Gloss-Based Semantic Similarity Metrics for Predominant Sense Acquisition

In recent years there have been various approaches aimed at automatic acquisition of predominant senses of words. This information can be exploited as a powerful backoff strategy for word sense disambiguation given the zipfian distribution of word senses. Approaches which do not require manually sense-tagged data have been proposed for English exploiting lexical resources available, notably WordNet. In these approaches distributional similarity is coupled with a semantic similarity measure which ties the distributionally related words to the sense inventory. The semantic similarity measures that have been used have all taken advantage of the hierarchical information in WordNet. We investigate the applicability to Japanese and demonstrate the feasibility of a measure which uses only information in the dictionary definitions, in contrast with previous work on English which uses hierarchical information in addition to dictionary definitions. We extend the definition based semantic similarity measure with distributional similarity applied to the words in different definitions. This increases the recall of our method and in some cases, precision as well.