Evaluating n-gram Models for a Bilingual Word Sense Disambiguation Task

Abstract

The problem of Word Sense Disambiguation (WSD) is about selecting the correct sense of an ambiguous word in a given context. However, even if the problem of WSD is difficult, when we consider its bilingual version, this problem becomes much more complex. In this case, it is necessary not only to find the correct translation, but such translation must consider the contextual senses of the original sentence (in the source language), in order to find the correct sense (in the target language) of the source word. In this paper we present a probabilistic model for bilingual WSD based on n-grams (2-grams, 3-grams, 5-grams and k- grams, for a sentence S of a length k). The aim is to analyze the behavior of the system with different representations of a given sentence containing an ambiguous word. We use a Naive Bayes classifier for determining the probability of the target sense (in the target language) given a sentence which contains an ambiguous word (in the source language). For this purpose, we use a bilingual statistical dictionary, which is calculated with Giza++ by using the EUROPARL parallel corpus. On the average, the representation model based on 5-grams with mutual information demonstrated the best performance.