Enriching EWN with Syntagmatic Information by means of WSD

Iulia Nica (1,2), Mª Antònia Martí (1), Andrés Montoyo (3) and Sonia Vázquez (3)

(1) CLiC - Centre de Llenguatge i Computació, Department of General Linguistics, University of Barcelona, Spain; (2) Department of General Linguistics, University of Iasi, Romania; (3) Research Group of Language Processing and Information Systems, Department of Software and Computing Systems, University of Alicante, Spain

O8-W

Word Sense Disambiguation confronts with the lack of syntagmatic information associated to word senses. In the present work we propose a method for the enrichment of EuroWordNet with syntagmatic information, by means of the WSD process itself. We consider that an ambiguous occurrence drastically reduces its ambiguity when considered together with the words it establishes syntactic relations in the sentence: the claim of "quasi one sense per syntactic relation". On this hypothesis, we obtain sense-tagged syntactic patterns for an ambiguous word intensively using the corpus, with the help of EWN and of associated WSD algorithms. For an occurrence disambiguation, we also consider the whole sentential context where we apply the same WSD algorithms, and combine the sense proposals from the syntactic patterns with the ones from the sentential context. We evaluate the hole WSD method on the nouns in the Spanish Senseval-2 exercise and also the utility of the syntactic patterns for the sense assignment. The annotated patterns we obtain in the WSD process are incorporated into EWN, associated to the synset of the assigned sense. As the syntactic pattern repeat themselves in the text, if sense-tagged, they are a valuable information for future WSD tasks.

EuroWordNet, Word Sense Disambiguation

579.pdf