Dialogue systems rely on a careful reinforcement learning (RL) design: the learning algorithm and its state space representation. In lack of more rigorous knowledge, the designer resorts to its ...