Manipuri-English Bidirectional Statistical Machine Translation Systems using Morphology and Dependency Relations

The present work reports the development of Manipuri-English bidirectional statistical machine translation systems. In the English-Manipuri statistical machine translation system, the role of the suffixes and dependency relations on the source side and case markers on the target side are identified as important translation factors. A parallel corpus of 10350 sentences from news domain is used for training and the system is tested with 500 sentences. Using the proposed translation factors, the output of the translation quality is improved as indicated by baseline BLEU score of 13.045 and factored BLEU score of 16.873 respectively. Similarly, for the Manipuri English system, the role of case markers and POS tags information at the source side and suffixes and dependency relations at the target side are identified as useful translation factors. The case markers and suffixes are not only responsible to determine the word classes but also to determine the dependency relations. Using these translation factors, the output of the translation quality is improved as indicated by baseline BLEU score of 13.452 and factored BLEU score of 17.573 respectively. Further, the subjective evaluation indicates the improvement in the fluency and adequacy of both the factored SMT outputs over the respective baseline systems.