Resolution of Indirect Anaphora 
in Japanese Sentences 
Using Examples "X no Y (Y of X)" 
Masaki Murata Hitoshi Isahara 
Communications Research Laboratory 
588-2, Iwaoka, Nishi-ku, Kobe, 651-2401, Japan 
{murat a, i sahara}¢crl, go. j p 
TEL: +81-78-969-2181, FAX: +81-78-969-2189 
Makoto Nagao 
Kyoto University 
Sakyo, Kyoto 606-8501, Japan 
nagao~pine, kuee. kyot o-u. ac. j p 
Abstract 
A noun phrase can indirectly refer to an entity that 
has already been mentioned. For example, "I went 
into an old house last night. The roof was leaking 
badly and ..." indicates that "the roof' is asso- 
ciated with "an old house", which was mentioned 
in the previous sentence. This kind of reference 
(indirect anaphora) has not been studied well in 
natural language processing, but is important for 
coherence resolution, language understanding, and 
machine translation. In order to analyze indirect 
anaphora, we need a case frame dictionary for nouns 
that contains knowledge of the relationships between 
two nouns but no such dictionary presently exists. 
Therefore, we are forced to use examples of "X no 
Y" (Y of X) and a verb case frame dictionary in- 
stead. We tried estimating indirect anaphora using 
this information and obtained a recall rate of 63% 
and a precision rate of 68% on test sentences. This 
indicates that the information of "X no Y" is use- 
ful to a certain extent when we cannot make use 
of a noun case frame dictionary. We estimated the 
results that would be given by a noun case frame 
dictionary, and obtained recall and precision rates 
of 71% and 82% respectively. Finally, we proposed 
a way to construct a noun case frame dictionary by 
using examples of "X no Y." 
1 Introduction 
A noun phrase can indirectly refer to an entity that 
has already been mentioned. For example, "I went 
into an old house last night. The roof was leaking 
badly and ..." indicates that "The roof' is associated 
with "an old house," which has already been men- 
tioned. This kind of reference (indirect anaphora) 
has not been thoroughly studied in natural language 
processing, but is important for coherence resolu- 
tion, language understanding, and machine trans- 
lation. We propose a method that will resolve the 
indirect anaphora in Japanese nouns by using the 
relationship between two nouns. 
When we analyze indirect anaphora, we need a 
case frame dictionary for nouns that contains infor- 
mation about the relationship between two nouns. 
For instance, in the above example, the knowledge 
that "roof" is a part of a "house" is required to an- 
alyze the indirect anaphora. But no such noun case 
frame dictionary exists at present. We considered 
using the example-based method to solve this prob- 
lem. In this case, the knowledge that "roof" is a part 
of "house" is analogous to "house of roofi" There- 
fore, we use examples of the form "X of Y" instead. 
In the above example, we use linguistic data such as 
"the roof of a house." In the case of verbal nouns, 
we do not use "X of Y" but a verb case frame dictio- 
nary. This is because a noun case frame is similar to 
a verb case frame and a verb case frame dictionary 
does exist. 
The next section describes a method for resolving 
indirect anaphora. 
2 How to Resolve Indirect Anaphora 
Anaphors and antecedents in indirect anaphora 
have a certain relationship. For example, "yane 
(roof)" and "hurui ie (old house)" are in an indi- 
rect anaphoric relationship which is a part-of rela- 
tionship. 
sakuban aru hurui ie-ni itta. 
(last night) (a certain) (old) (house) (go) 
(I went into an old house last night.) 
yane-wa hidoi amamoride ... 
(roof) (badly) (be leaking) 
(.The roof was leaking badly and ... ) (1) 
When we analyze indirect anaphora, we need a dic- 
tionary containing information about relationships 
between anaphors and their antecedents. 
We show examples of the relationships between 
anaphors and antecedents in Table 1. The form of 
Table 1 is similar to the form of a verb case frame dic- 
tionary. We would call a dictionary containing the 
relationships between two nouns a noun case frame 
dictionary but no noun case frame dictionary has 
yet been created. Therefore, we substitute it with 
examples of "X no Y" (Y of X) and with a verb case 
frame dictionary. "X no Y" is a Japanese expression. 
It means "Y of X," "Y in X," "Y for X," etc. 
31 
Table 1: Relationshi 
Anaphor 
...kazoku (family) 
kokumin (nation) 
genshu (the head of state) 
yane (roof) 
mokei (model) 
gyouji (event) 
jinkaku (personality) 
kyouiku (education) 
'kenkyuu (research) 
kaiseki (analysis) 
)s between anaphors and their antecedents 
Possible antecedents 
hito (human) 
kuni (country) 
kuni (country) 
tatemono (building) 
seisanbutsu (product) 
\[ex. hikouki (air plain), hune (ship)\] 
Relationship 
belong 
belong 
belong 
part of 
object 
soshiki (organization) agent 
hito (human) possessive 
hito (human) agent 
hito (human) recipient 
nouryoku (ability) object 
\[ex. suugaku (mathematics)\] 
hito (human), soshiki (organization) 
gakumon bun'ya (field of study) 
hito (human), kikai (machine) 
de-ta (data) 
agent 
object 
agent 
object 
Table 2: Case frame of verb "kaiseki-suru (analyze)" 
Surface case 
ga-case (subject) 
wo-case (object) 
Semantic constraint 
human 
abstract, product 
Examples 
seito (student), kate (he) 
atai (value), de-ta (data) 
We resolve the indirect anaphora" using the follow- 
ing steps: 
1. We detect some elements which could be ana- 
lyzed by indirect anaphora resolution using "X 
no Y" and a verb case frame dictionary. When 
a noun was a verbal noun, we use a verb case 
frame dictionary. Otherwise, we use examples 
of "X no V." 
For example, in the following example sentences 
kaiseki (analysis) is a verbal noun, and we use 
a case frame of a verb kaiseki-surn (analyze) 
for the indirect anaphora resolution of kaiseki 
(analysis). The case frame is shown in Table 
2. In this table there are two case components, 
the ga-case (subject) and the wo-case (object). 
These two case components are elements which 
will be analyzed in indirect anaphora resolution. 
denkishingou-no riyouni-ni yotte 
(electronic detectors) (use) (by) 
(By using electronic detectors. ) 
Butsurigakusha-wa tairyou-no deeta-wo 
(physicist) (a large amount) (data) 
shuushuudekiru-youni-natta. 
(collect) 
(physicists had been able to collect large 
amounts of data. ) 
(2) 
sokode subayai kaiseki-no houhou-ga hitsuyouni-natta. 
(then) (quick) (analysis) (method) (require) 
(Then, they required a method of quick analysis.) 
2. We take possible antecedents from topics or loci 
in the previous sentences. We assign them a 
certain weight based on the plausibility that 
they are antecedents. The topics/foci and their 
weights are defined in Table 3 and Table 4. 
For example, in the case of "I went into an old 
house last night. The roof was leaking badly and 
..., .... an old house" becomes a candidate of the 
desired antecedent. In the case of "analysis" 
in example sentence 2, "electronic detectors," 
"physicists," and "large amounts of data" be- 
come candidates of the two desired antecedents 
of "analysis." In Table 3 and Table 4 such can- 
didates are given certain weights which indicate 
preference. 
3. We determine the antecedent by combining the 
weight of topics and foci mentioned in step 2, 
the weight of semantic similarity in "X no Y" or 
in a verb case frame dictionary, and the weight 
of the distance between an anaphor and its pos- 
sible antecedent. 
For example, when we want to clarify the an- 
tecedent of vane (roof) in example sentence 
1, we gather examples of "Noun X no vane 
(roof)" (roof of Noun X), and select a possi- 
32 
Table 3: The weight as topic 
Surface expression 
Pronoun/zero-pronoun ga/wa 
Noun wa/niwa 
Example I Weight 
EDIml , 
Table 4: The weight as focus 
Surface expression (Not including "wa") 
Pronoun/zero-pronoun wo (object)/ni (to)/kara (from) 
Noun ga (subject)/mo/ da/nara/koso 
Noun wo (object)/ni/, /. 
Noun he (to)/de (in)/kara (from)/yor/ 
Example 
\[John ni (to)\] shita (done) 
John ga (subject) shita (done) 
John ni (object) shita (done) 
gakkou (school) he (to) iku (go) 
I Weight 
16 
15 
14 
13 
ble noun which is semantically similar to Noun 
X as its antecedent. In example sentence 2, 
when we want to have an antecedent of kaiseki 
(analysis) we select as its antecedent a possi- 
ble noun which satisfies the semantic constraint 
in the case frame of kuichigau (differ) in Ta- 
ble 2 or is semantically similar to examples of 
components in the case frame. In the ga-case 
(subject), of three candidates, "electronic de- 
tectors," "physicists," and "large amounts of 
data," only "physicists" satisfies the semantic 
constraint, human, in the case frame of the verb 
kaiseki-suru in Table 2. So "physicists" is se- 
lected as the desired antecedent of the ga-case. 
In the wo-case (object), two phrases, "electronic 
detectors" and "large amounts of data" satisfy 
the semantic constraints, abstract and product. 
By using the examples "value" and "data" in 
the case frame, the phrase "large amounts of 
data," which is semantically similar to "data" 
in the examples of the case frame, is selected as 
the desired antecedent of the wo-case. 
We think that errors made by the substitution of 
a verb case frame for a noun case frame are rare, but 
many errors occur when we substitute "X no Y" for 
a noun case frame. This is because "X no Y" (Y of 
X) has many semantic relationships, in particular a 
feature relationship (ex. "a man of ability"), which 
cannot be an indirect anaphoric relationship. To 
reduce the errors, we use the following procedures. 
1. We do not use an example of the form "Noun X 
no Noun ¥" (Y of X), when noun X is an adjec- 
tive noun \[ex. HONTOU (reality)\], a numeral, 
or a temporal noun. For example, we do not 
use honton (reality) no (of) hannin (criminal) 
(a real criminal). 
2. We do not use an example of the form "Noun 
X no Noun Y" (Y of X), when noun Y is a 
noun that cannot be an anaphor of an indirect 
anaphora. For example, we do not use "Noun X 
no tsurn (crane)," or "Noun X no ningen (hu- 
man being)." 
We cannot completely avoid errors by introducing 
the above procedure, but we can reduce them to a 
certain extent. 
Nouns such as ichibu (part), tonari (neighbor) and 
betsu (other) need further consideration. When such 
a noun is a case component of a verb, we use infor- 
mation on the semantic constraints of the verb. We 
use a verb case frame dictionary as shown in Table 
5. 
takusan-no kuruma-ga kouen-ni tomatte,ita. 
(many) (car) (in the park) (there were) 
(There were many cars in the park.) 
ichibu-wa kith-hi mukatta 
\[A part (of them)\] (to the north) (went) 
(A part of them went to the north.) (3) 
In this example, since ichibu (part) is a ga-case (sub- 
ject) of a verb mukau(go), we consult the ga-case 
(subject) of the case frame of mukau (go). Some 
noun phrases which can also be used in the case 
component are written in the ga-case (subject) of 
the case frame. In this case, kate (he) and hune 
(ship) are written as examples of things which can 
be used in the case component. This indicates that 
the antecedent is semantically similar to kare (he) 
and hune (ship). Since takusan no kuruma (many 
cars) is semantically similar to hune (ship) in the 
meaning of vehicles, it is judged to be the proper 
antecedent. 
When such a noun as tonari (neighbor or next) 
modifies a noun X as tonari no X, we consider the 
antecedent to be a noun which is similar to noun X 
33 
Table 5: Case frame of verb "mukad' (go to) 
Surface case Semantic constraint Examples 
ga-case (subject) concrete kare (he), hune (ship) 
n/-case (object) place kouen (park), minato (port) 
in meaning. 
ojiisan-wa ooyorokobi-wo-shite ie-ni kaerimashita. 
(the old man) (in great joy) (house) (returned) 
\[The old man returned home (house) in great joy,\] 
okotta koto-wo hitobito-ni hanashimashita 
(happened to him) (all things) (everybody) (told) (4) 
(and told everybody all that had happened to him.) 
tonari-no ie-ni ojiisan-ga mouhitori sunde-orimashita. 
(next) (house) (old man) (another) (llve) 
(There lived in the next house another old man.) 
For example, when tonari (neighbor or next) modi- 
fies ie (house), we judge that the antecedent of tonari 
(neighbor or next) is ie (house) in the first sentence. 
3 Anaphora Resolution System 
3.1 Procedure 
Before starting the anaphora resolution process, the 
syntactic structure analyzer transforms sentences 
into dependency structures (Kurohashi and Nagao, 
1994). Antecedents are determined by heuristic rules 
for each noun from left to right in the sentences. 
Using these rules, our system gives possible an- 
tecedents points, and it determines that the possible 
antecedent having the maximum total score is the 
desired antecedent. This is because a several types 
of information are combined in anaphora resolution. 
An increase in the points of a possible antecedent 
corresponds to an increase of the plausibility of the 
possible antecedent. 
The heuristic rules are given in the following form: 
Condition ~ { Proposal, Proposal, ... } 
Proposal := ( Possible-Antecedent, Point ) 
Surface expressions, semantic constraints, referential 
properties, for example, are written as conditions in 
the Condition part. A possible antecedent is written 
in the Possible-Antecedent part. Point refers to the 
plausibility of the possible antecedent. 
To implement the method mentioned in Section 2, 
we use the weights W of topics and foci, the distance 
D, the definiteness P, and the semantic similarity 
S (in R4 of Section 3.2) to determine points. The 
weights W oftopics and foci are given in Table 3 and 
Table 4 respectively in Section 2, and represent the 
preferability of the desired antecedent. In this work, 
a topic is defined as a theme which is described, and 
a focus is defined as a word which is stressed by the 
speaker (or the writer). But we cannot detect topics 
and foci correctly. Therefore we approximated them 
as shown in Table 3 and Table 4. The distance D is 
the number of the topics (foci) between the anaphor 
and a possible antecedent which is a topic (focus). 
The value P is given by the score of the definiteness 
in referential property analysis (Murata and Nagao, 
1993). This is because it is easier for a definite noun 
phrase to have an antecedent than for an indefinite 
noun phrase to have one. The value S is the semantic 
similarity between a possible antecedent and Noun 
X of "Noun X no Noun Y." Semantic similarity is 
shown by level in Bunrui Goi Hyou (NLRI, 1964). 
3.2 Heuristics for determining antecedents 
We wrote 15 heuristic rules for noun phrase 
anaphora resolution. Some of the rules are given 
below: 
R1 When the referential property of a noun phrase 
(an anaphor) is definite, and the same noun 
phrase A has already appeared, =¢, 
{ (the noun phrase A, 30)} 
A referential property is estimated by this 
method (Murata and Naga~, 1993). This is a 
rule for direct anaphora. 
R2 When the referential property of a noun phrase 
is generic, =¢, 
{ (generic, 10)} 
R3 When the referential property of a noun phrase 
is indefinite, 
{ (indefinite, 10)} 
R4 When a noun phrase Y is not a verbal noun, ==~ 
{ (A topic which has the weight W and the dis- 
tanceD, W-D+P+S), 
(A focus which has the weight W and the dis- 
tanceD, W-D+P+S), 
(A subject in a subordinate clause or a main 
clause of the clause, 23 -t- P + S) 
where the values W, D, P, and S are as they 
were defined in Section 3.1. 
R5 When a noun phrase is a verbal noun, =~ 
{ (A topic which satisfies the semantic con- 
straint in a verb case frame and has the weight 
W and the distance D, W-D+P+S), 
(A focus which satisfies the semantic constraint 
and has the weight W and the distance D, 
W-D+P+S), 
34 
kono dorudaka-wa kyoulyou-wo gikushaku saseteiru. 
(The dollar's surge) (cooperation) (is straining) 
(The dollar's surge is straining the cooperation. ) 
jikokutuuka-wo mamorouto nisidoku-ga kouleibuai-wo 
(own currency) (to protect) (West Germany) (official rate) 
(West Germany raised (its) official rate to protect the mark. ) 
hikiagela. 
(raised) 
Indefinite nisidoku jikokutuuka kyoutyou dorudaka 
West Germany own currency cooperation dollar's surge 
R3 10 
R4 25 -23 -24 -17 
23 Subject 
Topic Focus (W) 
Distance (D) 
Definiteness (P) 
Similarity ( S) 
Total Score 10 
-5 
7 
25 
14 
-2 
-5 
-30 
-23 
14 
-3 
-5 
-30 
-24 
20 
-2 
-5 
-30 
-17 
Examples of "noun X no kouteibuai (official rate)" 
"nihon (Japan) no kouteibuai (official rate)", 
"beikoku (USA) no kouteibuai (official rate)" 
Figure 1: Example of indirect anaphora resolution 
(A subject in a subordinate clause or a main 
clause of the clause, 23 + P + S) 
R6 When a noun phrase is a noun such as ichibu, 
tonari, and it modifies a noun X, =~ 
{ (the same noun as the noun X, 30)} 
3.3 Example of analysis 
An example of the resolution of an indirect anaphora 
is shown in Figure 1. Figure 1 shows that the noun 
koutei buai (official rate) is analyzed well. This is 
explained as follows: 
The system estimated the referential property 
of koutei buai (official rate) to be indefinite in 
the method (Murata and Nagao, 1993). Follow- 
ing rule R3 (ection 3.2) the system took a candi- 
date "Indefinite," which means that the candidate 
is an indefinite noun phrase that does not have 
an indirect anaphoric referent. Following R4 (Sec- 
tion 3.2) the system took four possible antecedents, 
nisidoku (West Germany), jikokutuuka (own cur- 
rency), kyoutyou (cooperation), dorudaka (dollar's 
surge). The possible antecedents were given points 
based on the weight of topics and foci, the distance 
from the anaphor, and so on. The system properly 
judged that nisidoku (West Germany), which had 
the best score, was the desired antecedent. 
4 Experiment and Discussion 
Before the antecedents in indirect anaphora were 
determined, sentences were transformed into a case 
structure by the case analyzer (Kurohashi and Na- 
gao, 1994). The errors made by the analyzer were 
corrected by hand. We used the IPAL dictionary 
(IPAL, 1987) as a verb case frame dictionary. We 
used the Japanese Co-occurrence Dictionary (EDR, 
1995) as a source of examples for "X no Y." 
We show the result of anaphora resolution using 
both "X no Y" and a verb case frame dictionary 
in Table 6. We obtained a recall rate of 63% and 
a precision rate of 68% when we estimated indirect 
anaphora in test sentences. This indicates that the 
information of "X no Y" is useful to a certain extent 
even though we cannot make use of a noun frame dic- 
tionary. We also tested the system when it did not 
have any semantic information. The precision and 
the recall were lower. This indicates that semantic 
information is necessary. The experiment was per- 
formed by fixing all the semantic similarity values S 
to 0. 
We also estimated the results for the hypothetical 
use of a noun case frame dictionary. We estimated 
these results in the following manner: We looked 
over the errors that had occured when we used "X 
no Y" and a verb case frame dictionary. We regarded 
errors made for one of the following three reasons as 
right answers: 
35 
Table 6: Results 
Non-verbal noun I Verbal noun Total 
Recall I Precisi°n I Recall I Precision Recall \] Precision 
Experiment made when the system does not use any semantic information 
85%(56/66) 67%(56/83) 40%(14/35) 44%(14/32) 69%(70/101) 61%(70/115) 
42%(15/36) (35/70) 1 46% (35/76) 50%(20/40) 47%(15/32) 
Experiment using "X no Y" and verb case frame 
91%(60/66) 86%(60/70) 66%(23/35) 79%(23/29) 82%(83/101) 84%(83/99) 83%(24/29) (44/70) 168% 56%(20/36) (44/65) 
Estimation for the hypothetical use of a noun case frame dictionary 91%(60/66) 88%(60/68) 69%(24/35) 89%(24/27) 83%(84/101) 88%(84/95) 
79%(30/38) 186%(30/35)63%(20/32)177%(20/26)I (50/70) I 82% 71% (50/61) 
The upper row and the lower row of this table show rates on training sentences and test sentences 
respectively. 
The training sentences are used to set the values given in the rules (Section 3.2) by hand. 
Training sentences {example sentences (Walker et al., 1994) (43 sentences), a folk tale Kobutori jiisan 
(Nakao, 1985) (93 sentences), an essay in Tenseijingo (26 sentences), an editorial (26 sentences)} 
Test sentences {a folk tale Tsuru no ongaeshi (Nakao, 1985) (91 sentences), two essays in Tenseijingo (50 
sentences), an editorial (30 sentences)} 
Precision is the fraction of the noun phrases which were judged to have the indirect anaphora as an- 
tecedents. Recall is the fraction of the noun phrases which have the antecedents of indirect anaphora. 
We use precision and recall to evaluate because the system judges that a noun which is not an antecedent 
of indirect anaphora is an antecedent of indirect anaphora, and we check these errors thoroughly. 
1. Proper examples do not exist in examples of "X 
no Y" or in the verb case frame dictionary. 
2. Wrong examples exist in examples of "X no Y" 
or in the verb case frame dictionary. 
3. A noun case frame is different from a verb case 
frame. 
If we were to make a noun case frame dictionary, it 
would have some errors, and the success ratio would 
be lower than the ratio shown in Table 6. 
Discussion of Errors 
Even if we had a noun case frame dictionary, there 
are certain pairs of nouns in indirect anaphoric rela- 
tionship that could not be resolved using our frame- 
work. 
kon'na hidoi hubuki-no naka-wo ittai dare-ga kita-no- 
ka-to ibukarinagara, obaasan-wa iimashita. 
(Wondering who could have come in such a heavy 
snowstorm, the old woman said:) 
"donata-jana" 
("Who is it?") 
to-wo aketemiruto, soko-niwa zenshin yuki-de masshi- 
roni natta musume-ga tatte orimashita. 
(She opened the door, and there stood before her 
a girl all covered with snow. ) (5) 
The underlined mnsnme has two main meanings: a 
daughter or a girl. In the above example, mnsnme 
means "girl" and has no indirect anaphora rela- 
tion but the system incorrectly judged that it is 
the daughter of obaasan (the old woman). This is 
a problem of noun role ambiguity and is very diffi- 
cult to solve. 
The following example also presents a difficult 
problem: 
shushou-wa teikou-no tsuyoi 
(prime minister) (resistance) (very hard) 
senkyoku-no kaishou-wo miokutta. 
(electoral district) (modification) (give up) 
(The prime minister gave up the modification of 
some electoral districts where the resistance was very 
hard.) (6) 
On the surface, the underlined leikou (resistance) 
appears to refer indirectly to senkyoku (electoral 
district). But actually teikou (resistance) refers to 
the candidates of senkyokn (electoral district) not 
to senkyoku (electoral district) itself. To arrive at 
this conclusion, in other words, to connect senkyoku 
(electoral district) and ~eikou (resistance), it is nec- 
essary to use a two-step relation, "an electoral dis- 
trict =¢, candidates," "candidates :=¢, resist" in se- 
quence. It is not easy, however, to change our system 
so it can deal with two-step relationships. If we ap- 
ply the use of two-step relationships to nouns, many 
nouns which are not in an indirect anaphoric rela- 
36 
Table 7: Examples of arranged "X no Y" 
Noun Y 
kokumin (nation) 
genshu (the head of 
state) 
yane (roof) 
mokei (model) 
gyouji (event) 
jinkaku 
(personality) 
Arranged noun X 
<Human> aite (partner) <Organization> kuni (country), senshinkoku (an ad- 
vanced country), vyoukoku (the two countries), naichi (inland), zenkoku (the whole 
country), nihon (Japan), soren (the Soviet Union), eikoku (England), amerika 
(America), suisu (Switzerland), denmaaku (Denmark), sekai (the world) 
<Human> raihin (visitor) <Organization> gaikoku (a foreign country), kakkoku 
(each country), poorando (Poland) 
<Organization> hokkaido (Hokkaido), sekai (the world), gakkou (school), kou- 
jou (factory), gasorinsutando (gas station), suupaa (supermarket), jilaku (one's 
home), honbu (the head office) <Product> kuruma (car), juutaku (housing), ie 
(house), shinden (temple), genkan (entrance), shinsha (new car) <Phenomenon> 
midori (green) <Action> kawarabuki (tile-roofed) <Mental> houshiki (method) 
<Character> keishiki (form) 
<Animal> zou (elephant) <Nature> fujisan (Mt. Fuji) <Product> imono (an 
article of cast metal), manshon (an apartment house), kapuseru (capsule), den- 
sha (train), hunt (ship), gunkan (warship), hikouki (airplane), jettoki (jet plane) 
<Action> zousen (shipbuilding) <Mental> puran (plan) <Character> unkou 
(movement) 
<Human> koushitsu (the Imperial Household), oushilsu (a Royal family), iemoto 
(the head of a school) <Organization> nouson (an agricultural village), ken (pre- 
fecture), nihon (Japan), soren (the Soviet Union), tera (temple), gakkou (school) 
<Action> shuunin (take up one's post), matsuri (festival), iwai (celebration), jun- 
rei (pilgrimage) <Mental> kourei (an established custom), koushiki (formal) 
<Human> watashi (myself), ningen (human), seishounen (young people), seijika 
(statesman) 
tion will be incorrectly judged as indirect anaphora. 
A new method is required in order to infer two rela- 
tionships in sequence. 
5 Consideration of Construction of 
Noun Case Frame Dictionary 
We used "X no Y" (Y of X) to resolve indirect 
anaphora. But we would achieve get a higher accu- 
racy rate if we could utilize a good noun case frame 
dictionary. Therefore we have to consider how to 
construct a noun case frame dictionary. A key is to 
get the detailed meaning of "no (of)" in "X no Y." 
If it is automatically obtainable, a noun case frame 
dictionary could be constructed automatically. Even 
if the semantic analysis of "X no Y" is not done well, 
we think that it is still possible to construct the dic- 
tionary using "X no Y." For example, we arrange 
"noun X no noun Y" by the meaning of "noun Y," 
arrange them by the meaning of "noun X", delete 
those where "noun X" is an adjective noun, and ob- 
tain the results shown in Table 7. In this case, we use 
the thesaurus dictionary "Bunrui Goi Hyou" (NLRI, 
1964) to learn the meanings of nouns. It should not 
be difficult to construct a noun case frame dictio- 
nary by hand using Table 7. We will make a noun 
case frame dictionary by removing aite (partner) in 
the line of kokumin (nation), raihin (visitor) in the 
line of genshu (the head of state), and noun phrases 
which mean characters and features. When we look 
over the noun phrases for kokumin (nation), we no- 
tice that almost all of them refer to countries. So 
we will also make the semantic constraint (or the se- 
mantic preference) that countries can be connected 
to kokumin (nation). When we make a noun case 
frame dictionary, we must remember that examples 
of "X no Y" are insufficient and we must add exam- 
ples. For example, in the line of genshu (the head of 
state) there are few nouns that mean countries. In 
this case, it is good to add examples by from the ar- 
ranged nouns for kokumin (nation), which is similar 
to genshu (the head of state). Since in this method 
examples are arranged by meaning in this method, 
it will not be very difficult to add examples. 
6 Conclusion 
We presented how to resolve indirect anaphora in 
Japanese nouns. We need a noun case frame dic- 
tionary containing information about noun relations 
to analyze indirect anaphora, but no such dictionary 
exists at present. Therefore, we used examples of "X 
no Y" (Y of X) and a verb case frame dictionary. We 
estimated indirect anaphora by using this informa- 
tion, and obtained a recall rate of 63% and a pre- 
cision rate of 68% on test sentences. This indicates 
37 
that information about "X no Y" is useful when we 
cannot make use of a noun case frame dictionary. 
We estimated the results that would be given by a 
noun case frame dictionary, and obtained recall and 
precision rates of 71% and 82% respectively. Finally, 
we proposed a way to construct a noun case frame 
dictionary by using examples of "X no Y." 

References 
Carterl D. 1987. Interpreting Anaphors in Natural 
Language Texts. Halsted Press. 
Electronic Dictionary Research Institute, LTD. 
1995. EDR Electronic Dictionary, Japanese Co- 
occurrence Dictionary, Version 1.5, (in Japanese). 
Information-technology Promotion Agency, Japan. 
1987. IPA Lexicon of the Japanese Language for 
Computers IPAL (Basic Verbs), (in Japanese). 
Kurohashi, S., Nagao, M. 1994. A Method of Case 
Structure Analysis for Japanese Sentences based 
on Examples in Case Frame Dictionary. The In- 
stitute of Electronics, Information and Commu- 
nication Engineers Transactions on Information 
and Systems E77-D (2), pages 227-239. 
Murata, M., Nagao, M. 1993. Determination of 
referential property and number of nouns in 
Japanese sentences for machine translation into 
English. In Proceedings of the 5th TMI, pages 
218-225, Kyoto, Japan, July. 
Murata, M., Nagao, M. 1995. An Estimate of Ref- 
erents of Pronouns in Japanese Sentences using 
Examples and Surface Expressions, (in Japanese) 
IPSJ 95-NL-108, I-Iukuoka, Japan, July. 
Nakao, K. 1985. The Old Man with a Wen. Eiyaku 
Nihon Mukashibanashi Series, Vol. 7, Nihon Eigo 
Kyouiku Kyoukai (in Japanese). 
National Language Research Institute (1964) Word 
List by Semantic Principles. Syuei Syuppan (in 
Japanese). 
Walker, M., Iida, M., Cote, S. 1994. Japanese Dis- 
course and the Process of Centering Computa- 
tional Linguistics, Vol. 20, No. 2, pages 193-232. 
Watanabe, Y., Kurohashi, S., Nagao, M. 1992. Con- 
struction of semantic dictionary by IPAL dictio- 
nary and a thesaurus (in Japanese). In Proceed- 
ings of the 45th Convention of 1PSJ, pages 213- 
214, Tokushima, Japan, July. 
