Improving reordering and modeling in statistical machine translation