EXPLOITING TAGGED AND UNTAGGED CORPORA FOR WORD SENSE DISAMBIGUATION

Abstract

Traditional supervised methods to sense disambiguation require a lot of sense tagged examples that are often difficult, expensive, or time consuming to obtain. Moreover, if there are no tagged examples for a sense (e.g., a domain specific sense) in the sense tagged corpus, then sense taggers built on this corpus using traditional learning technique will mis-tag the instances with the missed sense. We investigate a series of novel machine learning approaches on benchmark corpora for sense disambiguation and empirically compare them with other related state of the art sense disambiguation methods. They address following questions: How to automatically estimate the number of senses (or sense number, model order) of an ambiguous word from an untagged corpus? (Minimum Description Length criterion); How to use untagged corpora to build a better sense tagger? (label propagation); How to perform sense disambiguation with an incomplete sense tagged corpus? (partially supervised learning).