Best practices in measuring vowel merger

Vowel mergers are some of the most well-studied sound change phenomena. Yet the methods for assessing and characterizing an individual speaker's participation in an ongoing merger (or split) vary widely, especially among researchers analyzing naturalistic corpora. We consider four methodological approaches to representing and assessing vowel difference: Euclidean distances, mixed effects regression modeling (Nycz 2013), the Pillai-Bartlett trace (Hay, Warren, & Drager 2006), and the spectral overlap assessment metric (Wassink 2006). We discuss the strengths and weaknesses of each method and compare them by applying all of them to three different data sets, each of which contains low vowel data from speakers whose status with respect to a
vowel contrast may not be clear-cut: realizations of COT and CAUGHT in San Francisco, California; COT and CAUGHT among Canadians in the New York City region; and TRAP and BATH among Scots who work in Southern England. We conclude with some practical recommendations.