Farsi - German statistical machine translation through bridge language

Since statistical machine translation outperforms other approaches in the field of machine translation, we used this approach to develop a Farsi-German and a German-Farsi translation system. In this work, first we used an existing English-German bilingual corpus and then we manually translate a large part of the English corpus to Farsi, in order to have the required training data. Thereafter, we employ the idea of using English as the bridge language to build the Farsi-German translation systems, in addition to build the systems on the corresponding Farsi-German bilingual corpus. Because different amount of recourses exist for Farsi-English and German-English language pairs, we investigate different approaches to combine these two machine translation systems. We will show that the BLEU score of Farsi-German system has increased about 15% relatively compared to the baseline system.