These last years have seen the development of statistical approaches for machine translation. Nevertheless, the intrinsic variations of the natural language act upon the quality of statistical models. Studies have shown that in-domain corpora containwords that can occur in out-of-domain corpora (common words), but also contain domain specific words. This particularity can be handled by terminological resources like bilingual lexicons. However, if the vocabulary differs between out and in-domain data, the syntactic and semantic content may also vary. In our work, we consider the task of domain adaptation for statistical machine translation through two majoraxes : bilingual lexicon acquisition and post-edition of machine translation outputs.We evaluate our approaches on the medical domain. The quality of automatic translations in the medical domain are improved and the results are compared to other works in this field. Oracle evaluations tend to show that further gains are still possible
