Contextual semantics for WSD

Contextual Semantics for WSD ERIC CRESTAN(1,2) (1) Sinequa SAS 51-54, rue Ledru-Rollin 92400 Ivry-sur-Seine, France Crestan@sinequa.com (2) Laboratoire Informatique d’Avignon B.P. 1228 Agroparc 339 Chemin des Meinajaries 84911 Avignon Cedex 9, France Abstract For Sinequa’s second participation to the Senseval evaluation, two systems using contextual semantic have been proposed. Based on different approaches, they both share the same data preprocessing and enrichment. The first system is a combined approach using semantic classification trees and information retrieval techniques. For the second system, the words from the context are considered as clues. The final sense is determined by summing the weight assigned to each clue for a given example. 1 Introduction In the framework of the Senseval-3 evaluation campaign on Word Sense Disambiguation (WSD), we presented two systems relying on different strategy. The system SynLexEn is an evolution from the system used during the Senseval-2 campaign. It is based on two steps. The first step uses semantic classification trees on a short context size. A decision system based on document similarity is used as second step. The novelty of this system resides in a new vision level on the context. The semantic dictionary of Sinequa is extensively used in this process. The second system, SynLexEn2, is based on weighted clues summation over a short context size. From the training data, a score is computed for each word in a short context size, for each sense. In Section 2, the combined approach system for WSD is presented. We first give an overview of the data pre-processing that was applied (Section 2.1). Then, a brief description of Semantic Classification Trees is given (Section 2.2) along with a description of additional data used for semantic view of short and long context (Section 2.3 and Section 2.4). Next, a semantic information retrieval system used in order to select the appropriate sense is proposed (Section 2.5). Finally, the SynLexEn2 system is presented in Section 3. We then conclude with the evaluation results for both systems in Section 4. 2 Combined approach The SinLexEn system is quite similar to the system used during the last Senseval-2 evaluation campaign (Crestan et al., 2001). It is based on two stages: the first stage uses three Semantic Classification Trees in parallel, trained on different size of context. Then, the second stage brings in a decision system based on information retrieval techniques. The novelty of this approach dwells in the use of semantic resource as conceptual view on extended context in both stages. 2.1 Data pre-processing The first step in order to get the most from the data is to lemmatize and clean sentences. Each paragraph from the training and the test data are first passed though an internal tagger/lemmatizer. Then, some grammatical words are removed such as articles and possessive pronouns. Only one word is not handled in this process, it is the word to be disambiguated. Indeed, previous works (Loupy et al., 1998) have shown that the form of this word could bring interesting clues about its Association for Computational …

Content truncated to 500 words