Similar Items: Strong evidence for maintenance of gradient representations during language processing