On Admissible Linguistic Evidence