Going Beyond Word Cooccurrences in Global Lexical Selection for Statistical Machine Translation using a Multilayer Perceptron
Alexandre Patry, Philippe Langlais · 2011
Phrase-based statistical machine translation (PBSMT) decoders translate source sentences one phrase at a time using strong independence assumptions over the source phrases. Translation table scores are typically independent of context, language model scores depend on a few words surrounding the target phrase and distortion models do not influence directly the choice of target phrases. In this work, we propose to condition the selection of each target word on the whole source sentence using a multilayer perceptron (MLP). Our interest in MLP lies in their hidden layer which encodes source sentences in a representation that is not directly tied to the notion of word. We evaluated our approach on an English to French translation task. Our MLP model was able to improve BLEU scores over a standard PBSMT system. 1