tailieunhanh - Báo cáo khoa học: "A Hybrid Relational Approach for WSD – First Results"

We present a novel hybrid approach for Word Sense Disambiguation (WSD) which makes use of a relational formalism to represent instances and background knowledge. It is built using Inductive Logic Programming techniques to combine evidence coming from both sources during the learning process, producing a rule-based WSD model. We experimented with this approach to disambiguate 7 highly ambiguous verbs in EnglishPortuguese translation. | A Hybrid Relational Approach for WSD - First Results Lucia Specia Núcleo Interinstitucional de Linguistica Computational - ICMC - University of São Paulo Caixa Postal 668 13560-970 São Carlos SP Brazil lspecia@ Abstract We present a novel hybrid approach for Word Sense Disambiguation WSD which makes use of a relational formalism to represent instances and background knowledge. It is built using Inductive Logic Programming techniques to combine evidence coming from both sources during the learning process producing a rule-based WSD model. We experimented with this approach to disambiguate 7 highly ambiguous verbs in English-Portuguese translation. Results showed that the approach is promising achieving an average accuracy of 75 which outperforms the other machine learning techniques investigated 66 . 1 Introduction Word Sense Disambiguation WSD is concerned with the identification of the correct sense of an ambiguous word given its context. Although it can be thought of as an independent task its importance is more easily realized when it is applied to particular tasks such as Information Retrieval or Machine Translation MT . In MT the application we are focusing on a WSD or translation disambiguation module should identify the correct translation for a source word when options with different meanings are available. As shown by Vickrey et al. 2005 we believe that a WSD module can significantly improve the performance of MT systems provided that such module is developed following specific requirements of MT . employing multilingual sense repositories. Differences between monolingual and multilingual WSD are very significant for MT since it is concerned only with the ambiguities that appear in the translation Hutchins and Sommers 1992 . In this paper we present a novel approach for WSD designed focusing on MT. It follows a hybrid strategy . knowledge and corpus-based and employs a highly expressive relational formalism to represent both the examples .