Co-mingling effects of dialectal and stylistic variation due to choice of speaker normalization

Abstract

This paper explores how a speaker normalization routine can result in co-mingling effects of dialectal and stylistic variation. To this end, the paper examines a database of vowel formants in a passage and a wordlist task collected from 88 monolingual American English speakers in Michigan’s Upper Peninsula; this corpus is further stratified by ethnic-heritage, age, sex and educational attainment. Two different speaker normalization routines, Lobanov (Adank et al., 2004) and Labov ANAE (Labov et al., 2006), were used; however, each routine yielded differences in a paralleled analysis. The Lobanov normalization routine removes much of the ethnic-heritage effect that appears in the analysis using the Labov ANAE normalization routine. Examining the cause of these different results reveals that 1) the Lobanov routine removes any mean shifts in formant values that run orthogonal to anatomical variation (i.e., vocal tract length effects) and 2) the ANAE routine only removes shifts which are correlated across formants. The ethnic-heritage effect in this dataset involves mean shifts that are negatively correlated to typical anatomical variation, which results in the ANAE routine potentially amplifying such effects. We discuss the import of these findings for vowel space variation and language change.