A statistical model for near-synonym choice

We present an unsupervised statistical method for automatic choice of near-synonyms when the context is given. The method uses the Web as a corpus to compute scores based on mutual information. Our evaluation experiments show that this method performs better than two previous methods on the same task. We also describe experiments in using supervised learning for this task. We present an application to an intelligent thesaurus. This work is also useful in machine translation and natural language generation.