Underspecification

Abstract

Within generative phonology, there has always been a research strategy dictating that predictable aspects of a phonological string should be accounted for by rules, conventions or constraints. With respect to underlying lexical entries, a particularly strong hypothesis is to assume that all predictable information is excluded from underlying forms, that is, lexical entries are “minimally redundant” (Kiparsky 1982a). In this chapter, I will explore this hypothesis with respect to tone. It will be shown that evidence from tonal systems provides strong support for a theory where unmarked feature specifications are supplied by universal default rules. Moreover, it will be shown that the default fill-in rules apply in some cases after the application of regular phonological rules. The role and nature of tonal default rules, and how they are incorporated into the overall model of phonology, will be examined here. Special attention will be paid to problems that could result from underspecification, such as the surreptitious replacement of a binary feature system with a ternary one.