The effect of windowing in word sense disambiguation

Abstract

In this paper, the effect of different windowing schemes to the success rate of word sense disambiguation is probed. In these windowing schemes it is considered that the impact of a neighbor word to the correct sense of the target word should be somewhat related to it's distance to the target word. Several weighting functions are evaluated for their performance in representing this relation, Two semantic similarity measures, one of which is introduced by the authors of this paper, are used in a modified version of Maximum Relatedness Disambiguation algorithm for the experiments. This approach yielded improvements up to 4.24% in word sense disambiguation accuracy.