Unsupervised Multilingual Word Sense Disambiguation via an Interlingua

Book One

Volume

Proceedings of the AAAI Conference on Artificial Intelligence, 20

Natural Language Processing and Speech Recognition

We present an unsupervised method for resolving word sense ambiguities in one language by using statistical evidence assembled from other languages. It is crucial for this approach that texts are mapped into a language-independent interlingual representation. We also show that the coverage and accuracy resulting from multilingual sources outperform analyses where only monolingual training data is taken into account.