The bare minimal would be to implement the Accept-Languages http header correctly, not just a randomly picked subset in it. They cannot seem to even bother to do that.