| This thesis presents a two-step translation system with post-translation word re-ordering, which improves the grammatical and semantic accuracy of Korean to English statistical machine translation (SMT). Training and testing procedures for the translation system with post-ordering are explained, and results are given versus a phrase-based SMT baseline. To the author's knowledge, this is the first paper to explore post-ordering for Korean to English SMT, as well as the first to present Korean to English SMT results using the RIBES evaluation metric. The post-ordering method described in this thesis results in a gain of 3.98 RIBES points and 0.97 BLEU points over the baseline system, on a test set of 2047 sentences that cover a variety of topics. |
