Analyzing dialect variation in historical speech corpora

The Linguistic Atlas of the Gulf States is an extensive audio corpus of sociolinguistic interviews with 1121 speakers from eight southeastern U.S. states. Complete interviews have never been fully transcribed, leaving a wealth of phonetic information unexplored. This paper details methods for large-scale acoustic analysis of this historical speech corpus, providing a fuller picture of Southern speech than offered by previous impressionistic analyses. Interviews from 10 speakers (∼36 h) in southeast Georgia were transcribed and analyzed for dialectal features associated with the Southern Vowel Shift and African American Vowel Shift, also considering the effects of age, gender, and race. Multiple tokens of common words were annotated (N = 6085), and formant values of their stressed vowels were extracted. The effects of shifting on relative vowel placement were evaluated via Pillai scores, and vowel dynamics were estimated via functional data analysis and modeled with linear mixed-effects regression. Results indicate that European American speakers show features of the Southern Vowel Shift, though certain speakers shift in more ways than others, and African American speakers' productions are consistent with the African American Vowel Shift. Wide variation is apparent, even within this small geographic region, contributing evidence of the complexity of Southern speech