HOW CONCRETE IS PHONOLOGY

This paper presents evidence for abstract phonology-that is, a phonology that recognizes underlying segments which are not realized phonetically. Two underlying vowel segments /o/ and /e/ are posited for Nupe (a Kwa language of Central Nigeria), both of which are realized as [a], therefore merging ('absolutely neutralizing') with /a/. Arguments based on pattern congruity, naturalness, and a 'weak' form of alternation are brought forth in support of this solution. Central, however, is the demonstration of the psychological reality of these two segments. In addition, various alternate solutions are shown to be inadequate. Finally, implications of the abstract Nupe solution are discussed.l Much of the recent discussion in generative phonology has centered around the issue of abstraction: to what extent do the underlying forms of the most highly valued phonological system of a language differ from those forms encountered in the surface phonetics? Although generative phonology has never advocated a position as abstract as the one taken by Lamb 1966, for example, or by Fudge 1967, phonologists have generally assumed that a considerable degree of abstraction is correct in phonological solutions. Independent of any claims one might wish to make concerning the reality of these phonological solutions, many phonologists express the feeling that somehow these (and only these) are explanatory, since they account for many otherwise unaccountable exceptions to general rules and conditions by postulating a deeper, more regular pattern. That this notion of 'explanation' differs from the more general notion of EXPLANATORY ADEQUACY of grammars (and often, some would claim, is at odds with it) is perhaps a point that has not been properly emphasized. Does a phonology that 'explains' certain processes at an abstract level necessarily coincide with the one explanatorily adequate (that is, the most highly valued) form of phonology? In other words, are these abstractions justified on a principled basis, and do we want them in our grammar? Kiparsky, in an unpublished paper entitled 'How abstract is phonology?' (1968), presents evidence from linguistic change to show that many of the so-called abstractions put forth by phonologists have no place in generative phonology. It is to this issue of abstractness in phonology that I shall address myself in this paper. The term 'abstract' will be used specifically to refer to those solutions that permit non-alternating underlying segments which do not emerge on the phonetic surface. 1. THE PROBLEM. In writing grammars, the aim of reaching explanatory adequacy entails a quest for language universals and the 'psychologically real', 1 This paper is an extended version of the one I delivered in Urbana, Illinois, on 26 July 1969, at the Summer LSA Meeting. I would like to thank Victoria Fromkin for giving up much of her valuable time to help me revise the earlier version of this paper. I am also deeply indebted to my good friends and colleagues Isaac George and Danny Alford for many valuable insights.