Towards open domain spoken dialogue systems