Intensive Use of Lexicon and Corpus for WSD

The paper addresses the issue of how to use linguistic information in Word Sense Disambiguation (WSD). We introduce a knowledge-driven and unsupervised WSD method that requires only a large corpus previously tagged with POS and very little grammatical knowledge. The WSD process is performed taking into account the syntactic patterns in which the ambiguous occurrence appears, relaying in the hypothesis of “almost one sense per syntactic pattern”. This integration allows us to obtain, from corpora, paradigmatic and syntagmatic information related to the ambiguous occurrence. We also use variants of EWN information for word senses and different WSD algorithms. We report the results obtained when applying the method on the Spanish lexical sample task in Senseval-2. This methodology is easily transportable to other languages.