Évaluer Et Améliorer Une Ressource Distributionnelle : Protocole D'annotation De Liens Sémantiques En Contexte

Using distributional analysis methods to compute semantic proximity links between words has become commonplace in NLP. This paper focuses on the issues of evaluating a distributional resource. We consider that setting up an evaluation procedure is a first step towards the characterization of the resource, and a way to improve its overall quality. We then propose a new protocol for in-text annotation of distributional neighbors, which is used to build a reliable reference data set. The data generated are analyzed and used to guide the automatic categorization of distributional links. MOTS-CLÉS : Analyse distributionnelle, ressources lexicales, évaluation.