Asymmetry in incomplete neutralization

Abstract

Neutralizing patterns, such as final devoicing, are known to often be incomplete: “devoiced” final obstruents in languages such as Afrikaans may retain, e.g., longer preceding vowels than their voiceless counterparts. A common explanation for this phenomenon is that these partially devoiced obstruents are influenced by paradigmatically related obstruents that are fully voiced (e.g., Afrikaans hoe[t] ~ hoe[d]e, “hat(s)”). This proposal raises the question of whether such influence goes in the other direction too: is the [d] of hoe[d]e slightly devoiced, compared to a non-alternating [d] as in roe[d]e (“rod”)? The experiment presented here tests this hypothesis in 28 Afrikaans nouns as produced by nine native speakers; vowel length, closure duration, release duration, and glottal pulses were measured as cues for voicing. The results provide no evidence for partial devoicing of voiced stops under the influence of devoiced counterparts elsewhere in the paradigm; that is, the [d] of hoe[d]e is no less voiced than the [d] of roe[d]e. I conclude that incomplete neutralization is an asymmetrical phenomenon: segments subject to neutralization, as in hoe[t], may retain some contrasting cues; but these segments do not in turn encourage partial neutralization in morphologically related forms such as hoe[d]e.