Text this: Discrete state-action representations for hierarchical reinforcement learning