Resolving Ambiguities in Biomedical Text With Unsupervised Clustering Approaches

This paper explores the effectiveness of unsupervised clustering techniques developed for general English in resolving semantic ambiguities in the biomedical domain. Methods that use first and second order representations of context are evaluated on the National Library of Medicine Word Sense Disambiguation Corpus. We show that the method of clustering second order contexts in similarity space is especially effective on such domain-specific corpora. The significance of the current research lies in the method extension to a new, previously untested domain and the general exploration of method portability across domains.