Probing for referential information in language models