A central problem in any natural language processing application is determining the meaning of ambiguous words. Word-sense disambiguation (WSD) is often cast as a problem in supervised learning. These approaches assume the availability of text to train a learning algorithm where ambiguous words have been manually annotated with sense distinctions. If such text is available, supervised approaches are effective and we present several extensions to an existing method. However, since sense-tagged text is expensive to create and only exists for a small number of ambiguous words, unsupervised alternatives are presented that do not require such an expensive knowledge source.