The JAIST machine translation systems for WMT 17

Abstract

We describe the JAIST phrase-based machine translation systems that participated in the news translation shared task of the WMT17. In this work, we participated in the Turkish-English translation, in which only a small amount of bilingual training data is available, so that it is an example of the low-resource setting in machine translation. In order to solve the problem, we focus on two strategies: building a bilingual corpus from comparable data and exploiting existing parallel data based on phrase pivot translation. In order to utilize the strategies to enhance machine translation on the low-resource setting most effectively, we introduce a system combining the extracted corpus, the pivot translation, and the direct training data. Experimental results showed that our combined systems significantly improved the baseline models, which were trained on the small bilingual data.