Symmetric Pattern Matching Analysis for English Coordinate 
Structures 
Akitoshi Okumura and Kazunori Muraki 
NEC Corp. Information Technology Research Laboratories 
4-1-1 Miyazaki, Miyamae-ku, Kawasaki, Kanagawa 216, JAPAN 
okumura@hum.cl.nec.co.jp 
Abstract 
The authors propose a model for analyz- 
ing English sentences including coordinate 
conjunctions such as "and", "or", "but" and 
the equivalent words. Syntactic analysis 
of the English coordinate sentences is one 
of the most difficult problems for machine 
translation (MT) systems. The problem 
is selecting, from all possible candidates, 
the correct syntactic structure formed by 
an individual coordinate conjunction, i.e. 
determining which constituents are coor- 
dinated by the conjunction. Typically, so 
many possible structures are produced that 
MT systems cannot select the correct one, 
even if the grammars allow to write the 
rules in the simple notations. This pa- 
per presents an English coordinate struc- 
ture analysis model, which provides top- 
down scope information of the correct syn- 
tactic structure by taking advantage of the 
symmetric patterns of the parallelism. The 
model is based on a balance matching oper- 
ation for two lists of the feature sets, which 
provides four effects: the reduction of anal- 
ysis cost, the improvement of word dis- 
ambiguation, the interpretation of ellipses, 
and robust analysis. This model was prac- 
tically implemented and incorporated into 
the English-Japanese MT system, and pro- 
vided about 75% accuracy in the practical 
translation use. 
1 Introduction 
This paper presents a model for analyzing English 
sentences including coordinate conjunctions such as 
"and" , "or" , "but" and equivalent words. 
Syntactic analysis of English coordinate sentences 
is one of the most difficult problems for machine 
translation (MT) systems. The problem is select- 
ing, from all possible candidates, the correct syn- 
tactic structure formed by an individual coordinate 
conjunction, i.e., determining which constituents are 
conjoined by the conjunction. Although the con- 
junction appears to have a simple function in the 
English language, it has been researched as a con- 
junct scope problem by both theoretical and com- 
putational linguists. Theoretically, it is possible to 
describe the syntactic and semantic constraints that 
govern the acceptability of a structure in which two 
constituents are conjoined by the conjunction(Lesmo 
and Torasso, 1985; Gazdar, 1981; Schachter, 1977). 
Computationally, it is possible to describe the gram- 
mar and heuristic rules for these constraints by 
ATN networks, logic grarnmars~ HPSG~ and cate- 
gorical grammars(Kosy, 1986; Fong and Berwick, 
1985; Huang, 1984; Boguraev et al., 1983; Blackwell, 
1981; Niimi et al., 1986). However, it is not easy to 
apply these techniques to large-scale MT systems, 
because there exist a variety of conjoined patterns, 
many word ambiguities, some unknown words and 
ellipses of the words simultaneously, in real environ- 
ments. Also, there may be several conjunctions and 
the equivalent words, such as commas, in a single 
sentence. Typically, the methods produce so many 
possible structures that MT systems cannot select 
the correct one, even if the grammars allow to write 
the rules in the simple notations. 
Often, conjunctions might produce the reading 
difficulty even for the human readers. However, they 
also give the readers a kind of symmetry as a read- 
ing indication. They exhibit the tendency to con- 
join the same kind of syntactic patterns, which has 
been called "parallelism" (Beaugrande and Dressier, 
1981; Shinoda, 1981). In Japanese, the similar- 
ity is used for analyzing conjunctive structures and 
the method is found effective(Kurohashi and Nagao, 
1992). While Japanese language has several coordi- 
nate conjunctions according to the syntactic levels 
(a noun phrase and a predicative clause), English 
coordinate conjunctions are used for any level of the 
structures. More robust methods are necessary for 
dealing with English conjunctive structures. 
We propose here an English coordinate structure 
analysis model, which can determine the correct syn- 
tactic structure in real environments by taking ad- 
vantage of the symmetric patterns of the parallelism. 
41 
The model is based on a balance matching operation 
for two lists of the feature sets. The lists represent 
the left-side and right-side structures of the coordi- 
nate conjunctions, including word ambiguities. The 
operation determines the most symmetric structure 
by comparing both sides of the conjunction. 
First, the problems of the English coordinate sen- 
tences are explained. Second, it is mentioned that 
the parallelism can be effective information for top- 
down analysis of the sentences. Third, the balance 
matching analysis model is presented for solving the 
problems. This model, which was implemented in 
the PIVOT English-Japanese MT system with the 
dictionary of 100,000 words, has been working in 
the analysis module. Finally, the results in the MT 
system are reported together with the MT system 
configuration. 
2 Problems of the conjunctions 
Coordinate conjunctions for MT systems present 
three difficulties(Kosy, 1986; Huang, 1983; Niimi et 
al., 1986; Okumura et ai., 1987). 
1. Analysis cost: English coordinate conjunctions 
have a variety of linguistic functions. The con- 
junctions can syntactically conjoin any parts 
of speech; nouns, adjectives, verbs, etc., and 
all sorts of constituents; words, phrases and 
clauses. They produce so many possible struc- 
tures that MT systems cannot select the correct 
one, even if some grammars provide the simple 
notations of the rules. The complexity of the 
rules impose a burden on the analysis process. 
2. Ambiguities of the words: Most English words 
are ambiguous in their parts of speech ,such as 
Noun and Verb, Preposition and Conjunction, 
Auxiliary Verb and others, and Adjective and 
Noun. The complex rules governing the con- 
junctions make the word disambiguation more 
complicated, which results in the reduction of 
analysis precision. 
3. Ellipses of the words: It is possible for a con- 
junct to contain "gaps" (elided elements) which 
are not allowed if the conjunction is removed. 
These gaps must be filled with elements from 
other conjunct for a proper interpretation, as 
in (1) and (1)'. 
(1) NO.2 landing gear selector valves should be 
closed to the full position,and NO.3 to the hM£ 
(1)' NO.2 landing gear selector valves should be 
closed to the full position, and NO.3 \[landing 
gear selector valves should be closed\] to the half 
\[position\]. 
3 Parallelism of the conjunctions 
3.1 Symmetric patterns of parallelism 
English coordinate conjunctions have a tendency to 
conjoin the same kinds of syntactic patterns. We 
identify three levels of symmetric pattern. 
• Phrase(Clause) symmetric patterns: 
Phrase(Clause)-level symmetric patters such as 
\[Verb-Phase AND Verb-Phase\] in the conjunct 
scope of "as well as " in (2) and a series of com- 
mas in (3). 
(2) Such coupling is desirable because it enables a 
development engineer to move easily within this 
hierarchy as well as to exploit the distinctive 
features of each system. 
(3) The add operators cause POSTSCRIPT to pick 
up the top two numbers from the stack, re- 
move them, add them, and leave the sum on 
the stack. 
• Word symmetric patterns: Word-level sym- 
metric patterns such as \[Quantifier Preposi- 
tion Abstract-Noun AND Quantifier Preposi- 
tion Abstract-Noun\] in the conjunct scope of 
the "and" in (4). Some patterns are represented 
by the semantic features such as \[Instrument 
AND Instrument\] about "and1" of (5). 
(4) The container need not be large; if it is lOcm in 
diameter and 12cm in depth, that is enough. 
(5) Inspect the cockpit indicators and1 levers for 
cracked glass and missing control knobs. 
• Morphological symmetric patterns: Morpholog- 
ical symmetric patterns are recognized by the 
sorts of characters, uppercase or lowercase let- 
ters, as in (6) and (7) as well as an exactly same 
morphological pattern \[CIC ... hatches AND 
CIC ..., hatches\] in (8). 
(6) An atomic bomb is a device for producing 
an explosively rapid neutron chain reaction in 
uranium-235 or plutonium-239 which is called 
a fissile material. 
(7) Technical orders described in AFR 8-2 and 
PFR 7-2 are registered in the on-line file in the 
form of inspection workcards. 
(8) There are CIC1 ditching2 hatches3 and CIC4 
escape5 hatches6 in the compartment. 
Some symmetric patterns may appear in com- 
bined form \[Preposition Gerund Nominal-Phrase 
AND Preposition Gerund Nominal-Phrase\] in (9). 
(9) Radioisotopes have played an important part 
in1 developing effective insecticides in2 the 
country and in3 finding the best ways o£ ap- 
plying them. 
42 
8.2 Analysis by {he symmetric patterns 
The symmetric patterns can be effective information 
for top-down analysis of the conjunct scope. For ex- 
ample, though (9) allows another counterpart ("in2 
the country") as the conjoined phrase ("in3 find- 
ing ..."), the symmetric pattern information makes 
it easy to to select the correct counterpart of the 
phrase ("in1 developing .... " ). Also, where other ex- 
amples syntactically allow other counterparts of the 
conjoined noun phrases, the symmetric pattern in- 
formation enables easy selection. Often, the scope of 
each conjunct is explicitly demarcated with commas 
and morphological patterns, as in (3) and (8). The 
symmetric patterns can be also effective for word dis- 
ambiguation. For example, (8) contains verb/noun 
ambiguities for escape5 and hatches6. The symmet- 
ric patterning of ditching2 hatches3 facilitates their 
disambiguation. 
In the above example sentences, the words im- 
mediately following the conjunction play important 
roles for detecting the structures, because there is 
usually strong similarity between the starting words 
of each conjunct scope and the words following the 
conjunctions. However, the following examples also 
contain kinds of symmetric patterns, though the 
words following the conjunctions don't have similar- 
ity with the starting words of the conjunct scopes. 
(10) In 1985 the government offered offshore regis- 
tration to the companies, and, in consequence, 
in 1985 incorporation fees generated about two 
million dollars. 
(11) The damage of the landing gear selector valve 
caused the leakage of the hydraulic fluid, and 
completely blockaded the return path. 
(12) Close the cockpit ditching hatches, and 
the cabin pressure will be dumped to relieve the 
air loads on the hatches. 
In both (10) and (11), an adverbial modifier is 
inserted at the start of the second conjunct; this 
is a common pattern extension. In (12), there is 
no real parallelism: the first conjunct clause is an 
imperative, and the second its result. 
4 Balance matching analysis model 
The balance matching analysis model determines the 
correct structure by taking advantage of the sym- 
metric patterns. In this section, first, the represen- 
tation of symmetric patterns is presented. Then the 
balance matching operation is presented. Finally, 
analysis by the balance matching is described. 
4.1 The pattern representation 
The symmetric patterns are represented by a list of 
three feature sets; Phrase features, Word features, 
and Morphological features, based on the symmetric 
pattern levels. 
* Phrase feature \[¢\]: Values are Predica- 
tive, Nominal, Nominal-Premodifier, Nominal- 
Postmodifier, Predicate-modifier. These val- 
ues are assigned to all the constituents in the 
phrase. For example, all the words in "the ef- 
fective insecticides" have ¢.(Nominal) 
• Word feature \[7\]: This feature includes 120 
values which subclasses of general parts of 
speech, according to their grammatical and 
semantic function. For example, some val- 
ues are {NounInstrument}, {NounHuman}, 
{NounAction}, {PredicateStatic}, etc. 
• Morphological feature \[6\]: The values shows 
the morphological attributes of the words, 
which are a pair of the word and the mor- 
phological type. For example, "uranium- 
235" is represented by &(uranium-235, alpha- 
bet_hyphen_arabic-n umbers) . 
Each word in the sentence is represented by the set 
of the three features. ¢ and 7 can include ambiguous 
values. The ith word-feature set Hi and the n-word 
sentence S~ are respectively represented as follows. ( ~i ) ~i -= {¢iI,''',¢im} 
= = {Ti,'",Tin} (13) 
,% Ai =  12} 
= (Wl,..-, w,) (14) 
When the conjunction is the ruth word of the n- 
word sentence, the left-side list S~ -1 and the right- 
side list S~n+l are respectively represented as follows. 
S7 '-1 = (Wx,'", W~,-,) (15) 
S~+i = (W.~+I,'", W.) (16) 
The goal of the balance matching is to find the 
most symmetric pair of 5~-1(I _< x < m) and 
,$Ym+i(m < y ~ n), i.e., to find the values of x and 
y. 
4.2 The balance matching operation 
By definition, the most symmetric pair shares the 
maximum number of the word-feature sets in the 
lists. The pair is detected by three operations: the 
intersection operation for two features, the matching 
operation for two word-feature sets, and the balanc- 
ing operation for two lists. 
4.2.1 The intersection operation 
The intersection operation is one of the normal set 
operations for the features: 
@,~_ 1. {uranium-235,alphabet_hyphen_arabic- 
numbers} 
V~ ~,~+l.{plutonium-239, alphabet_hyphen_arabic- 
numbers} - {alphabet-hyphen_arabic-numbers} 
The mutual dependency information among the 
¢ij and Vik is managed by bi-directional lists in the 
background. If all ¢ij dependent of Vik are disam- 
biguated by the operation, 7ik are removed. 
43 
4.2.2 The matching operation 
The matching operation N for the the word- 
feature sets V, 142 is defined as follows: 
v~ n Wj = V~Wj = r~ n r;, (:7) 
A~ N A d 
The matching word-feature set V~lfl2j can exist, when 
¢i rl ¢.} ~ NULL or Fin F:. ¢ NULL. 
• 21 . • Especially, as is mentioned m Sectmn 3.2, in most 
of the conjoined sentence, the word-feature sets 
of Win+l, which immediately follows the conjunc- 
tion, play an important role for detecting the struc- 
ture, because there is strong similarity between the 
starting word-feature set of the conjunct scope and 
142m+1. 
4.2.3 The balancing operation 
The balancing operation ® for the lists £~ ,T~' is 
defined as follows: 
£'~ ® n7 = {(V~Wj,"', VkW,"')IV • £i,W • nT, 
k > j, c1 = (v~,..., v,), ~7 = (w:,.-., w,)} 
Every word-feature set in the list doesn't always 
match one of the other list. Some word-feature sets 
in £\]' can find matching counterparts in 7£ T but 
others cannot. In (18), The: and operationsa re- 
spectively matches the1 and results2. 
(18) The1 arithmetics operations3 and the~ 
results2. 
Therefore, the balancing operation creates a set of 
the lists for exhausting all the possible combinations. 
The lists consist of the matching word-feature sets 
(12~I,Y~) which are selected to avoid crossing any 
existing lines when ~i and I4~ are connected by a 
line as in the following: 
(v:, ,,,, v~, ,,,, vk, ,,,, v.~) 
/ 1 1 1 \ (:9) 
(w:, ,,,, w;, ,,,, w. ,,,, w.) 
For example,£ 3 ® 7~ is : 
C 3 = (V:, \])2, V3) ~2 = 0~Vl ' }/V2 ) 
Cl 3 @ "Z~21 = {(Vl~/Vl), (Vl\]/V2), (V2Wl),'(~22~V2), 
(~23~21), (V3~)2), (~)1~71, V2~V2), (Vl~)I, V3~V2), 
(v2w~, v3w2)} 
(20) 
The balancing degree 8 for a list 2- is defined as fob 
lOWS: 
o~ = ~_, wk.V'(k) z = (v~wj,..., v~w,,) 
dV'(¢) and Af(F) respectively represent the total 
number of the each feature in the list 2-. Af(A) is the 
total number of the values of A in the list 2". Herein, 
wk is defined as a binary value (1.0 or 0.0) for sim- 
plifying the model. Through the analysis of 10,000 
English conjunctive sentences in technical manuals, 
the structures could be divided into about 300 coor- 
dinate patterns, which are represented in the form 
of the word-feature sets. We manually assigned the 
weights to the features A, F and • according to the 
patterns, in order to select the correct structure. 
4.3 Analysis by balance matching 
The n-word sentence including the conjunction as 
the ruth word is analyzed according to the following 
steps. 
L Collect the word-feature set G(W): G(W) = 
{Wx\[Wx n Wm+l # NULL} 
This step collects a set of the words similar to 
W,~+I. The collection considers some definite 
concord markers and boundary markers, such 
as "and", "both" , "either", commas, periods, 
and colons. In order to deal with the cases of 
(10),(11) and (12), the starting words of the 
clause I42 k are added to G(IA~) when a comma 
II. 
is preceded by 
Create the list 
H(£) = {Wjl i 
the conjunction. 
set H(£): 
_<j <m-l, Wi6G0d))} 
n--m. III. Create the list ~1 • 
~,.~'-"' = {W,.+:, W~.+2,"., W.} 
IV. To create the list set F(£7~) by the balancing 
operation. 
F(Cn) = {q ® nr-~lq 6 g(£)} 
V. Select the list F(£7~)ma~ which has the highest 
balancing degree from all the possible lists in 
F(£7~). 
For example, (9) are analyzed by the following 
steps. Here, for the easy understanding, the word- 
feature set and the list are represented by the sim- 
plified expression. 
(9) Radioisotopes have played an important part 
in1 developing effective insecticides in2 the 
country and in3 t~nding the best ways of ap- 
plying them• 
I. G(W) = {in:, in2} 
II. H(£) = {(in1, developing, effective, 
insecticides, in, the, country), (in2, the, country)} 
III. Tt~ = 
(inn, finding, the, best, ways, of, applying, them) 
IV. F(CTt) = {..., (Preposition, Nominal), 
(Preposition, Gerund, Nominal)} 
V. 
F( E~ ) ..... = (Preposition, Gerund, Nominal) 
= (in1, developing, effective,..., in, the, country) 
7"4 = (in3, finding, the, best, ways, of, applying, them) 
The and5 in (21) can be analyzed as the same way. 
44 
(21) Use extreme care when using cleaning solvents 
like1 acetone and methyl ethyl ketone which are 
highly flammable2 and shall be used3 in areas 
with4 adequate fire extinguishing devices and5 
free6 of ignition sources. 
G(}/V) = {like1, flammable2, used3, with4, free6} 
f(~Tl)ma~ -- (NominalPostmodi fier) 
E. = (with4, adequate, fire, extinguishing, devices) 
n = (free6, of, ignition, sources) 
5 Empirical results on the MT 
system 
5.1 MT configuration 
The model was incorporated as a balance matching 
module into the PIVOT English-Japanese MT sys- 
tem as shown in Figure.l(Muraki, 1986; Okumura et 
al., 1987; Okumura et al., 1991). The system works 
for a practical use together with the dictionary of 
100,000 words. 
The input sentences, represented by the list of 
the feature sets based on the results of the mor- 
phological analysis, are transferred to the balance 
matching module. The module produces the con- 
junct scope information as well as the results of the 
balance matching. The syntactic-semantic analysis 
module analyzes the sentences according to this top- 
down information. After the analysis, the concep- 
tual analysis module creates an interlingua repre- 
sentation. From the interlingua, output sentences 
are generated(Muraki, 1986; Okumura et al., 1991). 
5.2 Effects of the model 
When 15,000 conjunctive sentences in the technical 
reports and manuals were translated apart from the 
analyzed 10,000 sentences, the following effects are 
confirmed about the model. 
1. Reduction of the analysis cost: This model pro- 
vides the correct top-down information about 
the conjunct scopes for about 75~0 of the sen- 
tences, which result in the accurate and effective 
syntactic analysis. Most of the sentences had 
required backtracking for the analysis without 
the model. The backtracking is almost all sup- 
pressed by the model. 
2. Improvement of the word disambiguation: The 
results of the balance matching improve word 
disambiguation and the inferences of the un- 
known words, because the ambiguities of each 
word are intersected by the counterparts of the 
symmetric list. The results provide top-down 
information for the analysis of the ambiguous 
words and unknown words, as in (8). Most of 
the sentences contained some word ambiguities 
and one sixth contained unknown words. By 
using the top-down information, the accuracy 
was twice better improved. 
. 
. 
Interpretation of the ellipses: The model makes 
it easy to interpret the elided elements, because 
the balance matching results can suggest the 
missing elements. In sentences such as (1), T~ 
is completely included £:. The differences of £: 
and 7~ complement the missing elements. 
Robust analysis: The model helps make the sys- 
tem robust because the balance matching op- 
eration is based on the three different kinds of 
features. In the MT domain of technical reports 
and manuals, there are some unknown words as 
well as some ambiguous words as in (6),(7) and 
(8). Robustness is achieved because the mor- 
phological features are considered as well as the 
other features. 
5.3 Discussions 
To increase the accuracy, the model is improved from 
three points: 
• Lexical disambiguation: The model is based 
on lexical information. Therefore, when many 
words provide too many feature ambiguities, 
the model cannot always determine a correct 
structure. In order to solve this problem, filter- 
ing rules are applied to the sentence before the 
balance matching operation. The rules are local 
constraint rules, which checks the two or three 
words before a focused word to remove some 
ambiguities of the focused word. The filtering 
rules improve the model. 
• Weight optimization: The weights for each fea- 
ture set are manually assigned based on 300 pat- 
terns. They should be more appropriately as- 
signed as real values instead of binary values ac- 
cording to the domain and text styles. We have 
developed a learning method for the feature 
structures(Okumura et al., 1992). The method 
is applicable for determining the weights ac- 
cording to the input patterns. 
• Semantic calculation: Some conjunctive struc- 
tures should be analyzed by the more subdi- 
vided semantic features and semantic similarity 
calculation. We are introducing some semantic 
taxonomy and the semantic distance measure- 
ment algorithm(Knight, 1993; Okumura and 
Hovy, 1994; Resnik, 1993). 
6 Conclusion 
The authors propose an English coordinate struc- 
ture analysis model, which provides top-down scope 
information of the correct syntactic structure by tak- 
ing advantage of the symmetric patterns of the par- 
allelism. 
This model was practically implemented and in- 
corporated into the PIVOT English-Japanese MT 
system with the dictionary of 100,000 words. The 
model provided about 75% accuracy, which lead to 
45 
input sentences ) 
I morphological \[_J syntactic-semantic I 
analysis I "1 analysis I 
T \[ balance matching module I 
output sentences ) 
I conceptual ~interlingua ~ sentence ' analysis generation 
Figure 1: A MT system configuration including the balance matching module 
robust analysis for the MT system, as well as helping 
make rule-based analysis effective. In the future, we 
plan to improve the model according to the "above 
discussion, and also to extend the analysis to other 
conjunctions and prepositions. 
7 Acknowledgments 
The authors thank Eduard Hovy and Kevin Knight 
of USC/ISI for their precious comments. 

References 

R.. Beaugrande and W. Dressier. 1981. INTRO- 
D UCTION TO TEXT LING UISTICS. Longman 
Group Limited, Harlow, Essex. 

S.A. Blackwell. 1981. Processing conjunctions in an 
atn parser. Technical report, University of Cam- 
bridge. Unpublished M.Phil. Dissertation. 

B.K Boguraev, K. Sparck-Jones, and Y.(eds) Wilks, 
editors. 1983. Recognizing Conjunctions within 
the ATN Framework. Ellis Horwood. 

S. Fong and R.C. Berwick. 1985. New approaches to 
parsing conjunctions using prolog. In Proceedings 
of the 23rd ACL Conference, Chicago. 

G. Gazdar. 1981. Unbounded Dependencies and Co- 
ordinate Structure. 

X. Huang. 1983. Dealing with conjunctions in a 
machine translation environment. In Proceedings 
of European ACL Conference. 

X. Huang. 1984. Dealing with conjunctions in a 
machine translation environment. In Proceedings 
of COLING 84. 

Kevin Knight. 1993. Building a large ontology for 
machine translation. In Proceedings of the ARPA 
Human Language Technology Workshop, Princeton, New Jersey. AR.PA. 

D. Kosy. 1986. Parsing conjunctions deterministically. In Proceedings of the 24th ACL Conference. 

S. Kurohashi and M. Nagao. 1992. Dynamic 
programming method for analyzing conjunctive structures in japanese. In Proceedings of COLING-92, August. 

L. Lesmo and P Torasso. 1985. Analysis of conjunctions in a rule-based parser. In Proceedings of the 23rd ACL Conference, Chicago. 

K. Muraki. 1986. Venus: Two-phase machine translation system. Future Generations Computer Systems 2. 

S. Niimi, Y. Suwa, J Tsujii., and .M Nagao. 1986. 
Coordinate sentences by mu-project english anal- 
ysis grammar. In Proceedings of Information Pro- 
cessing Society of Japan. in Japanese. 

A. Okumura and E. Hovy. 1994. Building a japanese 
dictionary associated with an ontology for ma- 
chine translation. In ARPA Human Language 
Technology Workshop. 

A. Okumura, S. Kamei, and K. Muraki. 1987. En- 
glish coordinate structure analysis. In Proceed- 
ings of Information Processing Society of Japan, 
September. in Japanese. 

A. Okumura, K. Muraki, and S. Akamine. 1991. 
Multi-lingual sentence generation from the pivot 
interlingua. In Proceedings of MT SUMMIT III, 
July. 

A. Okumura, K. Muraki, and K. Yamabana. 
1992. A pattern-learning based, hybrid model 
for the syntactic analysis of structural relation- 
ships among japanese clauses. In Proceeding of 
Fourth International Conference on the Theoreti- 
cal and Methodological Issues in Machine Trans- 
lation, June. 

P. Resnik. 1993. Semantic classes and syntactic am- 
biguity. In Proceedings of ARPA HLT Workshop. 

P. Schachter. 1977. Constraints On Coordination. . 

Y. Shinoda. 1981. TECHNICAL ENGLISH. Nagu- 
modou, Tokyo Japan. in Japanese. 
