Construction of Japanese semantically compatible words resource

Abstract

We have constructed a Japanese semantically compatible resource and attached it to a dictionary used in our language analyzer, which segments text into words. We expect that semantically compatible words solve the data sparseness problem of corpus-based Natural Language Processing. By grouping compatible words together, the amount of words to process can be much reduced. In this study, we define hyponymy-and-hypernymy relation groups and synonym groups as semantically compatible words. The semantically compatible resource contains 343 concepts as hyponymy-and-hypernymy relation groups and 21,784 concepts as synonymy groups. We can obtain semantically compatible words from a Japanese word analyzer, SNOWMAN. The constructed resource will be available to the public.