Word sense disambiguation of adjectives using dependency structure and degree of association between sentences
Kenichi Mishina, Seiji Tsuchiya, Hirokazu Watabe · 2017
Good WSD results has been shown by the supervised methods. But it is extremely costly to manually construct a corpus with semantic label as learning data. It is not for practical use. Therefore, the unsupervised methods using dictionary knowledge have been actively researched. Many conventional unsupervised WSD is using English dictionary. There are few cases of unsupervised WSD in Japanese. Therefore it is necessary to improve the accuracy of unsupervised WSD. In this paper, we propose a new unsupervised WSD method of adjectives using sentence structure information and dictionary knowledge. As the evaluation result, the accuracy was 29.1% in the conventional method, and it was 40.2% in the proposed method. Calculation of the z-test confirmed that the evaluation result was significant.