Use of linguistic features for improving English-Persian SMT

In this paper, we investigate the effects of using linguistic information for improvement of statistical machine translation for English-Persian language pair. We choose POS tags as helping linguistic feature. A monolingual Persian corpus with POS tags is prepared and variety of tags is chosen to be small. Using the POS tagger trained on this corpus, we apply a factored translation model. We also create manual reordering rules that try to harmonize the order of words in Persian and English languages. In the experiments, factored translation model shows better performance compared to unfactored model. Also using the manual rules,which just contain few local reordering rules, increases the BLEU score compared to monotone distortion model.