A Bayesian Model Predicts Human Parse Preference and Reading Times in Sentence Processing