Text this: Scaling Up Reinforcement Learning without Sacrificing Optimality by Constraining Exploration /