Restrictions on consonant harmony are usually accounted for in terms of transparency of intervening vowels to the consonant feature being spread: If the vowel is specified for the feature, spreading is blocked; if it is not specified, spreading is possible. This approach has been employed in attempts to explain the typological generalization that almost all cases of consonant harmony involve distinctions among coronals, as in Navaho, Chumash and Sanskrit. E.g. Shaw 1991 argues that consonant harmony in labial and dorsal features is impossible because vowels are generally specified for these features.

This analysis predicts a number of unattested harmony patterns: (1) Since vowels are not specified for [coronal], this articulator should be able to spread across vowels onto a following consonant yielding unattested assimilations such as /tap/ -> [tat]. (2) Given the standard assumption that unrounded vowels are not specified with a labial articulator, [labial] should be able to spread from consonant to consonant across unrounded vowels resulting in unattested assimilations such as /pak/ -> [pap].

A more satisfactory account of restrictions on consonant harmony can be derived from the hypothesis that assimilation is strictly local, so intervening vowels are not transparent to consonant harmony, they undergo it. The possibility of consonant harmony in a given feature then depends on compatibility of the feature with intervening vowels, not on transparency. The only consonantal gestures which can be extended through a vowel without significantly affecting its quality are the differences in tongue-tip posture (apical vs. laminal) that can distinguish coronals. Other features such as nasality and pharyngealization are compatible with a vocalic degree of constriction, but have clearly perceptible effects on vowels. So nasal harmony (as in Tucanoan (Piggott 1992)) and pharyngeal harmony (as in Arabic) have been characterised as applying to consonants and vowels. Extending a labial, dorsal or coronal constriction gesture would mask intervening vowels, so consonant harmonies of these types are not possible (cf. Ni Chiosain and Padgett 1994).

Strict locality in assimilation implies that there is no transparency to assimilation, but there are a number of prima facie cases of consonant harmony for which this claim seems implausible, e.g. Ponapean labialization harmony and identity between glottalized consonants in K'ekchi and Hausa. It will be argued that these are not in fact cases of assimilation but result from a dispreference for similar but non-identical consonants within the morpheme or word. Finally I will outline an approach to the analysis of transparency in vowel harmony which does not employ non-local feature spreading.