DALE: A Word Sense Disambiguation System for Biomedical Documents Trained using Automatically Labeled Examples

Judita Preiss, Mark Stevenson

Abstract: Automatic interpretation of documents is hampered by the fact that language contains terms which have multiple meanings. These ambiguities can still be found when language is restricted to a particular domain, such as biomedicine. Word Sense Disambiguation (WSD) systems attempt to resolve these ambiguities but are often only able to identify the meanings for a small set of ambiguous terms. DALE (Disambiguation using Automatically Labeled Examples) is a supervised WSD system that can disambiguate a wide range of ambiguities found in biomedical documents. DALE uses the UMLS Metathesaurus as both a sense inventory and as a source of information for automatically generating labeled training examples. DALE is able to disambiguate biomedical documents with the coverage of unsupervised approaches and accuracy of supervised methods.