Two Types of Retroflex Harmony in Kalasha: Implications for Phonological Typology

Abstract

Recent cross-linguistic surveys by Hansson (2001; 2010) and Rose & Walker (2004) have revealed that consonant harmony systems exhibit unique typological properties that set them apart from other assimilation patterns including vowel and vowel-consonant harmony systems. This has been taken as evidence for the hypothesis that there are two distinct mechanisms of assimilation at work in languages, feature/gesture repetition (or “agreement”) and extension (or “spreading”), each with its own functional underpinnings and associated typological properties. The properties that are most characteristic of consonant harmony systems are those that are said to arise from feature repetition. They include: (i) an inherent bias toward regressive directionality; (ii) transparency of intervening segments; and (iii) sensitivity to the similarity of interacting segments. This paper presents a case study of retroflex assimilation in Kalasha, a Dardic language of northern Pakistan, and argues that the evidence from Kalasha provides support for the typological distinction between feature repetition and gesture extension. Kalasha has a (typologically rare) phonological inventory in which retroflex stops, affricates, fricatives and vowels contrast with their non-retroflex counterparts. Moreover, Kalasha exhibits two distinct patterns of long-distance retroflex assimilation: retroflex consonant harmony and retroflex vowel (or vowel-consonant) harmony. Using statistics calculated over the Kalasha lexicon (based on Trail & Cooper 1999) and historical-comparative data, the study documents a pattern of retroflex consonant harmony that is regressive, skips intervening vowels (i.e., “transparency”) and is highly sensitive to the similarity of interacting consonants. It applies to pairs of coronal obstruents that agree in manner of articulation (i.e., two stops, two affricates or two fricatives) but not to pairs with different manners of articulation (i.e., stop-affricate, stop-fricative and some affricate-fricative pairs). This is contrasted with the pattern of retroflex vowel harmony described by Heegård & Mørch (2004), which is primarily progressive (but potentially bi-directional) and does not show clear evidence of similarity or transparency effects. It applies to any pair of vowels and can even target intervening coronal consonants (e.g., /a˞in/ → [a˞i˞n] ~ [a˞i˞ɳ] ‘millet’). The Kalasha case study makes two important contributions. First, Kalasha provides unambiguous evidence for the role of similarity in retroflex consonant harmony systems. This is significant because most previously documented cases of retroflex consonant harmony are ambiguous in this respect. Second, Kalasha is the only language known to exhibit both retroflex consonant and retroflex vowel harmony. The coexistence of two patterns of retroflex harmony in the same language, each with a distinct set of typological properties, lends support to the hypothesis that two independent mechanisms of assimilation are at work: feature/gesture repetition and extension.