Constituent reordering and syntax models for english-to- Japanese statistical machine translation

Abstract

We present a constituent parsing-based reordering technique that improves the performance of the state-of-the-art English- to-Japanese phrase translation system that includes distortion models by 4.76 BLEU points. The phrase translation model with reordering applied at the pre-processing stage outperforms a syntax- based translation system that incorporates a phrase translation model, a hierarchical phrase-based translation model and a tree-to-string grammar. We also show that combining constituent reordering and the syntax model improves the translation quality by additional 0.84 BLEU points.