A Rule-Based Source-Side Reordering on Phrase Structure Subtrees

Abstract

Since different languages put words in different orders, reordering is an important issue in statistical machine translation. The paper proposes a rule-based reordering method at the source side as a preprocessing step, which applies some syntactic reordering rules on the phrase structure subtree to reorder source language. The reordering rules integrate the phrase structure tree with part-of-speech tags, which can implement the reordering not only between words but also between words and phrases. And the problems of long-distance reordering and translation errors can be partly solved. Meanwhile, the interference between reordering rules of this method has been significantly reduced in this method. Experiments shows that our method can improve the performance of the state-of-the-art phrase translation models, achieving 1.71 BLEU score increase over the standard phrase-based machine translation system.