Lexicalized Reordering Models for English-Myanmar Phrase-based Statistical Machine Translation

Abstract

Reordering is extremely desirable for translation accuracy when translating between high disparity language pairs in word order. The aim of this paper is the comparative study of lexicalized reordering models (LRM) by Moses to investigate the translation performance for English-Myanmar statistical machine translation (SMT) system. The studied methods are word-based, phrase-based and hierarchical phrase-based LRM by using various orientations and distortion limits. This reordering model calculates reordering probability conditioned on the word of each phrase pair. We applied Moses phrase-based SMT (PBSMT) system to make experiments for the variants of LRM and evaluated the BLEU and RIBES scores to measure the performance of machine translation. According to this experiments, hierarchical phrase-based reordering model in MSD orientation gives the highest scores in English-Myanmar SMT system.