Hi Raimou,
We use our statistical technology for Arabic. That said, we do use syntactic information to inform the translations (our "treelet" system, which uses a dependency parser). If you're looking for a more technical review of the technology that underpins
how we use syntax to inform the translation, see the paper by Quirk, Menezes, and Cherry for ACL:
http://acl.ldc.upenn.edu/P/P05/P05-1034.pdf.
Best,
Will Lewis
Microsoft Translator Team