摘要：This paper extends a greedy decoder for statistical machine translation (SMT), which searches for an optimal translation by using SMT models starting from a decoder seed, i.e., the source language input paired with an initial translation hypothesis. First, the outputs generated by multiple translation engines are utilized as the initial translation hypotheses, whereby their variations reduce local optima problems inherent in the search. Second, a rescoring method based on the edit-distance between the initial translation hypothesis and the outputs of the decoder is used to compensate for problems of conventional greedy decoding solely based on statistical models. Our approach is evaluated for the translation of dialogues in the travel domain, and the results show that it drastically improves translation quality.
