A neural mapping hypothesis to explain why velar stops have an allophonic split.

Velar stops are phonetically characterized by two allophones-a palatal "g" when produced in front-vowel contexts and a velar "g" when produced in back-vowel contexts. Acoustic characterizations of velar stops based on plotting the onset and offset frequencies of the second formant transition clearly reveal two acoustically separated allophonic groups. Using cross-linguistic acoustic data, it is shown that alveolar stops fall precisely within the acoustic void created by the velar allophones. A neural-based mapping hypothesis is put forth claiming that the bimodal distribution of velars is an instance of articulatory behavior acting in the service of auditory representation. Copyright 1999 Academic Press.