Combining Supervised-Unsupervised Methods for Word Sense Disambiguation

This paper presents a method to combine two unsupervised methods (Specification Marks, Conceptual Density) and one supervised (Maximum Entropy) for the automatic resolution of lexical ambiguity of nouns in English texts. The main objective is to improved the accuracy of knowledge-based methods with statistical information supplied by the corpus-based method. We explore a way of combining the classification results of the three methods: “voting” is the way we have chosen to combine the three methods in one unique decision. These three methods have been applied both individually as in a combined way to disambiguate a set of polysemous words. Our results show that a combination of different knowledge-based methods and the addition of statistical information from a corpus-based method might eventually lead to improve accuracy of first ones.