Search

A sentence alignment model based on combined clues and Kernel Extensional Matrix Matching (KEMM) method is proposed. In this model, a similarity matrix for sentence aligning is formed by the similarities of bilingual sentences calculated by the combined clues, such as lexicon, morphology, length and special symbols, etc.; then this similarity matrix is used to construct a select matrix for sentence aligning; finally, obtains the sentence alignments by KEMM. Experimental results illustrated that our model outperforms over the Gale's system on handling any types of sentence alignments, with 30% total sentence alignment error rate decreasing.

Pattern extracting is one of key techniques in the field of pattern based machine translation. The step of extracting patterns, which influence the results of machine translation, are in low quality owing to the limitation of the technique of linguistic analysis. To avoid the effects, this paper proposes a direct matching model between the text and the pattern which directly matches the sentences and patterns. As shown in the experiment, the performance of pattern based machine translation system is improved by introducing this model.