Exploring new features to combine the output of machine translation paradigm

Machine translation can be seen as a part of information and knowledge technology. All current machine translation paradigms have their own shortcomings, and on the other hand their own non-overlapping advantages. Therefore, combining different machine translation systems could help to find or generate a better hypothesis. In this paper, we apply a hypothesis selection method. Using the limited number of features, we rescore hypotheses, and then the hypothesis with the highest score is selected as the final output. Experimental results on Persian-English task shows a significant improvement of 1.13% and 1.19% in BLEU on the tuning and unseen test sets as compared to the best individual system.