"Evaluation Functions for Word Sense Disambiguation"

Supervised by Dr RM Stevenson

Abstract

Word Sense Disambiguation is the process of automatically identifying the intended meaning of such a word in context. Various techniques have been proposed to solve this problem. One firstly introduced by Lesk (1986) is known as dictionary definition overlap, which seeks the most shares of the neighbouring words in context. This project discusses the variations of Lesk algorithm, especially Simulated Annealing method (Cowie et al. 1992). Rather than using traditional dictionaries, we introduce the lexical database WordNet, which not only stores words and definitions as the normal dictionaries, but also groups the related words together as individual synonym sets (or synsets). Finally, we will evaluate our algorithm by using the English lexical sample data from the SemCor Word Sense Disambiguation exercise together with other methods.