On the validity of remember–know judgments: Evidence from think aloud protocols