Phonetic Detail in the Perception of Ethnic Varieties of US English 321increased the statistical discrimination of tokens in conjunction with other
measures while not being statistically signi¿ cant on their own. In the present
study there do appear to be such measures that are not statistically signi¿ cant
but which reÀ ect underlying source and ¿ lter factors.
- Conclusion
In sum, from these two exploratory experiments involving ethnically af¿ li-
ated dialect data, we can observe variation in the mapping of acoustic charac-
teristics to perceptual cues. We can respond af¿ rmatively to the overarching
question of whether the mapping of acoustic characteristics to perceptual cues
allows for variation. For the two data sets previously mentioned we identi¿ ed
three questions regarding the interface between the acoustic characteristics
and perceptual cues. The ¿ rst question was: what acoustic characteristics
do or do not serve as perceptual cues? Using a forward stepwise discrimi-
nant analysis, the reduced set of predictors in experiment 1 focused largely
on vowel duration and percent glottal pulsing, while the set of predictors in
experiment 2 focused largely on the vowel space of the two vowels in “hello.”
The second question asked about the relevant strength of cues. For the Wis-
consin English data in experiment 1, while vowel duration and percent glottal
pulsing were identi¿ ed as acoustic variables with strong predictor qualities,
the contribution they make to the predictor model varied over subject groups.
Also, a change in F1 (claimed by Kingston and Diehl to be a strong factor)
was only a mild consideration in the data. For the “hello” data in experiment
2, vowel space was generally stronger than intensity, which in turn was stron-
ger than glottal airÀ ow and pitch. The third question was whether or not there
is variation in the acoustic-perception relation, i.e., either in the selection of
cues or the weighting of cues. In both analyses the set of perceptually strong
measures were not identical to the set of acoustically strong measures. More-
over, both data sets reveal some quirky behavior. For the Wisconsin English
data, the perceptual data highlighted the selection of more standard measures,
but weighted them like the older Wisconsin English weights. For the “hello”
data there was a preference for ¿ lter (vowel space) information over source
information, although both analyses included at least one intensity measure.
A number of ¿ ndings contribute directly to the importance of examining
the acoustic-perception mapping in light of perceptual dialectology beyond
just allowing for variation. The re-analysis of “hello” tokens by intended dia-
lect showed a marked difference to the analysis by perceived dialect. The
difference in identi¿ ed factors calls for more in-depth understanding of how