Deep Reinforcement Learning Conditioned on the Natural Language