Simple window selection strategies for the simplified lesk algorithm for word sense disambiguation

Abstract

The Simplified Lesk Algorithm (SLA) is frequently used for word sense disambiguation. It disambiguates by calculating the overlap of a set of dictionary definitions (senses) and the context words. The algorithm is simple and fast, but it has relatively low accuracy. We propose simple strategies for the context window selection that improve the performance of the SLA: (1) constructing the window only with words that have an overlap with some sense of the target word, (2) excluding the target word itself from matching, and (3) avoiding repetitions in the context window. This paper describes the corresponding experiments. Comparison with other more complex knowledge-based algorithms is presented. © Springer-Verlag 2013.