Japanese Hyponymy Extraction based on a Term Similarity Graph
Takuya Akiba, Tetsuya Sakai · Medical Entomology and Zoology · 2011
Semantic relations between words, such as hyponymy, synonymy and meronymy, have various information access applications (e.g. Web search) and the automatic extraction of such relations from corpora is an important research problem in natural language processing. For the Japanese language, there exist several linguistic resources that contain these relations, such as the Japanese Wordnet, Nihongo Goitaikei and EDR electric dictionary. However, the cost of maintaining such knowledge and of adapting to linguistic phenomena that keep evolving is very high. Therefore, many studies have been conducted for automatic extraction of these semantic relationships. In this study, we focus on automatic extraction of hyponymy relations from large corpora. Here, a word A is a hypernym of a word B (or the word B is a hyponym of the word A) if B is a kind of A or B is an instance of A. The relation is also called as is-a relation. Although there are already many existing studies on automatic hyponymy extraction, there still is a lot of room for improvement in terms of precision, recall or the trade-off between the two. Hyponymy relations are useful, for example, for an information access system