A Multi-Agent Approach to Using Redundancy and Reinforcement in Question Answering