How Abstract Is Nupe?

In recent years the basic principles of generative phonology have been subjected to increasingly critical scrutiny. Many questions have been raised; but answers have been few and, for the most part, have lacked firm empirical support. One important unresolved issue concerns the abstract relationship between the phonological shape of morphemes in the lexicon and their phonetic manifestations. In an attempt to restrict the abstractness of phonological lexical representation, Kiparsky 1968 proposed that grammars not permit 'absolute neutralization'-i.e. the positing of 'underlying distinctions for the sole purpose of classifying segments into those that do and those that do not meet the structural analysis of a rule', distinctions which 'are never realized on the phonetic surface' and which are lost 'regardless of environment' (pp. 10-11). Hyman 1970 proposes an analysis of Nupe which, he feels, requires three underlying low vowels /e o a/, all of which unconditionally become phonetic [a], a case of absolute neutralization. Hyman's Nupe solution, if valid, would constitute strong counter-evidence to Kiparsky's constraint, and thus more firmly establish the status of abstract phonological representation. I shall show, first, that a Nupe solution is possible which employs only the most general form of those rules assumed by Hyman, but which avoids the need for a rule of absolute neutralization. Second, I will show Hyman's rules to be inadequate to account for the data of Nupe (as reported in Hyman and in Smith 1967). Third, a plausible analysis of Nupe will be given which not only fails to support the case for abstract phonology, but is also consistent with one of the most concrete of phonological approaches, that of Sapir-one in which the lexical representation of a morpheme contains all those surface phonological features which remain constant in all its variants (cf. McCawley 1967:110). 1. Hyman presents the following rules (numbered as in his article):