Combining Distributional and Paradigmatic information in a lexical substitution task
Diego De Cao, Roberto Basili · 2009
Abstract. This paper describes an unsupervised approach to the Lexical Substitution task as applied to the Evalita 2009 competition. The applied approach builds on an extended Latent Semantic Analysis model that combines distributional and paradigmatic evidence in a unified vector space. All the systems employed in the ART laboratory are based on two main steps: 1) selection of a set of candidate substitute words and 2) ranking of each candidate according to combination of similarity metrics in extended LSA spaces. Comparative validation is reported in this paper.