364 Renée van Bezooijen and Vincent J. van Heuven
3.3 Vowel normalization
Formant values measured for perceptually identical vowels may differ
between individuals. The larger the differences are between two speakers in
shape and size of the cavities in their vocal tracts, the larger the differences in
formant values of perceptually identical vowel tokens will be. Given that the
vocal tracts of women are some 15% smaller than those of men, comparison
of formant values is hazardous across speakers of the opposed sex. Numer-
ous attempts have been made, therefore, to factor out the speaker-individual
component from the raw formant values such that phonetically identical vow-
els spoken by different individuals would come out with the same values.
None of these vowel normalization procedures have proven fully satisfactory
(Labov 2001: 157–164).
Broadly, two approaches to the normalization problem have been taken in
the literature (see also Nearey 1989). The ¿ rst approach, called intrinsic nor-
malization, tries to solve the problem by considering only information that is
contained in the single vowel token under consider ation, typically by comput-
ing ratios between pairs of formant values such as F1/F0, F2/F1.^2 The alterna-
tive, extrinsic normalization, looks at tokens of all the vowels in the phoneme
inventory of a speaker and expresses the position of one vowel token relative
to the other tokens within the individual speaker’s vowel space.
For the purpose of the present study we adopted a hybrid solution, which
combines virtues of both intrinsic and extrinsic normalization. The intrinsic
part of our normalization is just a transformation of the measured formant val-
ues from Hertz into Bark. The extrinsic part of the procedure is a new imple-
mentation of what has been called “end-point normalization” in the literature.
A vowel token is scaled according to its relative position between the extreme
(lowest and highest) values for F1 and F2 found for the individual speaker.
Since the study is limited to the sound change in /͑i/—a front, unrounded
vowel—we only require reference vowels that allow us to determine the indi-
vidual implementation of the front region of the speaker’s vowel space. All
that is required, therefore, is a reliable estimation of the speaker’s /i/ (maxi-
mally high front vowel) and /a/ (maximally open front vowel). We made the
explicit assumption that the point vowels /i/ and /a:/ do not participate in the
sound change in progress that affects the Dutch mid vowels (as is also implied
by Stroop’s vowel diagram, which does not indicate any involvement of the
point vowels /i, a:, u/ (1998: 28; see also our Figure 15.1).
In some cases, the vowel tokens of a speaker were dispersed in a perfectly
regular fashion, with the ¿ ve tokens of /i/ compactly clustered in the left-hand cor-
ner of the acoustic vowel space, the ¿ ve tokens of /a/ in the open-central area, and