Word Sense Disambiguation Using Random Indexing

Abstract

This paper presents the results of an experiment to apply a novel semantic representational formalism called Random Indexing for the supervised word sense disambiguation of English words. Random Indexing uses high-dimensional sparse vectors with random patterns modeling neural activation patterns in the brain to represent linguistic information. The presented learning and disambiguating method was trained and tested using manually sense-tagged corpora available from Senseval. The results are evaluated and compared to previous works using the same corpora, and the possible lacks and weaknesses of Random Indexing are pointed out both in general, both for the purpose of word sense disambiguation.

Resources

Details