Lexical Constellations and the Structure of Meaning: A Prototype Application to WSD

Abstract

This paper presents the initial stages of a WSD system based on Lexical Constellations. The system pursues two priorities: first, minimize computational costs, and second, deal with different degrees of sense granularity. Computationally, this model has the advantage of involving relatively low-dimensional feature space, because it runs on raw contextual data. We use discriminant function analysis as it allows us to compute distances between each occurrence and each semantic class; for each meaning, we determine the location of the point (group centroids) that represents the means for all variables (collocational data) and for each case we then compute the distances (of the respective case) from each of the group centroids. Finally, we classify cases as belonging to the group (meaning) to which it is closest. The transition from coarse-grained senses to finer-grained ones can be achieved by means of reiteration of the same algorithm on different levels of contextual differentiation.

This research is financially supported by the Spanish Ministry of Education and Science and by the Murcian Government. The first research project (Ref.: HUM2004-00080/FILO) is funded by the Plan Nacional de Investigación Científica, Desarrollo e Innovación Tecno lógica. The second project (00481/PI/04) is funded by Fundación Séneca, Comunidad Autó noma de la Región de Murcia.