Keywords

post-measurement processing choices, order effects in sociophonetic workflows, vowel data filtering procedures

Abstract

In modern sociophonetic data analysis, a researcher makes numerous decisions when processing their data. Many of the “big” decisions include how data is collected, how audio is transcribed, how these transcriptions are aligned to the audio, and how acoustic measurements are extracted. Once a spreadsheet of numbers has been generated, further decisions must be made regarding normalization, detecting and handling outliers, and excluding tokens irrelevant to the research question. This paper takes a closer look at this second group of decisions—those made after acoustic measurements have been extracted—and shows that the order in which they are applied can have significant effects on the final conclusions drawn from the data. Recent sociophonetic research on North American vowels typically involves at least half a dozen or so procedures that are applied to a spreadsheet of formant measurements before a particular metric can be generated. Such steps include detecting (and removing) outliers, removing stopwords (i.e. function words and other highly frequent words), removing unstressed vowels, normalizing the data, isolating midpoints from vowel trajectories, removing diphthongs and other vowel classes irrelevant to the study, and excluding allophones such as vowels in preliquid environments. The researcher’s task is to decide how and when each of these procedures should be carried out.

Original Publication Citation

Joseph A. Stanley. “Interpreting the order of operations in sociophonetic analysis.” Linguistics Vanguard. 8(1). 279–289. DOI: 10.1515/lingvan-2022-0065.