Text Meaning Representation for Chinese 
Wanying Jin 
Computing Research Laboratory 
New Mexico State University 
Las Cruces, NM 88003-8001 
U.S.A. 
wanying@crl.nm~u.edu 
Abstract 
This paper describes text meaning represen- 
tation for Chinese. Text meaning representa- 
tion is composed of a set of ontological concept 
instances along with ontological links among 
them. It integrates lexical, textual and world 
knowledge into a single hierarchical frame- 
work. In NLP application it serves as an inter- 
lingual representation for various processing. 
The methodology and implementation of text 
meaning representation is discussed in detail. 
1 
1 Introduction 
In natural  text processing, it be- 
comes inevitable to require a system having 
capability to automatically extract and repre- 
sent the information conveyed in a given text. 
The theory and methodology of text mean- 
ing representation (TMR) has been studied 
in the past decade (Nirenburg, 1991; Ma- 
hesh and Nirenburg, 1996; Beale et al., 1995; 
Onyshkevych 1997) and its application has 
been presented in machine translation sys- 
tems that employ interlingual approach. An 
ideal text meaning representation will be a 
-neutral description of the linguistic 
information conveyed in a natural  
text. TIVIR captures the meanings of words in 
the text and represents them in a set of onto- 
logical concepts interconnected through onto- 
logical relations. In a~:ldition, TMR provides 
information about the lexicon-semantic de- 
pendencies as well as stylistic factors. Based 
on this study, two supportive resources to 
compose TMR are ontology and semantic lex- 
icon. An ontology is a set of knowledge con- 
cepts about the world. It is composed of thou- 
l This work has been supported by the Department 
of Defense of the United States under contract number 
MDA-904-92-C,-5189. 
sands of concepts organized into a particular 
hierarchy so that each concept is related to 
other concepts through semantic links (Carl 
son et aI., 1990; Bateman, 1993; Dowellet al. 
1995; Nirenburg et al., 1995; Bouaud, et al: 
1995; Mahesh et al. 1995;). Semantic lexicon 
represents the senses of the words. Each lex- 
icon entry is a framework that maps a word 
sense to an ontological concept. It also in- 
dudes information about syntax, semantics, 
morphology and pragmatics, as well as anno- 
tation that keep the record of data manage- 
ment. Over the years, various methodologies 
have been investigated carrying out the struc- 
ture of a computational lexicon entry within a 
knowledge-based framework(Onyshkevych et 
aI, 1995; Viegas and Raskin, 1998; Viegas: 
1999; Viegas, et a/, 1998a, 1998b). A semantic 
parser uses information in the semantic lex/- 
con and makes a decision on word sense dis- 
ambiguation based on the strategy proposed 
by Beale et aI. (1995) and Beale(1997). 
This paper presents the development and 
application of text meaning representation 
for Chinese. Detailed discussion about the 
principle for building a computational se- 
mantic Chinese lexicon is illustrated in Jin 
(1999). The methodology and implementa- 
tion of word sense disambiguation is fully dis- 
cussed in Viegas, Jin and Beale(1999a: 1999b, 
1999c). 
2 Overview of Ontology 
An ontology is a body of knowledge about 
the world. It is a repository of concepts used 
in meaning representations. All concepts are 
organized in a tangled subsumption hierar- 
chy and further interconnected using a system 
of semantic relations defined among the con- 
cepts. The ontology is put into well-defined 
109 
relationships with knowledge sources in the 
system. In an NLP application the ontology 
supplies world knowledge to lexical, syntactic, 
semantic and pragmatic processes. 
In the MikroKosmos project, 2 the ontolog- 
ical concepts consist of: OBJECT, the static 
things existing in the world; EVENT, any ac- 
tivities happening in the world, and PROP- 
ERTY, the properties of OBJECTs and EVENTs. 
The ontology organizes terminologicalnouns 
into a taxonomy of objects, verbs into a taxon- 
omy of events, and adjectives into a taxonomy 
of attributes. It further includes many onto- 
logical relations between objects and events 
to support a variety of disambiguation tasks. 
Currently, the ontology in the IVIikroKosmos 
project contains about 5,000 concepts cover- 
ing a wide range of categories in the world. 
Each concept, on average, has 14 relation 
links. An example below presents the top 
three level of the ontology, which differentiates 
between OBJECT; EVENT and PROPERTY. 
+ ALL 
+EVENT 
+ MENTAL-EVENT 
+ PHYSICAL-EVENT 
+ SOCIAL-EVENT 
+ OBJECT 
+MENTAL-OBJECT 
+ INTANGIBLE-OBJECT 
+ PHYSICAL-OB/ECT 
+ SOCIAL-OBJECT 
+ PROPERTY 
+ ATTRIBUTE 
+ RELATION 
+ 0NTOLOGY-SLOT 
Each concept is a frame in which a col- 
lection of ontological slots such as DEFINI- 
TION, IS-A, SUBCLASSES, INVERSE~ case roles 
such as AGENT, THEME and properties such 
as HEADED-BY, HAS-MEMBER, etc. link one 
concept to other concepts in the ontology. 
Below is an example of a concept frame of 
GOVERNMENT-ACTIVITY: 
2The MikroKosmos project is a knowledge-based 
machine translation system using an interHngual ap- 
proach. The so~ce s are Spanish, Chinese 
and Japanese. The target  is English. 
Concept GOVERNMENT-ACTIVITY 
DEFINITION: an activity commonly 
carried out by a 
government. 
IS-A: POLITICAL-EVENT 
AGENT: HUMAN 
THEME: EVENT OBJECT 
ACCOMPANIER: HUMAN 
LOCATION: PLACE 
This example indicates that the concept 
GOVERNMENT-ACTIVITY is a subclass of 
the concept POLITICAL-EVENT~ its case role 
AGENT requires its semantic value as HUMAN 
and its THEME requires its value as either OB- 
JECT or EVENT~ the GOVERNMENT-ACTIVITY 
can also have case role ACCOMPANIER with 
value as HUMAN and LOCATION with value as 
PLACE. Any lexicon entry mapping to the con- 
cept GOVERNMENT-ACTIVITY gets extended 
information through this frame. 
3 Semantic Lexicon 
Semantic lexicon is another knowledge source 
for text meaning representation. In the 
M.ikroKosmos project each lexicon entry is de- 
signed as a frame with 11 zones corresponding 
to information relevant to orthography, mor- 
phology, syntax, semantics, syntax-semantic 
linking, stylistics, and database type manage- 
ment record, etc. The core of the lexicon 
frame is syntactic zone SYN-STRUC, semantic 
zone SEM-STRUC and their link SYNSEM zone. 
Syntax particular to a given  is de- 
scribed in the syntactical zone. The semantic 
zone maps a sense into an ontological concept 
in the case of single sense, or to several con- 
cepts in the case of multiple senses. Through 
the syntactic-semantic link zone the informa- 
tion of each word in the text can be extracted 
directly from lexicon database and its relevant 
world knowledge also can be retrieved. The 
general template of semantic lexicon entry is 
shown as follows. For a detailed description 
see Viegas and l~kin (1998). 
Entry Elements = 
\[FORM: Word Form 
CAT: Part of speech 
MORPH: Morphology 
ANNO: Annotation 
TRANS: Translation 
110 
SYNSEM: 
SYR-STRUC: 
SEM-STRUC: 
LEX-RULES: 
PRAGM: 
STYL: 
Syn1;ax-Semant ic link 
Syntactic strucCure 
Semant ic sl;ructure 
Lexical rules 
Pragmatic information 
Stylistic informazion\] 
'~k-Vl 
syn: I~jb.j t: \[\] : ~\]\[cat: NP\]I 
: \[\]\[cat: NP\]J 
sere: 
;ynsem: 
IT G°vernment-Activity an 1 
AGENT: ~-l Hum 
HEME: ~-~ Object J 
:bj ubj: E\]\[ sere: \[~\]1\] 
: ~\[se~: ~\]J 
Figure 1: Sense for the Chinese lexicon ~k. 
Figure 1. is a simplified structure of a 
Chinse lexicon entry ~k in the sense of 
{GOVERNMENT-ACTIVITY. The SYN zone in- 
dicates that when parsing a sentence contain- 
ing this entry, subcategories SUBJ and OBJ are 
required. The SEM zone presents the semantic 
value of each case role, i.e. AGENT with value 
HUMAN, and TttEM~ with value EVENT or OB- 
JECT. The SYNSEM zone provides information 
about the syntax-semantic linking. That is, 
SUBJ \[\] is linked to AGENT ~-l with value 
HUMAN and OBJ \[\] is linked to THEblE 
with value EVENT or OBJECT. 
Due to the lack of morphological informa- 
tion in Chinese, it is often the case that the 
same Chinese word form can be mapped to 
a different part of speech and has multiple 
senses, such as the word ~:g~: 
• in the context ~:~ ~:~ flowers bloom, 
can be an intransitive verb mapping to 
a concept BLOOM with the definition to 
produce flower. 
• in the context ~ ~k ~ ~ ~ the 
government opens the foreign trade pol- 
icy, can be a transitive verb mapping to 
a concept GOVERNMENT-ACTIVITY with 
the definition an activity that is com- 
monly carried out by a government at any 
level. 
• in the context ~ ~ ~k ~ the gov- 
ernment carries open policy, can be an 
adjective mapping to OPEN-TO-PUBLIC 
with the definition to be available to the 
public. 
• in the context I~l~ ~k the library is 
open, can be an intransitive verb with the 
same concept OPEN-TO-PUBLIC. 
Using the ontological concepts as the value of 
semantic variables and linking them to syn- 
tactic variables makes the lexicon very infor- 
mative. Figure 2. and Figure 3. present each 
sense and POS of the lexicon entries for the 
Chinese word ~. 
~-v2 
syn: \[root: \[\] 
Lsubj: \[~\]\[cat: N P\]\] 
IT Bl°°ra 1 HEME" FlowerJ ; SeN: . \[\] 
.s em: hubS: Elbe : 
Figure 2: Sense for the Chinese lexicon ~k. 
RE-v3 
syn: \[root: \[subj: \[~\[cat: NP\]\] 
Open-to-Public sere: \[THEME: ~ Object 
synsem: \[subj: \[~\]\[sem: \[~\]\]\] 
Figure 3: Sense for the Chinese lexicon R~. 
4 Semantic Analysis for Word 
Sense Disambiguation 
The task of a semantic analyzer is to com- 
bine the knowledge contained in the ontology 
and lexicon and apply it to the input text to 
produce text meaning representation output. 
The central tasks involved are to retrieve the 
appropriate semantic constraints for each pos- 
sible word sense, test each sense in context, 
and construct the output TMBs by instanti- 
ating the concepts in semantic zones of the 
word senses that best satisfy the combination 
of constraints. Figure 4. illustrates the pro- 
cess of text meaning representation. Below 
illustrates the process through a sentence ~ \[\] 
111 
~t~ ~:~ T ~J~ ~ ~. The Chinese gov- 
ernment has opened foreign trade policy. The 
syntactic analysis gives the following output: 
((ROOT ~)(CAT V)(TRANS open) 
(s~J 
(MODS ~\[\]) (CAT N) (TRANS China) 
(ROOT ~'~) (CAT N) (TRANS government) ) 
(0BJ 
(~0DS 
(HODS ~) (CAT ADJ) (TRAILS foreign) 
(ROOT ~ ~ ) (CAT N) (TRNAS trade) ) 
(ROOT ~)(CAT N)(TRANS policy) )) 
~PUTTEX~ S'VHTACIlCpA.~R J PANGLOSS ~n,s*l,.w u,t S~lem 
SYHTACllCO-SEMAINrnc P~.PROCESS~ 
TP~O PARSE 
FOREST 
IttSTANTLk'nON taXI SYNTAX-S~ANlrIC$ 
VARIABLE B~I~IG 
FR,~NtE$ & $&OF$ 
WORD SBCSE S~.ECnON 
(¢~.~.&,; ChKtdng) 
TMRs lira 
COMMNA~OW IMR SIJ~G 
~w 
C~ 
l~Fm 
| -- mC~Oll.H!O~.S 
Figure 4: The Architecture of TMR. 
The semantic analysis process takes the fol- 
lowing steps: 
• to gather all of the possible lexicon for 
each of the words with instantiated each 
concepts. 
~\[\] N CHINA-6 
~ ~ ZSDZ~TIO~-I 
~ VI BLOOM-2-1 
V2 GDVERNME~rr-ACTIVITY- 2-2 
V3 OPI~I-TO-PUBLIC-2-3 
ADJ OPEN-TO-PUBLIC-2-4 
~ M)J1 INTERNATIONAL-ATTRIBUTE-5-1 
ADVI OUTVARD-5-2 
~, N c0mrzacz-zvmrr-4-1 
V CO ~..RCE-EVEIWr-4-2 
Syntactic variables are bound to one an- 
other using the syntactic patterns in the 
lexical entries to establish syntactic de- 
pendencies. In addition: ontological con- 
cepts referred to the semantic zones of 
the lexical entries are instantiated and 
linked through ontological relations to es- 
tablish semantic dependencies. For ex- 
ample, the syntactic structure of the text 
requires ~k to be a verb. Thus: the ADJ 
category with sense OPEN-TO-PUBLIC-2-4 
is rejected. From Figure 2. and Figure 
3. both SYN zones indicate an intransi- 
tive verb that violates the required syn- 
tax. Therefore, the concepts BLOOM and 
OPEN-TO-PUBLIC are also rejected. In 
the same way, the adverb ~b with sense 
OUTWARD-5-2 and the verb ~ with 
sense COMMERCE-EVENT-4-2 are also re- 
jected because of the violation of required 
POS. Finally the ADJ .~ with the sense 
INTERNATIONAL-ATTRIBUTE-5-1 and the 
NOUN ~ with COMMERCE-EVENT-4-1 
are selected. After all senses are de- 
termined: SYN-SEM zone binds all syn- 
tactic variables with semantic variables, 
i.e. SUBJ FEDERATION-I is bound to 
the AGENT of GOVERNMENT-ACTIVITY- 
2-2, OBJ LAW-3 is bound to the THEME of 
GOVEB_NMENT-ACTIVITY-2-2. 
• In the next step, selectional constraints 
are retrieved from the ontology. Individ- 
ual selectional constraints are checked. In 
the example, the concept GOVEthNMENT- 
ACTIVITY requires AGENT to be HUMAN 
and THEME to be EVENT or OBJECT. 
The lexical information indicates that the 
SUBJ ~ with sense FEDERATION must 
satisfy the AGENT of GOVERNMENT- 
ACTIVITY with value HUMAN. An infer- 
ence rule described below checks the sat- 
isfaction. The OBJ ~ with the sense 
LAW satisfies THEME of GOVERNMENT- 
112 
ACTIVITY with value OBJECT. Through 
IS-A links it is found that LAW is a descen- 
dant of OBJECT. Therefore, the semantic 
constraints are satisfied. 
Seeking satisfaction through inference 
rules, the semantic analyzer does more 
than match selectional constraints or find 
the distance along IS-A links. The search 
inside the ontology also involves look- 
ing for metonymic type links, such as 
FEDERATION in a metonymic relation 
with HUMAN through the property HAS- 
REPRESENTATIVE: 
Concep%: 
IS-A: 
DOMAIN: 
RANGE: 
INVERSE : 
HAS-REPRESENTATIVE 
ORGANIZATION-RELATION 
ORGANIZATION 
HUMAN BUSINESS-ROLE 
GOVERNMENTAL-ROLE 
REPRESENTATIVE-OF 
in which DOMAIN is ORGANIZATION that 
has subclass FEDERATION and RANGE is 
HUMAN. Thus, the constraint of AGENT 
of GOVERNAIENT-ACTIVITY to be HU- 
MAN is satisfied. 
• In case multiple senses all satisfy the con- 
straints, the concept with the shortest 
path is selected as the best choice. An 
ontological search program, Onto-Search, 
is presented in Onyshkevych (1997). The 
resulting preference values for each con- 
stralnt are combined in an efficient con- 
trol and search algorithm called Hunter- 
Gatherer that combines constraint satis- 
faction~ branch and bound, and solution 
synthesis techniques to pick the best com- 
bination of word senses of the entire sen- 
tence in near linear time, as described in 
Beale (1997). 
• Chosen word senses are assembled into 
TMR frames. 
5 Text Meaning Representation 
i text meaning representation(TMR) is a 
-neutral description of the meaning 
conveyed in a text. It is derived by syn- 
tactic and semantic analysis. TMR captures 
not only the meaning of individual words in 
the text, but also the relation between those 
words. It provides information about the 
lexicon-semantic dependencies. In addition, it 
also represents stylistic and other factors pre- 
sented in the text. From the result of word 
sense disambiguation, TMR integrates lexical, 
ontological and textual information into a sin- 
gle hierarchical framework. Below is a TMR 
for the example sentence ~\[\] R~ ~:~ T 
~ ~ ~ R~. Chinese government has opened 
.foreign trade policy. 
(FEDERATION-1 
(AGENT-0F 
(VALUE GOVERNMENT-ACTIVITY-2) ) 
(RELATION (VALUE CHINA-6) ) 
(INSTANCE-OF (VALUE FEDERATION) ) ) 
( GOVERNMENT-ACTIVITY-2 
(AGENT (VALUE FEDERATION-i) ) 
(THEME (VALUE LAW-3)) 
(INSTANCE-OF 
(VALUE GOVERNMENT-ACTIVITY) ) ) 
(LAW-3 
(THmm-0Y 
(VALUE GOVERNMENT-ACTIVITY-2) ) 
(INSTANCE-OF (VALUE LAW) ) ) 
(COMMERCE-EVENT-4 
(EVENT-OBJECT-RELATION (VALUE LAW-3) ) 
(INSTANCE-OF (VALUE COMMERCE-EVENT) ) ) 
(INTERNATI ONAL-ATTRIBUTE-5 
(DOMAIN (VALUE COMMERCE-EVENT-4) ) 
(RANGE (VALUE INTmU~ATIONAL)) 
(INSTANCE-OF 
(VALUE INTERNATIONAL-ATTRIBUTE) ) ) 
(CHINA-6 
(INSTANCE-OF (VALUE CHINA) ) ) 
(ASPECT-7 
(SCOPE (VALUE GOVERNMENT-ACTIVITY-2) ) 
(TELIC (VALUE YES) ) ) 
(TIME-8 
(DOMAIN (VALUE GOVERNMENT-ACTIVITY-2) ) 
(RANGE (VALUE *speaker-time*))) 
(INFERENCE (TYPE METONYMY) 
(HUMAN 
(REPRESENTATIVE-OF 
(VALUE FEDERATION-1 ) ) ) ) 
After semantic analysis~ a variety of mi- 
crotheories are applied to further analyze el- 
ements of text meaning such as time, aspect: 
propositions, sets, co-reference, and so on.. 
to produce a complete TMR. In the exam- 
ple, ASPECT-7 is applied within the scope 
of GOVERNMENT-ACTIVITY in which TELIC 
with value YES indicates the GOVERNMENT- 
ACTIVITY is complete that means the action 
113 
of opening foreign trade policy is done. TIME- 
8 indicates the GOVERNMFENT-ACTIVITY hap- 
pens at the time the speaker make the ut- 
terance. Thus, the meaning of the Chinese 
sentence ~H\] ~ ~ ~ 3" ~t" ~ ~ is 
completely represented in the TlVIR. 
6 Discussion 
A knowledge-based machine translation can 
be viewed as extracting and representing the 
meaning of a text and generating a text in 
target  based on the meaning pre- 
sented. Thus, text meaning representation 
plays the key role in an interlingual approach 
to machine translation. The approach de- 
scribed in this article enables integrating lin- 
guistic knowledge of source s with 
general world knowledge to reach high qual- 
ity translation. It is because TMR represents 
meaning deeper and broader than what the 
context presents. For example, in the con- 
text ~. ~9~ provide service, the linguistic 
information indicates syntactic-semantic de- 
pendency as SUBJ (human) -V(service-event)- 
OBJ(event). In an ontology, the SERVICE- 
EVENT concept frame contains information 
about AGENT and THEME as well as BENEFI- 
CIARY, ACCOMPANIEtL INSTRUMENT, LOCA- 
TION, etc. which extends the meaning of the 
given text ~ ~ to lEA. :~ ~ ~ i~,A. 
~. ~ ~ Someone provides service to someone 
else at some place. If the extended informa- 
tion is not explicitly presented in the text, the 
default value provides the assumption based 
on the world knowledge. In machine transla- 
tion, it enables the generation of high quality 
text, especially in the case where the syntax 
in source  is different from that in 
target  or in the case where ellipsis 
is allowed in one  such as He plays 
Bach in English, but is not allowed in other 
 such as in Chinese, where one must 
say ~ ~ ~ fi.\] ~ ~ He plays Bach's work. 
With the rapid development of internet, in- 
formation retrieval plays the key role in search 
engin. The extended information about the 
world knowledge allows to retrieve relevant 
data through the ontology that is implicitly 
specified in the query. As result, more broad 
and deep information can be extracted. It is 
extremely valuable to the development of next 
generation of internet search engin. All in all, 
using ontology strengthens all NLP systems. 

References 
Bateman, J. A. 1993. Ontology Construc- 
tion and Natural Language. Proc. Inter- 
national Workshop on Formal Ontology, 
Padua, Italy. 
Bea.le, S. 1997. Hunter-Gatherer: Apply- 
ing Constraint Satisfaction, Branch-and- 
Bound and Solution Synthesis to Compu- 
tational Sernantics; Ph.D. Diss., Carnegie 
Mellon University. 
Beale, S., Nirenburg, S. and K. Mahesh. 1995. 
Semantic Analysis in the MikroKosmos Ma- 
chine Translation Project. Proc. of the 2nd 
SNLP-95, Bangkok, Thailand. 
Bouaud, J., Bachimont, B., Charlet, J., and 
Zweigenbaum, P. 1995. Methodological 
Principles for Structuring an ONTOLOGY. 
Proc. the Workshop on Basic Ontological 
Issues in Knowledge Sharing, International 
Joint Conference on Artificial Intelligence 
(IJCAI-95), Montreal, Canada. 
Carlson, L. and Nirenburg, S. 1990. World 
Modeling for NLP. Technical Report CMU- 
CMT-90-121, Center for Machine Trans- 
lation, Carnegie Mellon University, Pitts- 
burgh, PA. 
Dowell, M., Stephen, L., and Bonnell, R. 
1995. Using a Domain Knowledge Ontology 
as a Semantic Gateway among Database. 
Proc. the Workshop on Basic Ontological 
Issues in Knowledge Sharing, International 
Joint Conference on Artificial Intelligence 
(IJCAI-95), Montreal, Canada. 
Jin, W., Viegas, E. and Beale, S. 1999. Build- 
ing a Chinese Computational Semantic Lex- 
icon Proc. of International Symposium on 
Machine Translation and Computer Lan- 
guage Information Processing-1999 (ISMT 
~4 CLIP), Beijing, China. 
Mahesh, K. and Nirenburg, S. 1995c. Se- 
mantic Classification for Practical Natu- 
ral Language Processing. Proc. Sixth ASIS 
SIGICR Classification Research Workshop: 
An Interdisciplinary Meeting. Chicago, IL. 
Mahesh, K and Nirenburg, S. 1996. Mean- 
ing Representation for Knowledge Sharing 
in Practical Machine Translation. Proe. the 
FLAIRS-96 Track on Information Inter- 
change. Florida AI Research Symposium. 
Mahesh, K. 1996. Ontology Development for 
Machine Translation: Ideology and Method- 
114 
ology. MCCS-96-292, Computing Research 
Laboratory, New Mexico State University. 
Nirenburg, S., P~skin, V. and Onyshkevych, 
B. 1995. Apologiae Ontologiae. Proc. of 
The Conference on TMI, Leuven~ Belgium. 
Nirenburg, S. 1991. Application-Oriented 
Computational Semantics. Computational 
Linguistics and Formal Semantics, R. John- 
son and M. Rosner (eds.) 
Onyshkevych, B. 1997. An Ontological- 
Semantic Framework for Text Analysis. 
Ph.D. Diss., School of Computer Science, 
Carnegie Mellon University. 
Onyshkevych= B. and Nirenburg, S. 1995. 
A Lexicon for Knowledge-Based MT. Ma- 
chine Translation Issue on Building Lemcon 
for MT~ B. Dorr and J. Klavens (eds.) 10:1- 
2, 5-57. 
Viegas, E. and Raskin. V. 1998. Com- 
putational Semantic Lexicon Acquisition: 
Methodology and Guidelines. MCCS-98- 
315. Computing Research Laboratory, New 
Mexico State University. 
Viegas, E. 1999. An Overt Semantics with 
a Machine-guided Approach for Robust 
LKBs. In Proc. of SIGLEX99 Standardiz- 
ing Lexical Resources, University of Mary- 
land. 
Viegas, E., Beale, S. and S. Nirenburg. 1998. 
The Computational Lexical Semantics of 
Syntagmatic Relations. Proc. of the 36th 
ACL and the 17th COLING, Montreal, 
Qu&bec, Canada. 
Viegas, E., Jin, W. and Beah, S. 1999. 
A Knowledge-based Approach for Chinese- 
English Translations Proc. of 5th Natural 
Language Processing Pacific Rim Sympo- 
sium (NLPRS-99), Beijing, China. 
Viegas, E., Jin, W. and Beale, S. 1999. Long 
Time No See: Overt Semantics for Machine 
Translation Proc. of Conference on The- 
oretical and Methodological Issues in Ma- 
chine Trar~.¢lation (TMI-99), England. 
Viegas, E., Jin, W. and Beale, S. 1999. Us- 
ing Computational Semantics for Chinese 
Translation Proc. off Machine Translation 
Summit-99: Singapore. 
