Tekstiviesti: Adaptive representations for reinforcement learning