JU-SKNSB: Extended WordNet Based WSD on the English All-Words Task at SemEval-1

Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 203–206, Prague, June 2007. c⃝2007 Association for Computational Linguistics JU-SKNSB: Extended WordNet Based WSD on the English All-Words Task at SemEval-1 Sudip Kumar Naskar Computer Sc. & Engg. Dept., Jadavpur University, Kolkata, India sudip.naskar@gmail.com Sivaji Bandyopadhyay Computer Sc. & Engg. Dept., Jadavpur University, Kolkata, India sivaji_cse_ju@yahoo.com Abstract This paper presents an Extended WordNet based word sense disambiguation system using a major modification to the Lesk al- gorithm. The algorithm tries to disambigu- ate nouns, verbs and adjectives. The algo- rithm relies on the POS-sense tagged syn- set glosses provided by the Extended WordNet. The basic unit of disambiguation of our algorithm is the entire sentence un- der consideration. It takes a global ap- proach where all the words in the target sentence are simultaneously disambigu- ated. The context includes previous and next sentence. The system assigns the de- fault WordNet first sense to a word when the algorithm fails to predict the sense of the word. The system produces a precision and recall of .402 on the SemEval-2007 English All-Words test data. 1 Introduction In Senseval 1, most of the systems disambiguating English words, were outperformed by a Lesk vari- ant serving as baseline(Kilgariff & Rosenzweig, 2000). On the other hand, during Senseval 2 and Senseval 3, Lesk baselines were outperformed by most of the systems in the lexical sample track (Edmonds, 2002). In this paper, we explore variants of the Lesk al- gorithm on the English All Words SemEval 2007 test data (465 instances), as well as on the first 10 Semcor 2.0 files (9642 instances). The proposed WSD algorithm is POS-sense-tagged gloss (from Extended WordNet) based and is a major modifi- cation of the original Lesk algorithm. 2 Extended WordNet The eXtended WordNet (Harabagiu et al., 1999) project aims to transform the WordNet glosses into a format that allows the derivation of additional semantic and logic relations. It intends to syntacti- cally parse the glosses, transform glosses into logi- cal forms and tag semantically the nouns, verbs, adjectives and adverbs of the glosses automati- cally. The last release of the Extended WordNet is based on WordNet 2.0 and has three stages: POS tagging and parsing, logic form transformation, and semantic disambiguation. 3 Related Works Banerjee and Pedersen (2002) reports an adapta- tion of Lesk’s dictionary-based WSD algorithm which makes use of WordNet glosses and tests on English lexical sample from SENSEVAL-2. They de- fine overlap as the longest sequence of one or more consecutive content words that occurs in both glosses. Each overlap contributes a score equal to the square of the number of words in the overlap. A version of Lesk algorithm in combination with WordNet has been reported for achieving good results in (Ramakrishnan et al., 2004). Vasilescu et al. (2004) carried on a series of ex- periments on the Lesk algorithm, adapted to WordNet, and on some variants. They studied the effect of varying the number of words in the con- texts, centered around …

Content truncated to 500 words