Word Sense Disambiguation Based on Weight Distribution Model with Multiword Expression

This paper proposes a two-phase word sense disambiguation method, which filters only the relevant senses by utilizing the multiword expression and then disambiguates the senses based on Weight Distribution Model. Multiword expression usually constrains the possible senses of a polysemous word in a context. Weight Distribution Model is based on the hypotheses that every word surrounding a polysemous word in a context contributes to disambiguating the senses according to its discrimination power. The experiments on English data in SENSEVAL-1 and SENSEVAL-2 show that multiword expression is useful to filter out irrelevant senses of a polysemous word in a given context, and Weight Distribution Model is more effective than Decision Lists.