This paper introduces an unsupervised vector approach to disambiguate words in biomedical text that can be applied to all-word disambiguation. We explore using contextual information from the Unified Medical Language System (UMLS) to describe the possible senses of a word. We experiment with automatically creating individualized stoplists to help reduce the noise in our dataset. We compare our results to SenseClusters and Humphrey et al. (2006) using the NLM-WSD dataset and with SenseClusters using conflated data from the 2005 Medline Baseline. .
