IMPROVING WORD SENSE DISAMBIGUATION ACCURACY Ergin ALTINTAŞ

 

In this thesis, previous approaches to the problem of word sense disambiguation are reviewed and some new ideas to improve the accuracy in word sense disambiguation are presented. These ideas consist of a new conceptual hierarchy-based semantic similarity measure and a suggestion to use weighting functions.

The introduced similarity measure tries to improve over the Leacock and Chodorow's measure, using the specificity information derived from WordNet's concept hierarchy. The proposed weighting functions are based on the assumption that the words having different distances from the target word should have different impacts to target word's correct sense.

The similarity measure and the weighing functions are evaluated using the Maximum Relatedness Disambiguation Algorithm that assigns a target word the sense that is most similar to the sense of neighboring words in its local context.