Discovering Word Senses for Polysemous Words Using Feature Domain Similarity

This paper presents a new clustering algorithm called DSCBC which is designed to automatically discover word senses for polysemous words. DSCBC is an extension of CBC (Pantel and Lin, 2002), and incorporates feature domain similarity: the similarity between the features themselves, obtained a priori from sources external to the dataset. By incorporating the feature domain similarity in clustering, DSCBC produces monosemous clusters (a cluster in one domain), thereby discovering individual senses of polysemous words. For evaluation, we apply the algorithm to Japanese and English adjectives, and compare the derived senses against manually created lexicons. The results show significant improvements over other clustering algorithms including CBC.