Text this: Improving single and multi-agent deep reinforcement learning methods