Text this: Reward-optimizing learning using stochastic release plasticity