An Enhanced Lesk Word Sense Disambiguation Algorithm through a Distributional Semantic Model

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 1591–1600, Dublin, Ireland, August 23-29 2014. An Enhanced Lesk Word Sense Disambiguation Algorithm through a Distributional Semantic Model Pierpaolo Basile Annalina Caputo Department of Computer Science, University of Bari Aldo Moro Via E. Orabona, 4, Bari - 70125 Italy {pierpaolo.basile,annalina.caputo,giovanni.semeraro}@uniba.it Giovanni Semeraro Abstract This paper describes a new Word Sense Disambiguation (WSD) algorithm which extends two well-known variations of the Lesk WSD method. Given a word and its context, Lesk algorithm exploits the idea of maximum number of shared words (maximum overlaps) between the context of a word and each definition of its senses (gloss) in order to select the proper meaning. The main contribution of our approach relies on the use of a word similarity function defined on a distribu- tional semantic space to compute the gloss-context overlap. As sense inventory we adopt Babel- Net, a large multilingual semantic network built exploiting both WordNet and Wikipedia. Besides linguistic knowledge, BabelNet also represents encyclopedic concepts coming from Wikipedia. The evaluation performed on SemEval-2013 Multilingual Word Sense Disambiguation shows that our algorithm goes beyond the most frequent sense baseline and the simplified version of the Lesk algorithm. Moreover, when compared with the other participants in SemEval-2013 task, our approach is able to outperform the best system for English. 1 Introduction Unsupervised Word Sense Disambiguation (WSD) algorithms aim at resolving word ambiguity with- out the use of annotated corpora. Among these, two categories of knowledge-based algorithms gained popularity: overlap- and graph-based methods. The former owes its success to the simple intuition un- derlying that family of algorithms, while the diffusion of the latter started growing after the development of semantic networks. The overlap-based algorithms stem from the Lesk (1986) one, which inspired a whole family of meth- ods that exploit the number of common words in two sense definitions (glosses) to select the proper meaning in a context. Glosses play a key role in Lesk algorithm, which exploits only two types of in- formation: 1) the set of dictionary entries, one for each possible word meaning, and 2) the information about the context in which the word occurs. The idea is simple: given two words, the algorithm selects those senses whose definitions have the maximum overlap, i.e. the highest number of common words in the definition of the senses. In order to extract definitions, Lesk adopted the Oxford Advanced Learner’s dictionary. This approach suffers from two problems: 1) complexity, the number of comparisons in- creases combinatorially with the number of words in a text; and 2) definition expressiveness, the overlap is based only on word co-occurrences in glosses. The first problem was tackled by a “simplified” version of Lesk algorithm (Kilgarriff and Rosenzweig, 2000), which disambiguated each word separately. Given a word, the meaning whose gloss shows the maximum overlap with the current context, represented by the surrounding words, is selected. The simplified Lesk significantly outperforms the original Lesk al- gorithm as proved by Vasilescu et al. (2004). …

Content truncated to 500 words