Perceptual assessment of phonetic convergence between speakers: A Bayesian meta-analysis

Abstract

The present study aims to consolidate over two decades of research on phonetic convergence using Bayesian meta-analytic techniques. This project identified 68 individual experiments across 47 published and unpublished studies that used holistic AXB perceptual similarity assessment of phonetic convergence (through 2023). Of these, 21 researchers provided full datasets for 48 individual experiments. Data from the 33 experiments with the most consistent methodologies were subjected to a series of meta-analyses in order to provide an estimate of the overall 95% credible interval for mean AXB estimates, and to resolve inconsistencies across the literature with respect to the effects of talker sex and lexical factors. Importantly, the present study calculated new estimates of word frequency and density for all relevant studies using consistent modern databases (e.g., SUBTLEX-US, etc.). The main finding was that the AXB perceptual similarity mean probability was 0.55 (95% credible interval: [0.53, 0.57]) and that estimates of the effects of talker sex, word frequency, word phonological neighborhood density and word length all reflected very small effect sizes with very weak evidence against the null. These findings reflect a subtle phenomenon that most likely does not differ according to talker sex or lexical factors, and future investigations should use representative talker sets with well-balanced lexical items to enhance generalizability. The combined datasets and analysis scripts from this project are available for future investigations at osf.io/8s9r5.