Enhancing Translation from English to Arabic Using Two-Phase Decoder Translation

This paper describes an approach to enhance statistical machine translation. This approach uses a two-phase decoder system; the first decoder which we call the initial decoder translates from English to Arabic and the second is a post-processing decoder that re-translates the initial's decoder Arabic output to Arabic again to fix some of the translation errors. This new technique showed to be useful when trying to translate corpus from a different context other than the original corpus used in the training of the initial decoder. We recorded a BLEU score enhancement on out-of-context corpus close to 10 BLEU points on UN corpus and 2 BLEU points on TED 2013 corpus.