Is Voicing a Privative Feature?

A typology of voicing assimilation has been presented in Cho (1990a), whose result will be summarized in section 2. Like many other marked assimilations, voicing assimilation is characterized as spreading of only one value of the feature [voice]. The main body of this paper will compare a privative theory of voicing with a binary theory. It has often been noted that assimilation rules are natural rules since they are cross - linguistically very common. It has also been observed that they are asymmetric in nature (Schachter 1969, Schane 1972). For example, nasalization, palatalization, and assimilation of coronals to noncoronals are extremely common but the reverse processes are not frequently found in natural languages. On the other hand, voicing assimilation has been known to be relatively free in choosing its propagating value. Whereas the other assimilation rules are sensitive to the marked and the unmarked value of a given feature, assimilating a voiced consonant to a voiceless consonant has been assumed to be as natural as the reverse process (Anderson 1979, Mohanan (forthcoming)). I have argued that voicing assimilation is no different in its asymmetry from the other types of assimilation by demonstrating the need for two parameters and one universal delinking rule. A universal typology emerges from the possible interaction among the values associated with delinking and spreading parameters. The following theoretical assumptions will be utilized throughout this paper. First, I follow the standard assumption in Autosegmental Phonology that assimilation rules involve not a change or a copy but a reassociation of the features. This operation of reassociation called spreading is assumed to be the sole mechanism of assimilation rules (Goldsmith 1979, Steriade 1982, Hayes 1986). Second, I assume Underspecification Theory, which requires that some feature values be unspecified in the underlying representation ( Kiparsky. 1982, Archangeli and Pulleyblank (forthcoming)). Distinguishing different versions of Underspecification Theory will not be relevant in the discussion since I will discuss whether voicing is universally a privative feature or a binary opposition. Third, I assume the principle of Structure Preservation (Kiparsky 1985, Borowsky 1986), which is expressed in terms of constraints that apply in underlying representations and to each stage in the derivation up to the level at. which they are turned off (usually in the lexicon). Structure Preservation will be invoked to classify obstruents on the one hand, and sonorants and the other redundantly voiced segments on the other. Last, I translate the Classical Praguean conception of the relation between neutralization and assimilation into the autosegmental framework, and assume that assimilation is always feature-filling. All instances of the effect of feature-changing assimilation rules, then, are the result of two independent rules of (1) delinking and (2) spreading (Poser 1982, Mascaró 1987)