Abstract
Genetic algorithms (GAs) have widely been investigated to solve hard optimization problems, including the word sense disambiguation (WSD). This problem asks to determine which sense of a polysemous word is used in a given context. The performance of a GA may drastically vary with the description of its genetic operators and selection methods, as well as the tuning of its parameters. In this paper, we present a self-adaptive GA for the WSD problem with an automated tuning of its crossover and mutation probabilities. The experimental results obtained on standard corpora (Senseval-2 (Task#1), SensEval3 (Task#1), SemEval-2007 (Task#7)) show that the proposed algorithm significantly outperformed a GA with standard genetic operators in terms of precision and recall.