Calibration, Entropy Rates, and Memory in Language Models