplishment) depending on the 
types of Object NP's or existence of certain PP's. 
(1) a. Sue drank wine for/*in an hour. 
b. Sue drank a bottle of wine *for/in an hour. 
(2) a. Harry climbed for/*in an hour. 
b. Harry climbed to the top *for/in an hour. 
Situation types are intended to address the issue of cross-linguistic predicate-argument 
mapping generalization, rather than the semantics of aspect. 
2.2 Idiosyncrasies 
Idiosyncrasies slots in the lexicon specify word sense-specific idiosyncratic phenomena 
which cannot be captured by semantic concepts or situation types. In particular, subcat- 
egorized pre/postpositions of verbs are specified here. For example, the fact that "look" 
denotes its TItEME argument by the preposition "at" is captured by specifying idiosyn- 
crasies. Examples of lexical entries with idiosyncrasies in English, Spanish and Japanese 
are shown in Figure 1. As discussed in the next section, we derive this kind of word-specific 
information automatically from corpora. 
2There is a debate over whether the NP with "ga" is a subject or object. However, our approach can 
accommodate either analysis. 
3The GOAL of some INVERSE-STATE verbs in Japanese can be expressed by a "ni" postpositional 
phrase. However, as Kuno \[12\] points out, since this is an idiosyncratic phenomenon, such information 
does not go to the default mapping rule. 
109 
(LOOK (CATEGORY . V) (SENSE-NAME. LOOK-l) 
(SEMANTIC-CONCEPT #LOOK#) (IDIOSYNCRASIES (THEME (MAPPING (LITERAL "AT")))) 
(SITUATION-TYPE AGENTIVE-ACTION)) 
(INFECTAR (CATEGORY . V) (SENSE-NAME. INFECTAFt- 1) 
(SEMANTIC-CONCEPT #INFECT#) (IDIOSYNCRASIES (THEME (MAPPING (LITERAL "CON" "DE"))) 
(GOAL (MAPPING (SURFACE OBJECT)))) (SITUATION-TYPE CAUSED-PROCESS)) 
(NARU (CATEGORY . V) 
(SENSE-NAME. NARU- I) (SEMANTIC-CONCEPT #BECOME#) 
(IDIOSYNCRASIES (GOAL (MAPPING (LITERAL "TO" "NI")))) 
(SITUATION-TYPE PROCESS-OR-STATE)) 
Figure 1: Lexical entries for "look", "infectar", and "naru" 
2.3 Semantic Concepts 
Each lexical meaning of a verb is represented by a semantic concept (or frame) in our 
language-independent knowledge base, which is similar to the one described in Onyshkevych 
and Nirenburg \[17\]. Each verb frame has thematic role slots, which have two facets, 
TYPE and MAPPING. A TYPE facet value of a given slot provides a constraint on the 
type of objects which can be the value of the slot. In the MAPPING facets, we have 
encoded some cross-linguistically general predicate-argument mapping information. For 
example, we have defined that all the subclasses of #COMMUNICATION-EVENTS (e.g. 
#REPORT#, #CONFIRM#, etc.) map their sentential complements (SENT-COMP) 
to THEME, as shown below. 
(#COMMUNICATION-EVENT# (AKO #DYNAMIC-SITUATION#) 
(AGENT (TYPE #PERSON# #ORGANIZATION#)) (THEME (TYPE #SITUATION# #ENTITY#) 
(MAPPING (SENT-COMP T))) (GOAL (TYPE #PERSON# #ORGANIZATION#) 
(MAPPING (P-ARG GOAL)))) 
2.4 Merging Predicate-Argument Mapping Information 
For each verb, the information stored in the three levels discussed above is merged to form 
a complete set of mapping rules. During this merging process, the idiosyncrasies take 
precedence over the situation types and the semantic concepts, and the situation types 
over the semantic concepts. For example, the two derived mapping rules for "break"(i.e. 
one for "break" as in "John broke the window" and the other for "break" as in "The 
window broke") are shown in Figure 2. Notice that the semantic TYPE restriction and 
INSTRUMENT role stored in the knowledge base are also inherited at this time. 
110 
(MAPPING (P-ARG I~ 
CAUSED-PROCESS ~I ~ PROCESS-OR-STATE 
"break" 
(AGENT {TYPE (#CREATURE# #ORGANIZATION#}) \] i( THEME (TYPE #ENTITY#} {MAPPING (SURFACE SUBJECT)) | I (MAPPING (SURFA¢ 
(THE~ (TYPE #ENTITY#) j (INSTRUI~NT (TYPE (#PH¥ (MAPPING (SURFACE OBJECT)}) | (MAPPING (p 
(INSTRUMENT (TYPE #PHYSICAL-OBJECT#)) | (MAPPING (P-ARC INSTRUMENT})} 
D 
Sittumt Lon ~"l~s 
I,,,~,,.,4. oon 
ITY#) SURFACE SUBJECT))) I 
|PHYSICAL-OBJECT#)) | P-ARG INSTRU~NT)I)J 
Figure 2: Information from the KB, the situation type, and the lexicon all combine to 
form two predicate-argument mappings for the verb "break." 
3 Automatic Acquisition from Corpora 
In order to expand our lexicon to the size needed for broad coverage and to be able to tune 
the system to specific domains quickly, we have implemented algorithms to automatically 
build multilingual lexicons from corpora. In this section, we discuss how the situation 
types and lexical idiosyncrasies are determined for verbs. 
Our overall approach is to use simple robust parsing techniques that depend on a 
few language-dependent syntactic heuristics (e.g. in English and Spanish, a verb's object 
usually directly follows the verb), and a dictionary for part of speech information. We 
have used these techniques to acquire information from English, Spanish, and Japanese 
corpora varying in length from about 25000 words to 2.7 million words. 
3.1 Acquiring Situation Type Information 
We use two surface features to restrict the possible situation types of a verb: the verb's transitivity rating 
and its subject animacy. 
The transitivity rating of a verb is defined to be the number of transitive occurrences 
in the corpus divided by the total occurrences of the verb. In English, a verb appears in 
the transitive when either: 
• The verb is directly followed by a noun, determiner, personal pronoun, adjective, or 
wh-pronoun (e.g. "John owns a cow.") 
• The verb is directly followed by a "THAT" as a subordinate conjunction (e.g. "John 
said that he liked llamas.") 
• The verb is directly followed by an infinitive (e.g. "John promised to walk the dog.") 
• The verb past participle is preceded by "BE," as would occur in a passive construc- 
tion (e.g. "The apple was eaten by the pig.") 
111 
verb occs TR SA Pred. ST Correct ST Prepositional Idio 
SUFFICE 8 0.6250 0.0000 I$~ I~) 
TIME 15 0.8333 1.0000 C IS) C 
TRAIN 20 1.0000 1.0000 CP IS/ CP PS) at 
WRAP 22 0.7222 0.6667 CP IS CP) up over in with 
PSI I out SORT 25 0.4211 1.0000 CP IS AA CPAA UNITE 27 0.5833 1.0000 CP IS AA PS CPAA 
0. 1 00.  cPI 
SUSTAIN 32 0.9062 0.6842 CP IS CP 
SUBSTITUTE 33 0.7500 0.5000 IS) 'CP PS) for 
TARGET 36 0.7778 0.8000 CP IS 1 'CP 1 from°n STORE 36 0.9091 1.0000 CP IS :cCpP 
STEAL 36 0.9167 0.6667 CP IS 
SHUT 36 0.2400 0.5000 IS PS) 'CP PS) up for 
STRETCH 53 0.5278 0.5000 IS PSI 'CP PS) over into out from 
STRIP 57 0.7609 0.8571 'CP IS) 'CP) from into of 
THREATEN 58 0.8793 0.4419 IS) ~CP IS) over 
01 I iil sS I TREAT 77 0.8052 0.8000 'CP IS as TERMINATE 79 0.9726 1.0000 'CP IS 
'I,~ PS) on with into WEIGH 81 0.2069 0.5294 'C.r IS) 
'CP IS 
TEACH 82 0.7794 0.6875 'CP) at 
SURROUND 85 0.8000 0.6667 /CP/ 
TOTAL 97 0.0515 0.2759 PS) (CP PS) at 
VARY 112 0.1354 0.0294 IS PS) (CP PS) from over 
WAIT 130 0.1923 1.0000 CP IS AA PS) (AA) for up 
SPEAK 139 0.1667 0.7500 'CP IS AA PS) (AA CP) out at up 
SURVIVE 146 0.4754 0.3846 IS PS) (IS PS) 
UNDERSTAND 180 0.6946 0.8684 CP IS) IS) 
SURGE 188 0.0182 0.3125 PS) PS) 
SUPPLY 188 0.7176 0.8571 CP IS) CP) with 
SIT 199 0.0625 0.7027 AA PS) AA PS) on with at out in up 
TEND 200 0.8594 0.4340 (IS) CP IS) 
BREAK 219 0.4771 0.5000 (IS PS) CP PS) up into out 
WRITE 243 0.4637 0.9123 (CP IS AA PS) CP AA) off 
WATCH 268 0.7069 0.8462 (CP IS) CP) out over 
SUCCEED 277 0.5379 0.8899 (CP IS AA PS) CP PS) 
STAY 300 0.2156 0.6604 ~CP IS AA PS) PS) out up on with at 
STAND 310 0.2841 0.7237 (CP IS AA PS) PS CP AA) up at as out on 
TELL 368 0.8054 0.8101 ICP IS) CP) 
SPEND 445 0.3823 0.8125 (CP IS AA PS) CP) on over 
,°4 l, i c ,s I SUGGEST 570 0.7782 0.5918 IS CP IS 
TURN 852 0.3418 0.5891 ~IS PS) CP PS) out into up over 
START 890 0.3474 0.6221 (CP IS AA PS) CP PS) with off out 
LOOK 1084 0.1718 0.6520 (CP IS AA PS) AA PS) at into for up 
THINK 1227 0.7602 0.9237 (CP IS) CP) 
TRY 1272 0.7904 0.8743 (CP IS) CP ) 
WANT 1659 0.8559 0.8787 (CP IS) IS) 
USE 2211 0.8416 0.7725 (CP IS) CP) 
TAKE 2525 0.7447 0.5933 (IS) CP IS) over off out into up 
Table 4: Automatically Derived Situation Type and Idiosyncrasy Data 
112 
Transitivity: 
CP/IS 6.0 ().1 6.2 6.3 
Ambig. , , , ,, :.-., ° 
0.0 0.1 0.2 0.3 
AA/PS ,, , , ,',, 
0.0 O1 0.2 ().3 
*1 0.4 6.5 d.~ 
"''-'" 0.7 o8 0.9 i o 
i ~ 99 ~ qlQ e~ .Tt • 0.4 0.5 0.6 6 0.8 0.9 1.0 
d.4 d.5 6.6 6.7 6.8 6.9 1'.0 
Subject Animacy: 
CP/AA • o e~ we 6.0 6.1 d.2 d.3 d.4 6.5 6.6"' 8 ~ Y 
0.7 0.8 0.9 1.0 
6.1 d.2 d4 w • Ambig. 00"." ".30 T- ,~" ,- ", r 0.5 0.6 0.7 6.8 1.0 0.9 
IS/PS t • - 
O0 IJl 6.2 " 0.3 0.4 6.5 6.6 6.7 6.8 '0.9 i.O 
Figure 3: This graph shows the accuracy of the Transitivity and Subject Animacy metrics. 
For Spanish, we use a very similar algorithm, and for Japanese, we look for noun 
phrases with an object marker "-wo" near and to the left of the verb. A high transitivity 
is correlated with CAUSED-PROCESS and INVERSE-STATE while a low transitivity 
correlates with AGENTIVE.-ACTION and PROCESS-OR-STATE. Table 4 shows 50 verbs 
and their calculated transitivity rating. Figure 3 shows that for all but one of the verbs 
that are unambiguously transitive the transitivity rating is above 0.6. The verb "spend" 
has a transitivity rating of 0.38 because most of its direct objects are numeric dollar 
amounts, Phrases which begin with a number are not recognized as direct objects, since 
most numeric amounts following verbs are adjuncts as in "John ran 3 miles." 
We define a verb's subject animacy to be the number of times the verb appears with 
an animate subject over the total occurrences of the verb where we identified the subject. 
Any noun or pronoun directly preceding a verb is considered to be its subject. This 
heuristic fails in eases where the subject NP is modified by a PP or relative clause as in 
"The man under the car wore a red shirt." We have only implemented this metric for 
English. The verb's subject is considered to be animate if it is any one of the following: 
• A personal pronoun ("it" and "they" were excluded, since they may refer back to 
inanimate objects.) 
• A proper name 
• A word under "agent" or "people" in WordNet (cf. \[14\]) 
• A word that appears in a MUC-4 template slot that can be filled only with humans 
(cf. \[7\]) 
Verbs that have a low subject animacy cannot be either CAUSED-PROCESS or 
AGENTIVE-ACTION, since the syntactic subject must map to the AGENT thematic 
113 
role. A high subject animacy does not correlate with any particular situation type, since 
several stative verbs take only animate subjects (e.g. perception verbs). 
The predicted situation types shown in Figure 3 were calculated with the following 
algorithm: 
1. Assume that the verb can occur with every situation type. 
2. If the transitivity rating is greater than 0.6, then discard the AGENTIVE-ACTION 
and PROCESS-OR-STATE possibilities. 
3. If the transitivity rating is below 0.1, then discard the CAUSED-PROCESS and 
INVERSE-STATE possibilities. 
4. If the subject animacy is below 0.6, then discard the CAUSED-PROCESS and 
AGENTIVE-ACTION possibilities. 
We are planning several improvements to our situation type determination algorithms. 
First, because some stative verbs can take animate subjects (e.g. perception verbs like 
"see", "know", etc.), we sometimes cannot distinguish between INVERSE-STATE or 
PROCESS-OR-STATE and CAUSED-PROCESS or AGENTIVE-ACTION verbs. This 
problem, however, can be solved by using algorithms by Brent \[3\] or Dorr \[8\] for identifying 
stative verbs. 
Second, verbs ambiguous between CAUSED-PROCESS and PROCESS-OR-STATE 
(e.g. "break", "vary") often get inconclusive results because they appear transitively about 
50% of the time. When these verbs are transitive, the subjects are almost always animate 
and when they are intransitive, the subjects are nearly always inanimate. We plan to 
recognize these situations by calculating animacy separately for transitive and intransitive 
cases. 
3.2 Acquiring Idiosyncratic Information 
We automatically identify likely pre/postpositional argument structures for a given verb 
by looking for pre/postpositions in places where they are likely to attach to the verb (i.e. 
within a few words to the right for Spanish and English, and to the left for Japanese). 
When a particular pre/postposition appears here much more often than chance (based 
on either Mutual Information or a chi-squared test \[5, 4\]), we assume that it is a likely 
argument. A very similar strategy works well at identifying verbs that take sententiai 
complements by looking for complementizers (e.g. "that", "to") in positions of likely 
attachment. Some English examples are shown in Tables 4 and 5, and Spanish examples 
are shown in Tables 6 and 7. The details of the exact algorithms used for English are 
contained in McKee and Maloney \[13\]. Areas for improvement include distinguishing 
between cases where a verb takes a prepositional arguments, a prepositional particle, or 
a common adjunct. 
4 Conclusion 
We have automatically built lexicons with predicate-argument mapping information from 
English, Spanish and Japanese corpora. These lexicons have been used for several multi- 
lingual data extraction applications (cf. Aone et ai. \[2\]) and a prototype Japanese-English 
114 
word possible clausal complements 
know THATCOMP 
vow THATCOMP, TOCOMP 
eat 
want TOCOMP 
resume INGCOMP 
Table 5: English Verbs which Take Complementizers 
verb MI with "que" 
indicar 9.3 
sefialar 8.7 
estimar 8.6 
calcular 7.7 
precisar 7.7 
anunclar i 7.7 
Table 6: Spanish Verbs which Take Complementizers 
verb 
luchar 
unit 
vacunar 
cifrar 
consultar 
paaar 
acordar 
contar 
relacionar 
notificar 
oeurrir 
encontrar 
preposition 
contra 
contra 
contra 
sobre 
sobre 
sobre 
con 
con 
con 
en 
en 
en 
MI between verb and preposition 
12.4 
8.9 
8.9 
9.6 
9.6 
8.6 
10.8 
10.3 
9.7 
8.7 
8.0 
7.8 
Table 7: Spanish Verbs that Take Prepositional Arguments 
11 5 
machine translation system. The algorithms presented here have minimized our lexical 
acquisition effort considerably. 
Currently we are investigating ways in which thematic role slots of verb frames and 
semantic type restrictions on these slots can be derived automatically from corpora (cf. 
Dagan and Itai \[6\], Hindle and Rooth \[10\], Zernik and Jacobs \[20\]) so that knowledge 
acquisition at all three levels of predicate-argument mapping can be automated. 

References 
\[1\] Chinatsu Aone and Doug McKee. Three-Level Knowledge Representation of 
Predicate-Argument Mapping for Muitilingual Lexicons. In AAAI Spring Sympo- 
sium Working Notes on Building Lexicons for Machine Translation, 1993. 
\[2\] Chinatsu Aone, Doug McKee, Sandy Shinn, and Hatte Blejer. SRA: Description 
of the SOLOMON System as Used for MUC-4. In Proceedings of Fourth Message 
Understanding Conference (MUC-$), 1992. 
\[3\] Michael Brent. Automatic Semantic Classification of Verbs from Their Syntactic 
Contexts: An Implemented Classifier for Stativity. In Proceedings of the 5th European 
ACL Conference, 1991. 
\[4\] Kenneth Church and William Gale. Concordances for Parallel Text. In Proceedings 
of the Seventh Annual Conference of the University of Waterloo Centre for the New 
OED and Text Research: Using Corpora, 1991. 
\[5\] Kenneth Church and Patrick Hanks. Word Association Norms, Mutual Information, 
and Lexicography. Computational Linguistics, 16(1), 1990. 
\[6\] Ido Dagan and Alon Itai. Automatic Acquisition of Constraints for the Resolution 
of Anaphora References and Syntactic Ambiguities. In Proceedings of the 13th Inter- 
national Conference on Computational Linguistics, 1990. 
\[7\] Defense Advanced Research Projects Agency. Proceedings of Fourth Message Under- 
standing Conference (MUG-4). Morgan Kaufmann Publishers, 1992. 
\[8\] Bonnie Dorr. A Parameterized Approach to Integrating Aspect with Lexical- 
Semantics for Machine Translation. In Proceedings of 30th Annual Meeting of the 
ACL, 1992. 
\[9\] David Dowty. Word Meaning and Montague Grammar. D. Reidel, 1979. 
\[10\] Donald ttindle and Mats Rooth. Structural Ambiguity and Lexical Relations. In 
Proceedings of 29th Annual Meeting of the ACL, 1991. 
\[11\] Manfred Krifka. Nominal Reference, Temporal Construction, and Quantification in 
Event Semantics. In R. Bartsch et al., editors, Semantics and Contextual Expressions. 
Forts, Dordrecht, 1989. 
\[12\] Susumu Kuno. The Structure of the Japanese Language. M1T Press, 1973. 
\[13\] Doug McKee and John Maloney. Using Statistics Gained from Corpora in 
a Knowledge-Based NLP System. In Proceedings of The AAAI Workshop on 
Statistically-Based NLP Techniques, 1992. 
\[14\] George Miller, Richard Beckwith, Christiane Fellbaum, Derek Gross, and Katherine 
Miller. Five papers on WordNet. Technical Report CSL Report 43, Cognitive Science 
Laboratory, Princeton University, 1990. 
\[15\] Marc Moens and Mark Steedman. Temporal ontology and temporal reference. Com- 
putational Linguistics, 14(2), 1988. 
\[16\] Sergei Nirenburg and Lori Levin. Syntax-Driven and Ontology-Driven Lexical Se- 
mantics. In Proceedings of ACL Lexical Semantics and Knowledge Representation 
Workshop, 1991. 
\[17\] Boyan Onyshkevych and Sergei Nirenburg. Lexicon, Ontology and Text Meaning. 
In Proceedings of A CL Lezical Semantics and Knowledge Representation Workshop, 
1991. 
\[18\] Leonard Talmy. Lexicalization Patterns: Semantic Structure in Lexical Forms. In 
Timothy Shopen, editor, Language Typology and Syntactic Descriptions. Cambridge 
University Press, 1985. 
\[19\] Zeno Vendler. Linguistics in Philosophy. Cornell University Press, 1967. 
\[20\] Uri Zernik and Paul Jacobs. Tagging for Learning: Collecting Thematic Relations 
from Corpus. In Proceedings of the 13th International Conference on Computational 
Linguistics, 1990. 
