Improved Chinese-Japanese phrase-based MT quality using an extended quasi-parallel corpus

Abstract

State-of-the-art phrase-based machine translation (MT) systems usually demand large parallel corpora in the step of training. The quality and the quantity of the training data exert a direct influence on the performance of such translation systems. The lack of open-source bilingual corpora for a particular language pair results in lower translation scores reported for such a language pair. This is the case of Chinese-Japanese. In this paper, we propose to build an extension of an initial parallel corpus in the form of quasi-parallel sentences, instead of adding new parallel sentences. The extension of the initial corpus is obtained by using monolingual analogical associations. Our experiments show that the use of such quasi-parallel corpora improves the performance of Chinese-Japanese translation systems.