ATR Harmony: new typological patterns and diagnostics

Phonological studies of ATR harmony have led to advances in autosegmental phonology, markedness theory, and locality. Nevertheless, many of our theoretical assumptions may have been based on misanalyzed data or non-rigorous typological assumptions. For example, detailed acoustic and/or articulatory studies have shown that transparent vowels are not always transparent (Gick et al 2006, Richart & Rose 2017). In addition, Casali (2002, 2003, 2016) has argued that markedness based on combinatorial properties of [ATR] with other features is not a good fit for many ATR harmonic systems. In this talk, I build on Casali’s research in two ways. First, I present a large-scale typological study of 450 Nilo-Saharan and Niger-Congo vowel inventories from the Sudanic belt, both with and without ATR harmony. The survey reveals some striking typological properties: i) contrasts in high vowels (i ~ ɪ and u ~ ʊ) correlate with a strong likelihood of ATR harmony regardless of whether there are mid vowel contrasts or not; ii) the vowel system /i e ɛ a ɔ o u/ (termed 1IU in Casali 2016) is often touted as being a very common ATR harmony system, but in fact turns out not to correlate well with ATR harmony. Instead, this system is more likely to show no harmony (Nilo-Saharan and northeastern Niger Congo) or trace effects of harmony (western Niger-Congo). Second, I offer an interpretation of the results of the survey in terms of perceptual distance, driven by inventory contrast. The dominant role of high vowels in ATR systems and the weaker role of non-high vowels follows from perceptibility; high vowel ATR contrasts are perceptually more difficult than mid contrasts (Fulop et al 2008). High contrasts activate harmony, and the [+ATR] feature of the most peripheral vowels /i u/ of the pair. Furthermore, [+ATR] derives [e o] from /ɛ ɔ/ in languages that lack a mid vowel contrast. In the absence of high vowel contrasts, mid vowel contrasts do not present enough of a perceptual difficulty to reliably activate harmony. If harmony does occur, it tends to be driven by /ɛ ɔ/ with [-ATR] as the active feature (Casali 2008, 2016). High [-ATR] vowels can be created in these harmony systems (as in Ijesa or Ekiti Yoruba), but high [+ATR] vowels may also be neutral.