|Addresses: www mail|
GIZA++, which is to obtain the word alignment, has been the most popular used toolkit in statistical machine translation research field. To improve the quality of word alignment and the lexical translation performance, an word alignment approach based on Maximum Matching Method (MMM) and similarity is proposed. The algorithm applied in this presentation initially allows the words and phrases of parallel English and Chinese sentences are detected based on MMM, and then candidate alignment results are gotten by the constraint of both a dictionary and statistical GIZA++ results. Empirical study demonstrates that the proposed method gives a better alignment result than that of the GIZA++. Recently, one online system based on the word alignment method and Moses decoder has been developed, which is demonstrated in http://nlp2ct.sftw.umac.mo/MT/.