Advancements of phonetics in the 21st century: Quantitative data analysis

Phonetic research in the 21st century has relied heavily on quantitative analysis. This article reviews the evolution of common practices and the emergence of newer techniques. Using a detailed literature survey, we show that most work follows a mainstream, which has shifted from ANOVAs to mixed-effects regression models over time. Alongside this mainstream, we highlight the increasing use of a diverse methodological toolbox, especially Bayesian methods and dynamic methods, for which we provide comprehensive reviews. Bayesian methods, as well as frequentist methods beyond linear and logistic regression, offer flexibility in model specification, interpretation, and incorporation of prior knowledge. Dynamic methods, such as GAMs and functional data analysis, capture non-linear patterns in acoustic and articulatory data. Machine learning techniques, such as random forests, expand the questions and types of data phoneticians can analyze. We also discuss the growing importance of open science practices promoting replicability and transparency. We argue that the future lies in a diverse methodological toolbox, with techniques chosen based on research questions and data structure.