Cross-Lingual Word Sense Clustering for Sense Disambiguation

Translation is one of the areas where word disambiguation must be solved in order to find out adequate translations for such words in the contexts where they occur. In this paper, a Word Sense Disambiguation (WSD) approach using Word Sense Clustering within a cross-lingual strategy is proposed. Available sentence-aligned parallel corpora are used as a reliable knowledge source. English is taken as the source language, and Portuguese, French or Spanish as the targets. Clusters are built based on the correlation between senses, which is measured by a language-independent algorithm that uses as features the words near the ambiguous word and its translation in the parallel sentences, together with their relative positions. Clustering quality reached 81% (V-measure) and 92% (F-measure) in average for the three language pairs. Learned clusters are then used to train a support vector machine, whose classification results are used for sense disambiguation. Classification tests showed an average (for the three languages) F-measure of 81%.