Text this: Transferring Deep Reinforcement Learning with Adversarial Objective and Augmentation