Reinforcement learning for parameter estimation in statistical spoken dialogue systems