Finding a Deficiency of a Meaning in a Bunrui-goi-hyou Entry by Using Corpora

Language, Information and Computation(PACLIC 11), 1996, 177-186 Finding a deficiency of a meaning in a Bunrui-goi-hyou entry by using corpora Hiroyuki Shinnou Ibaraki University Dept. of Systems Engineering shinnou@lily.dse.ibaraki.ac.jp Abstract This paper presents a method to automatically find meanings which should be but are not entered in thesaurus. In this paper, we use Bunrui-goi-hyou as the thesaurus. To find the noun n with the meaning lacking in Bunrui-goi-hyou, we applies our presented method which extracts idioms from a corpus. We use the clue that many idioms with the noun n are extracted through that method. We have experimented with a corpus which consists of 5 years' worth of articles from a Japanese economic newspaper. As a result, we found 177 types of lacking meanings. Furthermore, our method can find not only a deficiency of a meaning but also meanings which are not used or are specifically used in the corpus domain. 1 Introduction Bunrui-goi-hyou (The National Language Research Institute 1994) repre- sents the Japanese thesaurus, and is used in many researches. Needless to say, it is important to expand and refine Bunrui-goi-hyou for the sake of Japanese natural language processing. This paper presents the method to automatically find meanings which must be entered in Bunrui-goi-hyou but not be entered. By this method, we can efficiently fill the deficiency of meanings in Bunrui-goi-hyou. We should notice that our method automatically executes the following: (1) to estimate that a certain noun n has the meaning g which is not entered as a meaning of this noun n in Bunrui-goi-hyou, (2) to show some nouns which have the similar meaning to the meaning however, our method does not trace what the meaning g is. This meaning is manually decided by observing nouns shown in (2). For example, the Japanese noun "J*" has meanings "voice" and "opinion ", but "r. " in Bunrui-goi-hyou has only "voice" and does not have "opinion 177 //. Our method points out that "r4 " in Bunrui-goi-hyou has a deficiency of a meaning, and shows some nouns such as "SR (opinion)" ,"NS (view)" ,"1 (insistence)" etc. which have the similar meaning to "opinion". By observing these nouns, we can manually decide that the lacking meaning of "P" is "opinion". The traditional research to acquire the unknown meaning was done by (Wilensky 1990). In his research, if a sense of a word is unknown, the sense is estimated from similar uses of other words to the use of the word. This approach, which estimates a feature of an unknown word from features of similar words, is taken to cope with the data sparseness problem of a corpus (Dagan 1993). In short, this approach is base on the idea that the unseen part can be estimated from seen similar parts. However, the simple use of this idea alone cannot find an unknown meaning of a word, because even similar words hardly have same polysemy to the word. For example, nouns such as "%� (laughter)" ,"'� (cheer)", "E PA (scream)", " 7..12 …

Content truncated to 500 words