RWTH Aachen University亚深工业大学.pdfVIP

The RWTH System Combination System for WMT 2009 Gregor Leusch, Evgeny Matusov, and Hermann Ney RWTH Aachen University Aachen, Germany Abstract RWTH participated in the System Combi- nation task of the Fourth Workshop on Sta- tistical Machine Translation (WMT 2009). Hypotheses from 9 GermanEnglish MT Figure 1: The system combination architecture. systems were combined into a consen- sus translation. This consensus transla- system prior weights and a language model (LM). tion scored 2.1% better in BLEU and 2.3% The single best path in this CN then constitutes the better in TER (abs.) than the best sin- consensus translation. gle system. In addition, cross-lingual output from 10 French, German, and 2.1 Word Alignment SpanishEnglish systems was combined The proposed alignment approach is a statistical into a consensus translation, which gave one. It takes advantage of multiple translations for an improvement of 2.0% in BLEU/3.5% in a whole corpus to compute a consensus translation TER (abs.) over the best single system. for each sentence in this corpus. It also takes ad- vantage of the fact that the sentences to be aligned 1 Introduction are in the same language. For each source sentence in the test corpus, The RWTH approach to MT system combination we select one of its translations is a refined version of the ROVER approach in as the primary hypothesis. Then we ali