Envoyer par SMS: Reinforcement Learning for Adaptive Dialogue Systems