Do Words Reflect Beliefs? Evaluating Belief Depth in Large Language Models