Automatic Construction of a Japanese Onomatopoeic Dictionary Using Text Data on the WWW

Abstract

As new onomatopoeic words are often created at short notice, existing dictionaries tend to have an insufficient number of their entries. Furthermore, onomatopoeic words seldom appear in collections of newspaper articles, that have been used as corpora in natural language processing. In this work, we present a method of automatically acquiring lexical knowledge for Japanese onomatopoeic words from the WWW. As a result, we could automatically construct a onomatopoeic dictionary that contained 5,130 entries. By manually evaluating 487 newly acquired words that were not in the existing dictionary, we found that we could acquire 266 new onomatopoeic words, and if words in the existing dictionary were regarded as being correct, precision of our automatic acquisition was 83.6%.