Clustering word senses from semantic mirroring data
Hampus Lilliehöök, Magnus Merkel · 2013
In this article we describe work on creating word clusters in two steps. First, a graph-based approach to semantic mirroring is used to create primary synonym clusters from a bilingual lexicon. Secondly, the data is represented by vectors in a large vector space and a resource of synonym clusters is then constructed by performing K-means centroid-based clustering on the vectors. We evaluate the results automatically against WordNet and evaluate a sample of word clusters manually. Prospects and applications of the approach are also discussed.