WORD SENSE CLUSTERING BASED ON TRANSLATION EQUIVALENCE IN PARALLEL TEXTS ; A CASE STUDY IN ROMANIAN

The lexical ambiguity is one of the most difficult problem to solve in the natural language processing. The Word sense disambiguation (wsd) is a task of utmost importance and as such, it is not surprinsing the great interest for this area of research. We argue that the difficulty of solving the problem of sense disambiguation depends to a large extent on the application which requires a solution to this question. For many of the computationally interesting applications of the natural language processing (such as the multilingual intelligent information retrieval, the semantic indexing, the machine translation) the sense distinctions should consider the so-called "strong" senses, namely those that more often than not imply different lexicalisations in another language. We describe our approach based on the concept of translation equivalence and a case-study providing support for claiming that a multilingual approach to wsd is much more reliable and more precise than the traditional monolingual tackling of the problem.