Lexical Micro-adaptation in Statistical Machine Translation

We introduce a generic framework in Statistical Machine Translation (SMT) in which lexical hypotheses, in the form of a target language model local to the input sentence, are used to guide the search for the best translation, thus performing a lexical microadaptation. An instantiation of this framework is presented and evaluated on three language pairs, where these auxiliary hypotheses are derived through triangulation via an auxiliairy language. Our first experiments consider nine auxiliary languages, allowing us to measure their individual contribution. We then combine all their hypotheses through a decoding by consensus. Our experiments show that SMT systems can be improved by automatically produced auxiliary hypotheses. MOTS-CLÉS : traduction automatique statistique, traduction par pivot.