Hierarchical pre-reordering model for patent machine translation

Renfen Hu, Kai Zhao, Hongzheng Li, Yun Zhu, Yaohong Jin

Abstract: Patents constitute one of the most challenging domains for machine translation because patent sentences can be quite long with complex structures. This paper presents a hierarchical reordering method based on three-level parsing for Chinese-English patent MT. After integrating into a Phrase based SMT system, our method improves the BLEU score by 1.92. There are also improvements in NIST and METEOR scores. Besides, we investigate the precision and recall of the reordering rule set by conducting manual evaluations.