Diversion of Hierarchical Phrases as Reordering Templates
Mikio Yamamoto, Jyunya Norimatsu, Mitsuru Koshikawa, Takahiro Fukutomi, Taku Nishio, Kugatsu Sadamitsu, Takehito Utsuro, Masao Utiyama, Shunji Umetani, Tomomi Matsui · 2008
In the hierarchical phrase-based translation model (Chiang 2007), translation rules handle both context-sensitive translation and reordering of phrases at the same time. This simultaneity is strengths and weak-nesses of the model. Although it enables the rules to be applied to the accurate and correct context, it deteriorates the applicability of the rules. In other words, the rules work very well in domains of train-ing data, but they lost robustness in out of the do-mains. In this paper, we will try to improve the appli-cability of the original model by adding extra reorder-ing templates which are separated out from hierarchi-cal phrase translation rules. An original hierarchical phrase rule with two non-terminals is regarded as ei-ther monotone or swap reordering template according to if the two non-terminals in the source side have monotone or swap relation to the target side in the original rule. We will describe experiments in which the original model compares with our extensions in BLEU as a metric of translation quality using shared data at the NTCIR-7 patent translation task.