The Pre-ordering Model for Statistical Machine Translation of Enhancing the N-best Syntactic Knowledge

Syntactic heterogeneity between source and target languages has an important impact on the performance of Statistical Machine Translation (SMT). On the basis of phrase-based Chinese-English SMT, a method of source language pre-ordering based on N-best syntactic knowledge enhancement is proposed. First, the source language input sentences are analyzed by N-best Syntax, and the high reliability sub-tree structure is obtained by calculating statistical probability. Two optimization strategies are used to optimize the initial rule set: rule deduction and rule probability threshold control mechanism. Second, the source language phrase translation table is used as a constraint to control the sequence between phrases. Finally, the syntax analysis tree of the source-side sentences is pre-ordered. The experimental results of Chinese-English SMT based on the NIST 2005 and 2008 test data sets show that comparing to the baseline system, the BLEU score of automatic evaluation criterion of the N-best syntactic knowledge-enhanced SMT pre-ordering method increased by 0.68 and 0.83 respectively.