Levée d’ambiguïtés sur les mots lexicaux et grammaticaux

SummaryAmbiguity is a central problem in computational linguistics. In this paper, we propose a reflexion based on our work of morpho-syntaxic disambiguation with the intex system. intex is an environment of linguistic development which includes large coverage dictionaries constructed at LADL. With the examples of verbs and adjectives, we show that choices on the encoding of linguistic information in the dictionary are central in the definition of ambiguity. The expected accuracy in tags (in linear tagging mode) considerably influences the results of disambiguation. In the second part of the paper, we examine the most frequent ambiguous words in a big corpus, in terms of words or of parts of speech. Based on these figures, we apply our method with local grammars to disambiguate words: determinants or pronouns; names, adjectives or verbs.