Reinforcement Learning of Question-Answering Dialogue Policies for Virtual Museum Guides