Quality Translation Enhancement Using Sequence Knowledge and Pruning in Statistical Machine Translation

Abstract

Machine translation has two important parts, a learning process which followed by a translation process. Unfortunately, most of the translation process requires complex operations and in-depth knowledge of the languages in order to give a good quality translation. This study proposes a better approach, which does not require in-depth knowledge of the linguistic properties of the languages, but it produces a good quality translation. This study evaluated 28 different parameters in IRSTLM language modeling, which resulting 270 millions experiments, and proposes a sequence evaluation mechanism based on a maximum evaluation of each parameter in producing a good quality translation based on NIST and BLEU. The parallel corpus and statistical machine learning for English and Bahasa Indonesia were used in this study. The pruning process, user interface, and the personalization of translation have a very important role in implementing of this machine translation. The result is quite promising. It shows that pruning process increases of the translation process time. The particular sequence knowledge/value parameter in translation process has a better performance than the other method using in-depth linguistic knowledge approaches. All these processes, including the process of parsing from a stand-alone mode to an online mode, are also discussed in detail.