Statistical Corpus-Based Word Sense Disambiguation: Pseudowords vs. Real Ambiguous Words

Abstract

In this paper we investigate whether the task of disambiguating pseudowords (artificial ambiguous words) is comparable to the disambiguation of real ambiguous words. Since the two methods are inherently different, a direct comparison is not possible. An indirect approach is taken where the setup for both systems is as similar as possible, i.e. using the same corpus and settings. The results obtained clearly indicate that the tasks are quite different. We conclude that the current practice of using pseudowords cannot be taken as a substitute for testing with real ambiguous words. 1