Perceptions of /a/ fronting Across Two Michigan Dialects 225vowel shifts, including /D/ fronting, /æ/ fronting and raising, and /͑/ lower-
ing. Shifts of this kind represent a substantial source of acoustical variation
in the production of vowels and a potential source of confusion for vowel
perceivers. A possibility entertained in the present study is that perceivers
who have knowledge of NCCS will minimize their confusion by taking its
inÀ uences into account when they interpret vowel cues. In other words, their
larger perception of a talker’s dialect will set a frame for the perceptual
identi¿ cation of vowels that participate in NCCS.
To test this possibility, we compared listeners’ perceptions of vowel
tokens that might be subject to /D/ fronting, depending on whether those
tokens occurred (a) at the end of sentences produced by a talker from the
Detroit area whose speech showed clear evidence of NCCS (including /D/
fronting); or (b) at the end of sentences produced by a talker from the
Ishpeming area whose speech showed no evidence of NCCS. The details
of that test and its results are presented in Section 4. Section 2 ¿ rst sum-
marizes several ¿ ndings regarding the phenomenon of talker normaliza-
tion, which is a vowel perception effect closely related to the sociophonetic
effect under study here. Section 3 then describes a ¿ eld-based method of
data collection that we employed to gather the perceptual data that are
reported here.
- Talker normalization in vowel perception
The speech signals that a listener must process are extremely variable. The
sources of this variability are of several different kinds. One of the most
notable is between-talker variation in vocal tract size, especially overall
vocal tract length (VTL). VTL varies across both age (children vs. adults)
and gender (women vs. men), and it directly affects the formant frequencies
of all vowels produced by a talker (Fant 1970). Effects of VTL variation can
be substantial. As an example, the formant frequencies of the vowel /æ/,
as in “back,” produced by male and female talkers from the same speech
community can differ by as much as 300 Hz along F1 and F2. Differences
of this magnitude are large enough to produce overlap between the formant
frequencies of neighboring vowel categories (Peterson and Barney 1952).
Despite this, perceivers are generally accurate in their categorization of
vowels. This strongly suggests that an adjustment for VTL differences is
made routinely, as part of the vowel perception process. This adjustment is
commonly referred to as talker normalization (Strange 1999).