An approach to knowledge-based Word Sense Disambiguation using semantic trees built on a WordNet lexicon network

This paper proposes a knowledge-based WSD (Word Sense Disambiguation) method derived from the Lesk algorithm. The proposed method considers an extension of the definition domain of the Lesk algorithm by creating a lexicon network from tagged lexicon glosses. We present several methods that adjust the lexicon network in order to better describe the natural language. Further, on the pre-processed lexicon network we build competence and definition semantic trees for each sense that will be used to compute costs of semantic similarity between words senses. For this purpose we use a WSD window limited to a phrase, and a similar reasoning for larger contexts. For testing we apply our methods to the recently WordNet tagged glosses.