Using recurrent phrases as text-type discriminators: A quantitative method and some findings