As interaction between speakers of different languages continues to increase, the everpresent problem of language barriers must be overcome. For the same reason, automatic language translation (Machine Translation) has become an attractive area of research and development. Statistical Machine Translation (SMT) has been used for translation between many language pairs, the results of which have shown considerable success. The focus of this research is on the English/Persian language pair. This paper investigates the development and evaluation of the performance of a statistical machine translation system by building a baseline system using subtitles from Persian films. We present an overview of previous related work in English/Persian machine translation, and examine the available corpora for this language pair. We finally show the results of the experiments of our system using an in-house corpus and compare the results we obtained when building a language model with different sized monolingual corpora. Different automatic evaluation metrics like BLEU, NIST and IBM-BLEU were used to evaluate the performance of the system on half of the corpus built. Finally, we look at future work by outlining ways of getting highly accurate translations as fast as possible.
展开▼
