Introduction

Given an inventory of segments and a set of features, there are two ways to perform the feature analysis. Either all segments are specified for all features (maximal or full specification) or the feature specification is kept maximally simple (minimal or impoverished specification). RCVP subscribes to minimal specification, because it is my assumption that the distinction between contrastive and non-contrastive properties is foundational to phonology. Using Occam's razor, we might then see how far we can push minimal specification, possibly also omitting contrastive properties in some circumstances. Of course, any specification, whichever view is adopted, must be correlated with phonetic properties. In a full specification approach this correlation is rather straightforward: every feature specification is implemented in terms of a phonetic correlate and these correlations must be subject to co-articulation, which creates the overlap that is characteristic of speech (and sign), and other local and global influences on phonetic implementation, such as in tonal domain, downstep and declination. When we adopt minimal specification, the question arises whether, prior to phonetic implementation, feature specifications are filled in so that we arrive at a full specification, or whether phonetic implementation can bypass that step, doing all the work that is necessary to arrive at a phonetic representation.

What counts as a minimal specification depends on certain assumptions regarding exactly what can be left unspecified, with a major difference between assuming that only non-contrastive, redundant specifications count as such (contrastive underspecification) and that, when a contrast exists, one specification, referred to as a default specification, can also be left out (radical underspecification). With respect to the choice of the feature system, a major difference exists between using binary features and using unary features (or elements). See § 2.2.1 for a discussion of these various distinctions.

An important issue is also how exactly one arrives at a minimal specification, given that, as we will see, there is not always only one way to get there. This means that we need a procedure or algorithm that guides the way to a minimal specification.

This chapter investigates these various issues, and related ones, with particular attention to the possibility of using radical underspecification in a unary system, and how, within such a system, we can arrive at a minimal representation.