ROBUST METHOD OF PRONOUN RESOLUTION 
USING FULL-TEXT INFORMATION 
Tetsuya Nasukawa 
IBM Research, Tokyo Research Laboratory 
1623-14, Shimotsuruma, Yamato-shi, Kanagawa-ken 242, Japan 
Abstract 
A consistent text contains rich inforlnation for resolv- 
ing mnbiguities within its sentences. Even simple syn- 
tactic information such as word occurrence and col- 
location patterns, which can be extra(:ted fi'om the 
text without deep discourse analysis, lint)roves the 
accuracy of sentence analysis. Pronoml resolution 
is a typical proceeding that utilizes this information. 
Through the use of this inforlnation, along with in- 
formation on the syntactic position of each candidate, 
93.8% of pronoun references were resolved correctly 
in an experiment on computer mammls. 
1 Introduction 
Resolving t>ronoun reference is a diifieult task that 
requires consideration of both linguistic and cogni- 
tive aspects of a language. As a linguistic phe- 
nomenon, the use of pronouns is treated a.s a co- 
referential prohlem in which both the antecedent and 
the pronoun co-refer to some object. From this point 
of view, finding the object that is co-referred to by 
a pronoun is the main problem in t)ronoun resolu- 
tion, and much research has been devoted to focus- 
ing on or inferring the referent object by consider- 
ing the grammatical and semantic roles of each en- 
tity in the sentences \[Sidner, 1983; Brennan, 1987; 
Kameyama, 1993\]. This task is especially difficult 
when the referent object is not explicitly stated in 
a text, and common sense and deep inference are re- 
quired in order to figure out the object, sLs in the clas- 
sic problems descrit)ed by Charniak \[Charniak, 1973\]. 
Since this approach of considering the grammatical 
and semantic roles of each entity depends heavily on 
accurate syntactic anMysis avd semantic analysis, it 
is not yet applicable to practicM systems. 
However, if we do not aim for perfect analy- 
sis, a simple syntactic-based heuristic rule for s- 
electing a correct antecedent from several candi- 
date noun phra~ses i)erforms quite well, especiMly 
in technical documents such as computer manual- 
s, in which we can usuMly expect an explMt an- 
tecedent within the same sentence or in a previous 
sentence. In this domain, a correct ante(:edent can 
be selected in ahnost 90% of all causes without any 
world knowledge other than simple semantic con- 
straints \[Hobbs, 1978; Walker, 1989; Lappin, 1990\]. 
Moreover, several heuristic rules can be combined 
to improve the accuracy of the analysis \[Rich, 1988; 
Carboi,ell, 1988\]. 
This approach of resolving pronoun reference by 
applying simple heuristic rules seeins to be ade.quate 
for a practical naturM language processing system, 
yet in order to achieve a success ratio of over 90%, 
some kind of knowledge processing is required, such 
as the use of world knowledge or deep inference mech- 
anisms for constructing and referring to a discourse 
structure. While the advantage of knowledge pro- 
cessing is widely recognized, this approach presup- 
poses a large quantity of knowledge resources, and 
leads to a knowledge acquisition bottleneck. In or- 
der to solve this problem, wmous studies have been 
done on methods of using on-line text databa~ses with 
less human intervention for word sense disambigua- 
tion attd structural disambiguation \[Jensen, 1987; 
Nagao, 1990; Uramoto, 1991; Hindle, 1993\]. These 
methods can be applied to knowledge processing in 
pronoun resohltion; however, no research has yet re- 
veMed sutficient world knowledge to cover general 
1)roblems. In other words, methods of using world 
knowledge have not reached a level sufficiently ma- 
ture for them to be used in broad-coverage systems. 
This paper prol)oses a simple and robust approach 
that utilizes inte.r-sentential information, extracted 
from a source text by means of a simple algorithm, 
to improve the accuracy of pronoun resolution. For 
example, (:olloeation patterns within a text offer in- 
formation that eorrcsl)onds to ease frames in world 
knowledge, and word frequency also gives informa- 
tion reh!wmt to the tot)it or focus of the subjects. 
Thus, instead of using outside knowledge resources, 
such information serves as world knowledge appropri- 
ate to the narrow domain of the source text. The ef- 
fectiveness of each type of information extracted from 
a source text is evaluated in the light of the results of 
experiments on comlmter lnanuals. 
In the next section, we introduce three effective 
factors in the selection of an antecedent from candi- 
date noun phra.ses. Then, in the third section, we 
1157 
specify the implementation of this method. Finally, 
in the fourth section, we evaluate the effectiveness of 
this approach on the basis of the results of an exper- 
iment. 
2 Three factors for evaluating 
salience in candidates 
In our approach, pronoun resolution basically consist- 
s of collecting candidate noun phrases and selecting 
the most preferable candidate as the antecedent of 
a pronoun by applying several rules to filter out in- 
appropriate candidates and to attach preferences to 
appropriate candidates. Rules are divided into two 
types. One type represents grammatical constraints 
that must be satisfied, such as number and gender 
agreement. Since rules of this type can filter out i- 
nappropriate candidates, we apply them at an ear- 
ly stage of pronoun resolution. The remaining rules 
constitute the other type, which attaches a preference 
to each candidate noun phrase. After inappropriate 
candidates have been filtered out by the former rules, 
the latter rules determine the most appropriate can- 
didate by measuring the sMience of each remaining 
candidate noun phrase. Thus, the latter rules are im- 
portant for selecting the exact antecedent from the 
remaining candidates and for improving the accuracy 
of pronoun resolution. 
In this section~ we describe three effective factors 
that utilize full-text information for measuring the 
salience of each candidate noun phrase. The reasons 
for their effectiveness are that they cover many as- 
pects of linguistic phenomena and that their inter- 
pretation is simple enough to be used in a practical 
system. 
2.1 Collocation patterns within a 
source text 
In previous approaches, semantic constraints have 
been among the most basic factors for filtering out 
candidates that would be inappropriate as modifier- 
s of the modifiee of a pronoun. However, in order 
to apply semantic constraints with broad coverage, a 
large amount of knowledge is required. For example, 
in processing a sample sentence provided by Hobbs 
\[Hobbs, 1978\], 
The castle in Camelot remained the resi- 
dence o/ the king until 536 when he moved 
it to London, 
the following knowledge must be supplied in order to 
filter out the candidates 536, castle, and Camelot, and 
leave the correct antecedent, residence: 
• Dates cannot move. 
• Places cannot l~love. 
• Large fixed objects cannot move. 
In order to apply this knowledge, we also prcsul)pose 
a correct analysis that categorizes each nmm phrase 
as a date, place, large fixed object, and so on. Since 
many of the words in these noun phrases have word 
sense ambiguities, it is not practical to presuppose 
the correct application of such knowledge. Assem- 
bling a large body of knowledge poses another major 
problem. 
instead of such worhl knowledge, collocation pat- 
terns (namely, modifiee-modifier relationships) ex- 
tracted from a discourse can be applied. Since word 
sense is usually unified within a discourse, and most 
words with the same lemma are frequently repeat- 
ed \[Gale, 1992; Nasukawa, 1993\], the collocation pat- 
terns in the same discourse provide valuable data 
for determining whether a candidate can modify the 
modifiee of Ct pronoun. For example, if the sentence 
He moved his residence 
is found in the discourse, this information indicates 
that the word residence can be the object of the verb 
move. Thus~ the inh)rmation works as a selectional 
constraint that the candidate can be an argument of 
a predicate that dominates the pronoun. 
Moreover, since statements tend to be repeated 
in a discourse, the existence of an identical colloca- 
tion pattern in a discourse may support selection of 
the candidate as the antecedent. In this sense, the 
preference for an identicM collocation pattern also 
reflects case role persistence preference and syntac- 
tic parM1elism preference, prolmsed by Carbonell and 
Brown \[Carbonell, 1988\]. The case role persistence 
rule prefers a candidate noun phrase that filled an i- 
dentical case role in an earlier sentence. For example, 
after the sentence 
Mary gave an apple to Susan, 
Susan is referred to by her in 
John also gave her an apple, 
while Mary is referred to by she in 
She also gave John an apple. 
The syntactic 1)arM1elism rule prefers a candidate 
noun phrase that preserves its surface syntactic role 
from the first of two or more coordinate clauses. For 
example, in 
7158 
The girl scout leader paired Mary with. Su- 
san, but she had paired her with Nancy 
last time, 
she refl,'rs to leader, and her refers to Mary, 
whereas in 
The girl scout leader paired Mary with Su- 
san, but she had paired Nancy with her 
last time, 
she refers go leader, ait(\[ her refers to ,S'usan. By re- 
ferring to the i(lenticM collocation Iiatterns, we (:~ut 
resolve all the pronouns in the above examples (:or- 
rectly. 
Since the idcntitication of modifier-modifiee rela 
tionshii)s is a basic feature of syntactic analysis, a pro- 
cedure for identifying identical collocation patterns is 
not a hard task, as long as M1 of the sentences are 
parsed by a single system and share a single l'orm~fl- 
ism for expressing modifier-modiiiee relationships. 
2.2 l~'equency of repetition in preced- 
ing sentences 
A characteristic of the i)ronolnimflization on whi(:h 
the centering apln'oach \[Sidner, 1983; Ilr(,mlan, 1987; 
Kameyama, 1993\] is l);Lsed is that an object in focus 
is likely to hc pronomimtlized. If this characteristic 
is expanded to all definite anaphoras, which include 
detinite noun phrases as well as pronouns, a candidate 
noun phrase that is in focus inay be repeated as a def- 
inite noun phrase before it is pronomiilalized. Thus, 
the frequen(:y in 1)receding sentences of a nouu i)hrase 
with the same lemma ~s ~t candidate noun l)hr~use (:aa 
be an index for the preference with whi(:h it is selected 
as the antecedent. The 1)roeess for a.ssigning this 1)ref- 
erence consists of a simple string match that cheeks 
words with the same lcmma in preceding sentences. 
In addition, when the source text is marked up 
with SGML or other su(-h tags, the roles of some 
phrases such as titles mid headings call he easily re(> 
ognized, and words with such roles tend to represent 
the topics of the sentences fifth)wing them. Thus, ~(\[- 
ditional preference can be assigned by checking the 
tags of each word. 
2.3 Syntactic position 
As shown by tIobtis \[ltobl)s, 1978\], a heuristic t'ule fa- 
voring subjects over objects l)erforlns remarkably well 
in English text. By traversing th(' surface parse tree 
of a sentence, a 1)referen(:e vMue can be provided for 
each candidate noun phrase according to its syntactic 
position. This factor has mt advantage over other fa(> 
tors shown in 1)revious subsections in the sense that 
it: assigns a definite ranking for ea(:h candidate noun 
I)hrase, since ea(:h o(:eul)ies a syntactic position in a 
text. Thus, this factor l)rovides a default value for 
the l)reference of ea(;h (:an(lidate itOUlt t)hrase when 
no oth('r factor provides wfiid information, and it is 
ad('(luat(~ for a r(It)ust approach since it is basically 
assigned t)y traversing tho surface l)arse trees of a 
sen t.enee. 
3 Implementation 
In this sect;ion, we describe the actual imple- 
lnelltation of (;he l)rottoUll resohltiott procedllre ill 
ml Fmglish-to-Japanese. machine translation system, 
Shalt2 \[Takeda, 1992\]. 
The procedure consists of two steps: 
1. Extraction of ca.ndidat(:s for an mltec(,dent 
2. Selecti(m of I;he correct antece(tent fl'om the 
candidates. 
\[11 or(let t(/ achieve ttigh(!r ac(;llracy itl l)rolloull res- 
olution with robust lirocessing, our straeegy consists 
of 
1. Extending a list of ('andidate noun ptm~scs so 
tht~t it does noL ex(:lude a correct antecedent 
2. \]leferring t() all information in the source text 
that can be syntactically extracted without re- 
ferring to outside knowledge resources, in orde, r 
to sele(:t the corre('.t ante(:e(lent. 
3.1 Extraction of candidates 
'fo ensure that the correct antecedent is in(:luded in 
a list of candidate noun t)hrases, candidates are ex- 
tra(:ted from exa(:tly two sentences with lninimum til- 
tering. First, the system cheeks whether any noun 
tihrases earlier in the same sentence satisfy the num-- 
ber and gender constraints. It then checks the t)reced - 
ing sentences in order of proximity until candidates 
have bct}ll YOlllI(\[ ill exactly two sentell(:es. 
l)uring the extraction of the candidates, the sys- 
tem filters out noun phrases that do not satisfy the 
numl)er and gender constrmnts, and Mso direct mod- 
ifiees of the pronoun and its arguments, so that a 
noll-reIlexive 1)rOllOlllt all(t its anl;eced(mt nlay not oc- 
cur in I:he same siml)lex sentence, as would be the 
case if data were the antecedent of it in the following 
sentenc(~; 
The device that writes onto a magnetic 
di.sk and reads data from it is called a disk 
drive. 
1159 
3.2 Selection of an antecedent 
In our implementation, the preference values provid- 
ed by the algorithms described in the following para- 
graphs are combined into a single value, and the can- 
didate noun phrase with the largest preference value 
is selected as the antecedent. 
Preference according to the existence of iden- 
tical collocation patterns in the text 
As a preference value that indicates the satisfaction 
of selectional constraints and repetition of an iden- 
tical statement, we assigned a constant value 3 for 
a candidate that has an identical collocation pattern 
with the modifiee of a pronoun within the source tex- 
t. Furthermore, in order to extend the use of colloca- 
tion patterns as knowledge on seleetional constraints, 
an on-line synonym dictionary \[3\] is referred to, and 
thus a collocation pattern with a synonym can sup- 
port candidates other than exactly identical colloca- 
tion patterns. 
Preference according to the frequency of rep- 
etition in preceding sentences 
In order to provide a larger preference value for clos- 
er and more frequent occurrences of a lemma, the 
preference value is given by the total score calculated 
according to the following formula, for each appear- 
ance of a noun phrase with the same lemma as tim 
candidate noun phrase that is found within the ten 
preceding sentences: 
1 
(Number of sentencea to the identical noun phrase)+1. 
Preference according to syntactic position 
Among the candidate noun phrases, a candidate in a 
closer sentence, or the one nearest the beginning of 
the same sentence is preferred. Besides the left-to- 
right order within a sentence, a negative preference 
value is given for tile distance (number of sentences) 
from the sentence that contains the pronoun to the 
sentence that contains the candidate. While the order 
of preference of candidates that is obtained in this 
manner is similar to that given by the naive algorithm 
proposed by Hobbs \[Hobbs, 1978\], our algorithm is 
much simpler, and does not even require the results 
of syntactic analysis. 
3.3 Example 
Figure 1 gives an example of system output that 
contains data on the preference of each candidate an- 
tecedent for a pronoun in a sample text extracted 
f,'om the second chapter of a computer manual \[2\] in 
the mam, er described in the previous paragraphs. 
In this figure, the number in brackets before each 
sentence indicates the sentence number in the text. 
As shown by these numbers, the output consists of 
eleven consecutive sentences, front the 104th to the 
114th in the second chapter of the manuM.* The or- 
der of candidates following the message 
Candidates for the referent of CFRAME106579 
("it") are : 
reflects tile order of preference values obtained by re- 
ferring to the positimt of each candidate. As shown 
in this list, key is the most preferable candidate from 
the viewpoint of syntactic position. In this candi- 
date list, CFRhMEwuwo indicates an instance of each 
content word in the dis(:ourse. Information on the 
position and on tile whole sentence can be extract- 
ed from each of these CFlt.AMEs. A number in arrow- 
head brackets next to CFRAMEu~uu~u, such as <ll3>, 
indicates the number of the sentence in which it oc- 
curs. A number in parentheses, such as 0.48571432 
in key (0.48871432), indicates the preference value 
obtained by referring to the frequency of repetition. 2 
Thus, from the viewpoint of the number of repeti- 
tions, cursor is the most preferable candidate for the 
aI, tecedent. At the bottmn of this figure, information 
on modifier-modifiee relationships that support can- 
didates is shown. In this case, there is a collocation 
pattern such that cursor modifies the verb reaches, 
which is the modifiee of tile pronoun it; tiros, this 
information prefers cursor for the antecedent of it. 
Finally, after combination of all the preferences, cur- 
sor is selected as the most preferable antecedent of 
it. 
4 Results 
We. examined 112 third-person pronouns in 1904 con- 
secutive sentences fl'om eight chapters of two differ- 
ent computer manuals. One \[1\] is a typical computer 
manual for computer experts such as programmers 
and system operators, and the other \[2\] is a primer 
for novice users of a computer. 
In this experiment, we excluded instances in which 
it pronmninalizes a sentence, as in 
do it, 
those in which it refers to a syntactically recoverable 
that-clmlse or to-infinitive clause, and those in which 
~In the sentences contained in Figure 1, the underlining and 
tile change of font for the target pronoun it were done by tile 
attthor. 
2Noun phrases with the same lemma referred to in the pre- 
ceding sentences are indicated by underlines. 
1160 
(104) 
(105) 
(lO6) 
(lO~) 
(108) 
(109) 
(11o) 
(111) 
(112) 
(11a) 
O14) 
All four of the cursor movement keys are typematic; 
they kee t) repeating as long ~s they are held down. 
The Cursor Up key moves the cnrsor up one line. 
Like the other cnrsor movement keys, this key moves the cursor one line or many 
lines depending on how long you hold down the key. 
The Cursor Right key moves the cursor to the right. 
Hoht the key down. 
When the cursor reaches the right end of the line, it goes off the screen and 
reappears on the left side, one line bdow the line it was on. 
If the cursor is on the bottom line of the screen and is run M1 the way to the right, 
it goes off the screen and reappears in the upper left corner. 
The Cursor Left key mow's the cursor one position to the left. 
1told this key down. 
When it reaches the left end of tile line, it goes off the screen and reappears on 
tile rigtlt, one line above the line it was on. 
Candidates for the referent of CFRAMEIO6579("it") are: 
CFRAMEI06573<113> .... key (0.48571432) 
CFKAMEI06564<I12> .... Cursor Left key (0) 
CFKAME106565<lli> .... cursor (1.4337664) 
CFKAME106568<l12> .... left (0) 
>> With DIANA << 
>> To support CFRAMElO6565(cursor) << 
i with SAME-ATTACHMENT-CAND-MODIFIEE in: 
" When the cursor reaches the right end of the line, 
it goes off the screen and reappears on the left side, 
one line below the line it was on ." 
(reaches<CFRhME106454> in SENTENCEIO6453<No. IIO>) 
Fig. 1: Saml)le data h)r resolving the first pronoun it in sentence 114 
it occurs in a time or weather construction. When an 
identical pronoun is included in the candidate list, the 
system assmnes that these pronouns share tile same 
antecedent. For example, we assumed that M1 the 
instances of it in 
When it reaches the left end of the line, 
it goes off the screen and reappears on the 
right, one line above the line it was on 
have the same antecedent. 
As a result of our strategy of enlarging the scope 
for selection of cmtdidates, tile average nuinber of can- 
didate noun phrases was 4.1. 
Our algorithm chose a correct antecedent in 105 
(:a.ses, giving a success ratio of 93.8%. In 28 of those 
112 cases, there was among the candidates an iden- 
tical t)ronoun that; referred to the same anteceden- 
t; thus, in 84 cases, antecedents were selected by e- 
valuating the syntactic position, frequency of repeti- 
tion, and collocation pattern of each candidate noun 
phrase. 
As shown in Table 1, without any information on 
repetition or collocation patterns, the success rate of 
selection based only on syntactic position was 82.7%, 
while tile success rate for selection based only on fre- 
quency of repetition was 60.7%. This result indicates 
that pronominMized noun phrases were actually re- 
peated more than twice within ten consecutive sen- 
tences in over 60% of the cases. Thus, preference 
according to the frequency of repetition contributed 
to the selection of the correct antecedent. In 16 of the 
22 cases in which this information preferred a wrong 
candidate noun phrase, the preference value was over- 
ridden by the negative preference value caused by a 
syntactic position far from tile sentence of the pro- 
noun, or by a larger preference assigned to some other 
candidate with an identicM collocation pattern. 
Identical collocation patterns were found within 
the same chapter in 22 of 84 cases in which the pref- 
erence value was ev',duated in selecting all anteceden- 
t. Although this is only 26.2% of the cases, collo- 
cation preference (lid not support any wrong candi- 
dates. Moreover, in 50.0% of the 22 cases, another 
preference value, either syntactic position or repeti- 
tion, supported a wrong candidate. Therefore, pref- 
erence according to collocation pattern contributed 
to the selection of the correct antecedent. 
Table 2 shows the distances and directions of sell- 
1161 
Table 1: Correlation between correct selection and selection in accordance with each type of pret~rence 
Number of c,~ses 
in which tile correct 
antecedent was selected 
Number of cases 
in which the wrong 
antecedent was selected 
Nulnber of cases 
without any 
valid information 
Syntactic 69 15 0 
position (82.1%) (17.9%) 
Frequency of 51 22 11 
repetition (60.7%) (26.2%) (13.1%) 
Existence of 22 0 62 
sinfilar collocation (26.2%) (0%) (73.8%) 
Table 2: Distribution of information relative to a sentence that contains information for modification preference 
Forward 
Backward 
Distance (number of sentences) 
Number of occurrences 
Distance (number of sentences) 
Number of occurreuces 
10&-4 5-L I 4 - 2 
1-51  10 I i11  116-20 21 
1211 I 0 \[ 2 0 
tences in which a collocation pattern supporting a 
candidate noun phrase to modify the modifiee of tile 
pronoun was found. The results indicate that such in- 
formation was extracted from a relatively small area 
of a text. In addition, relative collocation patterns 
were extracted from both previous and following sen- 
tences. 
Out of the 37 cases in which the identical collo- 
cation patterns were found, synonym relations were 
used in seven cases (18.9%). 
5 Conclusion 
We have proposed a robust method of pronoun res- 
olutimL that refers to information within tile source 
text in order to determine the preference value of each 
noun phrase that is a candidate for selection as the 
antecedent of a pronoun. This approach is practical 
in terms of the amount of knowledge it presupposes 
and the amount of computation it requires, since it 
basically relies only on the surface information in a 
text, and is fl'ee fi'om the knowledge acquisition bot- 
tleneck. 
In experiments on computer manuals, we achieved 
a success rate of 93.8%. A remarkable ~uspect of this 
result is that we achieved it without referring to any 
outside knowledge resource except for the synonym 
relations in an on-line synonym dictionary. By com- 
bining heuristic rules to utilize wtrious information 
extracted from M1 the sentences in the source text, 
high accuracy can be achieved in pronoun resolution 
for a practical naturM language processing system. 
Tilt advantages of this approach are that a simple 
algorithm can extract information on syntactic posi- 
tion, repetition, and collocation pattervs hy referring 
to morphological informatiml within a source text, 
and that it does rot even assume a correct syntac- 
tic analysis or depend ov the formalism of syntactic 
parse trees, since it does not rely on any grammatical 
information except for modifier-nmdifiec relationship- 
s. This at)proach is especially effective in technical 
documents such as computer manuMs or patent doe- 
uments in which words are use, d consistently in order 
to avoid ambiguity, and in which identical collocation 
patterns are frequently repeated in detailed descrip- 
tions of target objects or procedures. 
Acknowledgements 
I would like to thank Michael McDonald for invalu- 
able hell) in proofreading this paper. I would also 
like to thank Taijiro Tsutsumi, Masayuki Morohashi, 
Koichi Takeda, tIiroshi Maruyama, Hiroshi Nomiya- 
ma, Hideo Watanabe, Shiho Ogino, Naohiko Uramo- 
to, and the anonymous reviewers for their commeuts 
and suggestions. 
References 
\[Brennan, 1987\] Brennan, S. E., M. W. Friedman, 
and C. J. Pollard (1987). A Centering Approach 
to Pronomls. /:n Proceedings of A CL-8Z 
I'/62 
\[Carbonell, 1988\] Carbonell, J. G., and l.I.. D. 
Brown (1(,)88). Anaphora Resolution: A 
Multi-Strategy At)proach. in Proceedings of 
COLING-88. 
\[Charniak, 1973\] Char,dak, E. (1973)..lack and 
Janet in Search of a Theory of Knowledge. /n 
Proceedings of IJCAI-72. 
\[GMe, 1992\] Gale, W. A., K. W. Church, a.ltd D. 
Yarowsky (1992). One Sense Per l)iscourse. /n 
Proceedings of the 4th DARPA 5'pecch and Nat- 
ural Language Workshop 
\[Hindle, 1993\] Hindle, D., ~md M. Rooth (1993). 
Strnctural Ambiguity and Lexieal Relations. 
Computational Linguistics, Vol. 19, No. 1. 
\[nobbs, 1978\] Ilobbs, J. R. (1978). Resolving Pro- 
noun l/,eferenees. Lingua, 4~' 
\[Jensen, 1987\] Jensen, K., and J.-I,. Binot (1987). 
Disambiguating Prel)ositionaJ Phrase Attach-. 
ments by Using ()n-Line Dietioimry 1)eiinitions. 
Computational Linguistics, Vol. 13, No. 2- 4. 
\[Lal)pin , 1990\] Lapi)in , S., and M. McCord (1990). 
Anaphora ll~esolution in Sh)t Grammar. Com- 
putational Linguistics, Vol. 16, No. 4. 
\[Kameyama, 1993\] Kameyama, M., R. Passonneau, 
and M. I)oesio (1993). Temporal Centering. In 
Proceedings of A CL- 93. 
\[Nagao, 199(I\] Nagao, K. (1990). Del)endency Am 
alyzer: A Knowle(lge-ba.sed A1)proaeh to 
Structural Disambiguation. \[n Procecding.s of 
COLING-90. 
\[Nasuk~wa, 1993\] iasukawa, T. (1993). l)iscourse 
Constraint in Computer Manuals. In Proceed- 
ings of TM\[-93. 
\[Rich, 1988\] Rich, E.A. and S. Lul)erFoy (1988). An 
Architeeturc for AnaI)ttora Resolution. In Pro- 
ceedings of ANLP-88. 
\[Sidner, 1983\] Sidner, C. L. (1983). Focusing in the 
Coinprehension of Definite Anaphora. In Com- 
putational Models of Discourse, M. Brady and 
R. Bcrwick, cds., Cambridge, Mass.: MIT 
Press. 
\['\['~keda, 1992\] Ta.keda, K., N. Uramoto, T. Nw 
sukawa, and T. Tsutsmni (1992). Shalt2 - 
a Syimuetric Machine Translation System 
with Conceptual Transfer. In Proceedings of 
COLING-92. 
\[Uramoto, 1991\] Uramoto, N. (J991). Lexical and 
Structural Disambiguation Lrsing all Example- 
Base./n Proceedings of the 2nd Japan-Australia 
Joint Symposium on Natural Language Process- 
ing. 
\[Walker, 1989\] Walker, M. (1989). Evaluating Dis- 
course Processing Algorithms. In Proceedings of 
A CL-89. 
\[1\] "IBM SAA bnagePlus Object Distribution 
Manager MVS/ESA High-Speed Capture Sub- 
system Guide Version 2 Release 1.1," IBM Cor- 
1, (1991) 
\[2\] "IBM Application System/400 New User's 
Guide Version 2," IBM Corp. (1992) 
\[3\] "The New Collins Thesaurus," Collins Publish- 
ers, Glasgow (1.984) 
1163 
