Variable vowel convergence in a novel cooperative task

Abstract

Phonetic convergence is partly automatic, yet mediated by linguistic and attitudinal factors; salient social identity can suppress convergence or lead to divergence (Babel 2010). We assessed convergence across vowels among 12 pairs of speakers using a novel task that minimizes the salience of identity. Each person separately read aloud a 45-item word list, once before the main task and again afterwards. Between readings, the pair played a version of the game Taboo, in which players take turns attempting to elicit specific words from their partner while avoiding forbidden words. Lobanov-normalized formant values were extracted from the word lists. A convergence measure was calculated for each word for each pair, by subtracting the Euclidean distance between the speakers’ vowels in that word in post-game lists from that of the pre-game lists. Mixed-effects models of convergence were fit, with random effects for word and pair and fixed effects for vowel and pre-game distance. Greater initial distance was associated with greater convergence. /i/ and /a/ converged most consistently, while diphthongs typically diverged. This suggests that listeners focus on point vowels to model an interlocutor’s vowel space, facilitating convergence in these vowels, and that even when identity is not salient, divergence may occur.