Prolegomena to a theory of word formation