Analyzing the potential of source sentence reordering in statistical machine translation