Reinforcement learning approaches to natural language generation in interactive systems