Feature expansion for word sense disambiguation

One of the most serious obstacles in research on word sense disambiguation (WSD) is sparseness of training data. This paper describes and motivates a method of feature expansion as a means of resolving the data sparseness problem in supervised corpus-based WSD. The expanded features are extracted from an existing corpus and WordNet automatically. We use our method to supplement the feature expansion approach of [Leacock and Chodorow 1998]. In the experiments, the addition of our method more than doubled the precision improvement over baseline that was obtained by using Leacock and Chodorow's approach alone.