Word Sense Induction for Machine Translation

We have witnessed the research progress of machine translation from phrase/syntax-based to semanticsbased and from single sentence-based to discourse and document-based. This talk presents our work of word sense-based translation model for statistical machine translation, which is one of semantics-based SMT research at word sense level. The sense in which a word is used determines the translation of the word. The talk begins with how to build a broad-coverage sense tagger based on a nonparametric Bayesian topic model that automatically learns sense clusters for words in the source language, and then focuses on the proposed word sense-based translation model that enables the decoder to select appropriate translations for source words according to the inferred senses for these words using maximum entropy classifiers. The talk ends with experiential results and some conclusions. To the best of our knowledge, this is the first attempt to empirically verify the positive impact of lexical semantics (word sense) on translation quality. This is a joint work with Deyi Xiong, Soochow University.