A Reinforcement Learning approach to evaluating state representations in spoken dialogue systems