Text this: Bayesian curriculum generation in sparse reward reinforcement learning environments