Improved Reordering Rules for Hierarchical Phrase-Based Translation

Abstract

Hierarchical phrase-based translation model has been proven to be a simple and powerful machine translation model. However, due to the computational complexity constraints, the extraction and use of hierarchical rules are usually restricted under certain limits, and these limits could have a negative impact on the performance of the translation model, especially for reordering. This paper presents a solution to improve the reordering of hierarchical phrase-based translation model. We propose a two-step method to extract improved reordering rules with less limits. These reordering rules help both local and non-local reordering, and could be incorporated to a hierarchical phrase-based translation system easily. Experiments show that our approach achieves statistically significant improvements over the baseline system in Chinese-English translation.