Text this: An environment model in multi-agent reinforcement learning with decentralized training