Robust utilization of context in word sense disambiguation

Abstract

Context is the only means to identify the sense of a polysemous word. All algorithms for word sense disambiguation make use of information within a context window of the target word. What is the best window size for word sense disambiguation has been long a problem. Different contexts generally give different results even for a same algorithm. In this paper, we exploit an algorithm which is more robust with the varying of different context used. This method aims to lower the uncertainty brought by classifiers using different context window sizes and make more robust utilization of context while perform well. Experiments show our approach outperforms some other algorithms on both robustness and performance. © Springer-Verlag Berlin Heidelberg 2005.