A New Decision Rule for Statistical Word Sense Disambiguation

Word Sense Disambiguation (WSD) is usually considered to be a pattern classification to research and it has always being a key problem and one of difficult points in natural language processing. Statistical learning theory is a mainstream of the research method for WSD. The distribution of the word-senses of an ambiguous word is always not symmetrical and the distinction between word-senses' emergence frequency is great sometimes, so the judgment results are inclined to the maximum probability word-sense in the word-sense classification. The reflection of this phenomenon is obviously in the Bayesian model. When using the Bayesian model to carry on some research we find a new word-sense decision rule, which have a better precision than Bayesian model in WSD. In order to validate the credibility and stabilization of this method we carry through the experiment time and again, and acquire lots of experiment data. The results of the experiment indicate that new decision rule is more excellent than Bayesian decision rule. Furthermore this paper provides a theoretical foundation for this new decision rule.