Applying a Naive Bayes Similarity Measure to Word Sense Disambiguation

We replace the overlap mechanism of the Lesk algorithm with a simple, general-purpose Naive Bayes model that mea-sures many-to-many association between two sets of random variables. Even with simple probability estimates such as max-imum likelihood, the model gains signifi-cant improvement over the Lesk algorithm on word sense disambiguation tasks. With additional lexical knowledge from Word-Net, performance is further improved to surpass the state-of-the-art results.