An improved approach to word sense disambiguation

Words in the English language often correspond to different meanings in different contexts. Such words are referred to as polysemous words i.e. words having more than one sense. This paper presents a knowledge based algorithm for disambiguating polysemous words using computational linguistics tool, WordNet. The task of word sense disambiguation requires finding out the similarity between the target word (word to be disambiguated) and the nearby words (words surrounding the target word in input text). Algorithms in the past have calculated similarity either by finding out the number of common words (intersection) between the glosses (definitions/meanings) of the target and nearby words, or by finding out the exact occurrence of the nearby word´s sense in the hierarchy (hypernyms) of the target word´s senses. This paper proposes an algorithm which modifies the above two parameters by computing intersection using not only the glosses but also by including the related words. Also the intersection is computed for the entire hierarchy of the target and nearby words. It also incorporates a third parameter `distance´ (between target and nearby words). The proposed approach incorporates more parameters for calculating similarity, which has not been attempted by any of the previous approaches. It scores the senses based on the overall impact of three parameters i.e. intersection, hierarchy and distance and then chooses the sense with the highest score. The algorithm has been evaluated on SemCor which is the largest available sense-tagged corpus. The proposed algorithm achieves a precision of 53.12% for Top1 results, 59.91% for Top2 results and that of 62.13% for Top3 results which is better than other knowledge based approaches.