Language models and reranking for machine translation

Abstract

Complex Language Models cannot be easily integrated in the first pass decoding of a Statistical Machine Translation system - the decoder queries the LM a very large number of times; the search process in the decoding builds the hypotheses incrementally and cannot make use of LMs that analyze the whole sentence. We present in this paper the Language Computer's system for WMT06 that employs LMpowered reranking on hypotheses generated by phrase-based SMT systems.