In Search of a New Model for Wsd: a Multi-layered Lexical Constellation Model

Research on WSD has been rather intensive in the last quarter of the 20th c. (Agirre & Edmond 2006). Difficulties for a successful disambiguation algorithm lie mainly in polysemous words with high granularity. A research project recently conducted by Sanchez, Cantos, Almela (Sánchez, Cantos & Almela 2007; 2008), grounded on a supervised method, proved that the success rate is clearly lower in words with high granularity, compared to the percentage of success in words with poor semantic granularity. The analysis of the results reveals that failure in the identification of the meaningful contextual features is the main reason leading to disambiguation errors. The analysis concludes that the model of meaning organization used by most WSD algorithms is not adequate to successfully deal with fine-grained sense distinctions. The model of lexical constellation is advanced and sketched as a possibility to initiate a break through in WSD. Its non-linear character and three dimensionality fits better the non-linear and hierarchical structure of meaning organization in real language.