Vowel cooccurrence as a predictor of perceived vowel similarity in Turkish

Abstract

From an exemplar models perspective, vowel harmony classes emerge from vowel cooccurrence patterns in the lexicon through similarity-based coactivation of exemplars. Although the vowel inventory of Turkish is phonologically symmetrical and Turkish vowel harmony is highly systematic and pervasive, Turkish vowel harmony as it manifests in the lexicon is not exceptionless—Turkish vowel pairs vary in their likelihood of cooccurrence in the lexicon. We hypothesize that Turkish vowel pairs with higher likelihood of cooccurrence will have higher perceived similarity than vowel pairs with lower likelihood of cooccurrence. We test this hypothesis in a Turkish vowel identification experiment with 40 native Turkish listeners. We analyze the perceptual confusion responses as a measure of perceived vowel similarity and find that, contrary to our hypothesis, higher likelihood of vowel cooccurrence yields fewer perceptual confusions indicating lower perceived vowel similarity. In a post-hoc analysis, we find that the relationship between likelihood of vowel cooccurrence and perceived vowel similarity is qualified by vowel peripherality, with non-peripheral vowels driving the unexpected relationship between vowel cooccurrence and perceived similarity. We discuss the results in relation to the role of vowel peripherality in Turkish vowel harmony (cf. Clements & Sezer, 1982; Suomi, 1983).