Using LazyBoosting for Word Sense Disambiguation

This paper describes the architecture and results of the TALP system presented at the SENSEVAL-2 exercise for the English lexical–sample task. This system is based on the LazyBoosting algorithm for Word Sense Disambiguation (Escudero et al., 2000), and incorporates some improvements and adaptations to this task. The evaluation reported here includes an analysis of the contribution of each component to the overall system performance. 1 System Description The TALP system has been developed on the basis of LazyBoosting (Escudero et al., 2000), a boosting–based approach for Word Sense Disambiguation. In order to better fit the SENSEVAL2 domain, some improvements have been made on the basic system, including: features that take into account domain information, an specific treatment of multiwords, and a hierarchical decomposition of the multiclass classification problem, similar to that of (Yarowsky, 2000). All these issues will be briefly described in the following sections.