When vocabulary cannot be found in bilingual dictionaries, it is frequently obtained by using a third language as an intermediary. 'This indicates that, supplemental information may Iii; in other forms in other dictionaries. Here we try using; electronic dictionaries which can he reformed on a large scale, to extract, this informations so that we can obtain subsidiary data and refine a direct bilingual dictionary.

Looking up words in bilingual dictionaries in­termediating the third language is a method often used by people who handle uncommon languages in a specific domain. If this process can be auto­mated, bilingual dictionaries of any kind between any languages may he obtained as long as these concerned languages have dictionaries to a com­mon language. One objective of the research re­ported here is to establish a first; step in automat ing this process.

To construct a Japanese«-»French dictionary, we chose English as the intermediary language because Japanesc<->English and English*->French dictionaries exist in electronic forms and because published Japanese<->French dictionaries provide enough vocabulary in comparison with the result­ing dictionary.

In Section 2 we describe a method for extract­ing equivalencies for a given word. Its fundamen­tal concepts are stated in Section 3. The whole procedure used to construct the new dictionary is shown in Section 4 and in Section 5 the resulting dictionary is evaluated.

Japanese-English, English-Japanese, English-French, French-English, Japanese-French and French-Japanese dictionaries are respectively de­noted l)icj_,a, Dice.^j, Dice. ,t, Dicf _ ,0, Dicj. + f,

Kyo.ji UMEMURA

NTT basic Research Laboratories. 3-9-11 Midori, Musashino, Tokyo, 180, Japan, umemuraQnuesun.NTT.JP and Dicf. ,j. Die*..,,, is called an inverse, dic­tionary of Dicy„f!e. Japanese words have infor­mation in the following' format: pronunciation in romaji, and its equivalence in English. English words are written in this font and French words in this font.

The most naive way to use English to obtain French words corresponding to a Japanese word is to look up the Japanese word in a Dicj„,aand then look up the resultant English words in a Dicp^.j. 'The resulting French words can be regarded as equivalence candidates (ECs) of the original Japanese word. For example, in Fig. I, ECs for a Japanese word "ffil"""(kyousou : com­petition) are compétition, concours, race etc. Among' these, race and hâte are inadequate as equivalents of "{■it"''/".

As for race, the English word race has several meanings with the same spelling: one is to com­pete and another is human race. It, is human race which induces the inadequate EC race. As for hâte, the English word race has the wider mean­ing to hairy which the original Japanese word "Si If'-" does not. Since hâte is a direct translation of to hurry, it is inappropriate as an equivalence.

'The following three cases generate irrelevant ECs.

1. An English word with the same spelling; but with dilferenl meanings is intermediated, (race in the above example)

2. An English word with a wider meaning than that of original Japanoso word is intermedi­ated, (hâte in the above example) 3. 'There are mistakes in dictionaries.

The first two cases are due to the ambiguity in En­glish. An English word with a narrower meaning; than the Japanese source; may miss some French equivalents. We think thai; if the original word has ambiguity and several meanings, the dictio­nary gives the corresponding; English words.

We could handle the; problem of choosing equiv­alencies from among ECs by semantic processing;, but, expressing meaning of vocabulary in dictio­naries is a great; problem. A simpler way is to look the ECs up in the inverse dictionary. For ex­ample, one can consult Dicj_ ,j for competition, concours, and race and thereby get "vM", "fx

Japanese English French

Fig. 1 Equivalence candidates (ECs) for Fig. 2 One time inverse consultation (/Cj).

and "Affi" (jinshu : human race) as their respective equivalencies. Since "Ai !'" has noth­ing to do with "jiyK/i-", race is excluded. We call this method of looking up ECs in the inverse dictionaries when choosing relevant equivalencies inverse consultation, and we call the words ob­tained by looking up inverse dictionaries the se­lection area (SA). Inverse consultation utilizes the structure of dictionaries to measure the nearness of the meanings of words in different languages.

The simplest application of inverse consultation is to use Dicf_,e. In the above example, each EC is looked up in Dicj_>0 and the results are com­pared with the English equivalencies of "SSK/'-" (E); namely, competition, contest, and race. The SA of competition is competition, contest and match, which have the elements contest and competition in common with E (Fig. 2). As compétition derived from competition, competition should be put aside, but contest is still left as a common element and thus compétition is selected as an equivalence of "'»M". As for race, the SA of race consists of race and ancestry, whose inter­section with E only gives race; so race is judged as an inadequate EC. In short, the number of ele­ments in common between the selection area and E indicates the nearness of the meaning between the EC and the original word.

For the inverse consultation described above, the SA was in English. If we use Dice^j as an inverse dictionary successively after consult­ing Dicf_a, then the SA is in Japanese and we compare "Wfr" with the SA (Fig. 3). The SA for

Fig. 3 Two times inverse consultation (/Cj) .

compétition consists of two "»"/'"s and three "■$£$"s (kyougi : game). For race, the SA has W, "AW, «ftfll"(some : ancestry). Since "«8K/'" appears only once for race, we discard the EC race.

There can be a infinite number of inverse con­sultations according to the number of consulted inverse dictionaries. If the inverse dictionaries are consulted n times, we call the method n times inverse consultation (ICn). Which inverse dictio­nary to use does not always have a unique answer. For IC'2 for example, we may consult Dice^f af­ter consulting Dicf_0 with the SA in French.

Equivalencies obtained for the Japanese word "»i are concours, rivalité, competition, course, concurrence, emulation and the intermediated English words are

competition, contest, rival, rivalry, race. The entry "Wf>" is of Type B. The number of ECs is 41 (including overlaps), and 13 of them are selected by IC\. The number of ECs in each category of irrelevant words described in Section 2.1 is listed in Table 1, which indicates that in­verse consultation can detect; the relevant words even when there are mistakes in the dictionar­ies. The word which should not be dropped was joute.

Equivalencies for "SK/" in [Tam85] are

Table 2 Classification of entries.

Table 3 Entries with no EC.

concours, rivalité, competition, course, concurrence.

Our result contains émulation (which means ri­valry) in addition to the entries in the published dictionary.

Table 6 Japanese entries whose III are 0 %.

The entries tend to have either Rl=80%~100% or Rl=0%~20%. Of the latter, some examples are listed in Table 5 and 6. For distique in Ta­ble 5, which is a term in French literature, Df_j translates it into the term for the corresponding kind of Japanese literature. Although it resem­bles to the direct translation of distique, it is only an analogy. On the other hand, since the di­rect translation of distique in the published dic­tionary adopts the concept of French poems, it is uncommon and cannot be understood by most Japanese readers. The same is true for boulier except that the common Japanese is indicated in a published dictionary. Pull is borrowed from the English word pull-over whose direct translation is contained in a published dictionary, and it is not a common word in Japanese.

In the first example in Table G, grive is the generic name equivalent to thrush, whereas merle is a kind of thrush. The second example shows that more equivalencies are found in Dj_,f.

To sum up, the resulting dictionary can be uti­lized in conjunction with the published dictionar­ies as follows:

• To revise the equivalencies.

• To supplement the equivalencies. Sections of the resulting dictionaries is listed

in Table 7. List 1 is Dj_,f and List 2 is Df^j.

For each list, entries are in the first row and their equivalencies are in the third row. Symbols in the second row indicate how appropriate each equiv­alence is. (Refer to the notes beside.) G Related Work

The use of a third language English as an inter­mediary in the construction of a bilingual dictio­nary was tested manually on a large scale on edit­ing the Spanish-Japanese dictionary [KuwOO]. It is now a representative middle-sized dictionary hav­ing a large quantity of information.

Tokunaga and Tanalca [TokOO] tried to extract a conceptual dictionary from Japanese-English and English-Japanese dictionaries. Although they used a concept similar to ours that is the graph structure of a dictionary related to meaning of words, their frameworks and final product differ from ours.

The proposed method for using a intermediate language to construct a bilingual dictionary uti­lizes the structure of dictionaries and morphemes and can choose appropriate equivalencies for most entries. Comparing the resulting dictionary with published dictionaries showed that data obtained are useful for revising and supplementing the vo­cabulary of existing dictionaires.

To increase the accuracy with which equiva­lencies can be selected, mistakes in word-to-word dictionaries must be corrected even if our method may detect appropriate equivalencies. One way to do this would be to use thesaurus to check whether the extracted correspondences arc rele­vant.

Nouns were taken into consideration in this re­search, and the next step will be to apply the proposed method to other parts of speech. We also need to establish a way to handle compound words in European languages. Acknowledgements

We express our gratitude to Prof. Hideya Iwasaki of University of Tokyo for his guidance on theoret­ical aspects and essential remarks on experiments. We are also thankful to Prof. Nigel Ward of Uni­versity of Tokyo for his lexicographical advice and for his invaluable aid in shaping our idea.