Uniformity and the Phonetics - Phonology Boundary

This is a study of paradigmatic relations and of their significance for the link between phonology and phonetics *. A paradigm is a set of words sharing a morpheme of phrases sharing a word (e.g.{bomb, the bomb, ...}). The main component of the analyses presented here is the preference for uniform paradigms, that is paradigms sharing contextually invariant morphemes. A Paradigm Uniformity (PU) condition is a statement of the type shown in (1), which promotes invariance of some sound property within a given paradigm: (1) All surface realizations of µ, where µ is the morpheme shared by the members of paradigm x, must have identical values for property P. Examples of uniform and non-uniform paradigms appear in (2). In both cases, the shared morpheme is a root: it alternates in (2.a), but is invariant in (2.b). (2) a. {bAm, bAmb-Ard, bAmb-Ard-i " }: not uniform wrt stem final C quality b. {bAm, bAm-IN, bAm-" , bAm-z}: uniform wrt stem final C quality I discuss here the phenomenon of paradigm levelling, which represents the systematic generalization of one allomorph to positions where it is phonologically unjustified or unexpected, as a means of satisfying a PU condition. The degree of phonological invariance of shared morphemes stands in direct relation to the paradigm's productivity and to the transparency of the derivative's relation to its base (Bybee 1988): the paradigms like (2.b) that are generated by productive word formation processes involve less contextual variability than the unproductive paradigm (2.a). Bearing this in mind we concentrate here on highly productive and compositional paradigms. Paradigm levelling is a staple of the phonological literature. Different aspects of it have been studied under