Leveraging Translations for Word Sense Disambiguation

Abstract

Word sense disambiguation (WSD) is one of the core tasks in natural language processing and its objective is to identify the sense of a content word (nouns, verbs, adjectives, and adverbs) in context, given a predefined sense inventory. Although WSD is a monolingual task, it has been conjectured that multilingual information, e.g., translations, can be helpful. However, existing WSD systems rarely consider multilingual information, and no effective method has been proposed for improving WSD with machine translation. In this thesis, we propose methods of leveraging translations from multiple languages as a constraint to boost the accuracy of existing WSD systems. Since it is necessary to identify word-level translations from translated sentences, we also develop a novel knowledge-based word alignment algorithm, which outperforms an existing word alignment tool in our intrinsic and extrinsic evaluations. Since our approach is language-independent, we perform WSD experiments on standard benchmark datasets representing several languages. The results demonstrate that our methods can consistently improve the performance of various WSD systems, and obtain state-of-the-art results in both English and multilingual WSD.