A study of classifier combination and semi-supervised learning for word sense disambiguation
Anh-Cuong Le · Institutional Repositories DataBase (IRDB) · 2007
Word Sense Disambiguation (WSD) involves the association of a polysemous word in a text or discourse with a particular sense among numerous potential senses of that word.This is an "intermediate task" necessary to accomplish most natural language processing tasks.It is obviously essential for language understanding application, such as message understanding and human-machine communication; it is also at least helpful for other applications whose aim is not language understanding, such as machine translation and information retrieval, among others.The automatic disambiguation of word senses has been an interest and concern since the 1950s [Ide et al. (1998)].Although there have been many studies investigated on various methods for this problem, the performance of available WSD systems or published results are limited (accuracy around 70%).Therefore, WSD is still an open problem and is a challenge in Natural Language Processing (NLP) community.Nowadays, with the strong and fast development of machine learning methods and their success in applying to many NLP tasks, the use of machine learning techniques in WSD has been becoming more interest and attractive.This thesis also lies in this research direction, in which we present a study of classifier combination and semi-supervised learning for WSD.In addition, we also work on context representation and feature selection which play important roles in obtaining high accuracy of WSD task.Particularly, the following three problems are targeted in this research.