The SketchNet project : knowledge-based word sense disambiguation

Abstract

Advances in the science of Natural Language Processing have, throughout the years, made researchers more aware of the problem of Word Sense Disambiguation. The need for automatic processing of the human language has emphasized the necessity of a good automatic disambiguation solution, which would be able to determine the correct sense for a word, given the context within which it occurs. The SketchNet project proposes, implements and evaluates a new knowledge-based Word Sense Disambiguation method. The proposed method makes use of two readily available knowledge sources: Word Net and Sketch Engine. Though WordNet has been used by many existing automatic disambiguation methods, this is the first time that services offered by the Sketch Engine are being applied to a Word Sense Disambiguation task. The proposed method attempts to perform disambiguation through the combined use of three main components: polysemy, synonymy and semantic richness. Disambiguation is performed with respect to the relational context of a word. Testing of the proposed method was performed through the use of the SketchNet application, a .Net Windows Application implementing it. When evaluated on a test set of nouns with very fine sense distinctions, the proposed method ranked the correct sense in the top two places on 80% of occasions.