Naive Bayes and Exemplar-based Approaches to Word Sense Disambiguation Revisited