Text this: Deep variational reinforcement learning for POMDPs