On the quality of lexical resources for word sense disambiguation

Abstract

Word Sense Disambiguation (WSD) systems are usually evaluated by comparing their absolute performance, in a fixed experimental setting, to other alternative algorithms and methods. However, little attention has been paid to analyze the lexical resources and the corpora defining the experimental settings and their possible interactions with the overall results obtained. In this paper we present some experiments supporting the hypothesis that the quality of lexical resources used for tagging the training corpora of WSD systems partly determines the quality of the results. In order to verify this initial hypothesis we have developed two kinds of experiments. At the linguistic level, we have tested the quality of lexical resources in terms of the annotators' agreement degree. From the computational point of view, we have evaluated how those different lexical resources affect the accuracy of the resulting WSD classifiers. We have carried out these experiments using three different lexical resources as sense inventories and a fixed WSD system based on Support Vector Machines.