A clustering-based Approach for Unsupervised Word Sense Disambiguation

Clustering methods have been extensively used in many Information Processing tasks in order to capture unknown object categories. However, clustering has been scarcely used as a sense labeling method for Word Sense Disambiguation (WSD), that is, as a way to identify groups of semantically related word senses that can be successfully used in a disambiguation process. In this paper, we present an unsupervised disambiguation method relying on word sense clustering that also reveals the implicit relationships (not asserted in WordNet) existing among these word senses. We also investigate in depth the role of clustering and its contribution to WSD. Experimental results demonstrate the usefulness of clustering for unsupervised WSD.