Underlying representations and Bantu segmental phonology

A number of phonologists have claimed that we don’t need underlying representations (URs) in phonology. The arguments against URs come from a number of sources which variously appear to accuse URs of being either wrong (speakers don’t “know” them), redundant (we have other mechanisms to deal with the facts that have motivated URs), or insufficient (much more is stored in lexical entries than “just” URs). The basic question is whether lexical entries have URs that are not identical to their surface pronunciations, e.g. in isolated words. While this is easier to show when there is an alternation that occurs across word boundaries, the question is whether word-internal alternations should be captures in terms of input/output relations or by listed allomorphs. A major argument for the latter is that word-internal phonology seems to be unstable, often subject to morphological restrictions and exceptions. In this paper I consider such instability with respect to one phonetically “natural” and one phonetically “unnatural” segmental alternation occurring widely in Bantu languages to ask why this should be so.