More on post-nasal devoicing: The case of Shekgalagari

Post-nasal voicing is a typologically and historically common process. This process, by which voiceless obstruents become voiced after nasals, has a well-known aerodynamic and perceptual basis: (i) prolonged voicing into the stop closure, vis-à-vis postvocalic stops, due to nasal leakage before full velic closure is achieved and oral cavity expansion due to the velum continuing to raise after velic closure has occurred (Rothenberg 1968; Westbury 1983; Hayes and Stivers 2000), and (ii) the reinterpretation of these phonological voiceless stops, partially voiced and with a weaker stop burst, as voiced. This tendency to avoid voiceless stops after nasals is a supposedly universal tendency for which a *NT constraint has been proposed (Pater 1996, 1999; Hayes 1999). A number of Bantu languages, however, show a process of post-nasal obstruent DEVOICING which has been argued to provide evidence for a corresponding bias against nasal + voiced obstruent and the competing constraint *ND (Hyman 2001), which is responsible for such alternations as [bç@n-a] ‘see!’ vs. [m.pç@n-a] ‘see me!’ in Tswana. Although Hyman goes into great detail justifying *ND as an active synchronic constraint in Tswana and other languages, the proposal raises the following questions for those who assume that phonology must be “phonetically based” (Hayes et al. 2004): (i) Can phonologies exploit an allegedly unnatural phonetic constraint such as *ND? (ii) If so, what does this say about the alleged universal preference for ND over NT assumed by Hayes, Pater and others before them (e.g. Herbert 1986)? In optimality theory terms, what would it mean to assume a universal ranking of the two “markedness” constraints, *NT >> *ND, if languages such as Tswana can reverse the ranking as Hyman proposes? Since Hyman (2001) two groups of researchers have taken a new look at Tswana both from a phonetic and phonological point of view. While Coetzee et al. (2007; in press) generally confirm the postnasal devoicing process that has been noted for quite some time in languages of the Sotho-Tswana subgroup of Bantu, Zsiga et al. (2006, 2007) take issue with Hyman’s analysis on two fronts. First, finding that their speakers have variable realization of voiced stops (including devoicing) in other environments as well, that is, not only postnasally, they cast doubt on the phonetic process (vs. Coetzee et al., whose speakers produced the expected devoicing exclusively postnasally). Second, they reject *ND as a phonological constraint and propose to attribute any tendency towards postnasal devoicing to other constraints. On the one hand they view devoicing, in particular postnasal devoicing, as a “strengthening” process, as in virtually all previous literature on the history and description of the group (e.g. Tucker 1929, Dickens 1977,