Evaluating and Inducing Personality in Pre-trained Language Models