Working paper Open Access
Cheyenne has both vowel devoicing and tone. The resolution of the conflict between these two phenomena differs depending upon the position of the vowel in the word. The tone of a non-final devoiced vowel is lost completely (or, in some contexts, is inferable because of a particular tonal restriction on devoicing), whereas the tone of a final devoiced vowel is always recoverable. Leman  claims that the mechanism of recoverability is special sandhi tones on the penult, which result from tonal assimilation rules applying prior to devoicing.
In this paper I will present phonetic evidence that Leman's sandhi tones do not exist, and therefore that his analysis cannot correctly characterize the nature of tone recoverability in Cheyenne. An alternative autosegmental analysis will be proposed in terms of tone spreading, and an independently motivated accent will be suggested as an explanation for the difference in behavior between final and non-final tones.