212 Cynthia G. Clopper
performance was much more accurate on the Mid-Atlantic talkers than the
New England talkers overall. This ¿ nding suggests that when the listeners in
our experiment were making their categorization judgments, they may have
been relying on different mappings between acoustic characteristics of the
speech signal and regional labels than those predicted based on descriptive
variationist research (e.g., Labov et al. 2006). We have conducted two addi-
tional experiments which allowed us to investigate the role of category labels
in the perception of regional dialects of American English.
First, a free classi¿ cation experiment was developed to investigate naïve
listeners’ classi¿ cation behavior in the absence of experimenter-provided dia-
lect regions or labels (Clopper and Pisoni 2007). The listeners were asked to
group a set of unfamiliar talkers by regional dialect based on sentence-length
utterances. They were per mitted to make as many groups as they wanted with
as many talkers in each group as they wished. Two sets of talkers from six
different regional dialects of American English produced the stimulus mate-
rials. One set of talkers included only males and was identical to the talkers
who produced the stimulus materials in our original forced-choice categoriza-
tion task (Clopper and Pisoni 2004). The second set of talkers included both
males and females and was identical to the second set of mixed male and
female talkers from our earlier categorization task (Clopper and Pisoni 2006).
By using the same sets of talkers and stimulus materials, we could directly
compare the results of the free classi¿ cation experiments to the forced-choice
categorization experiments.
Given that the pattern of errors produced by the listeners in our earlier
forced-choice categorization experiments consistently revealed three broad
dialect categories, we predicted that the naïve listeners would exhibit a rela-
tively high tolerance for within-group acoustic variability and make a rela-
tively small number of groups in the free classi¿ cation task. Instead, however,
the listeners made an average of eight to ten groups of talkers, suggesting
that they were able to represent ¿ ne-grained acoustic differences between the
talkers. However, their grouping accuracy was still rather poor overall, which
may indicate attention to talker-speci¿ c differences instead of dialect-speci¿ c
variation. Thus, the dif¿ culties naïve listeners exhibit in explicit categoriza-
tion tasks may reÀ ect poorly speci¿ ed perceptual dialect categories, and not
simply a mismatch between the experimenter-provided labels and the listen-
ers’ own cognitive categories.
Multidimensional scaling analyses of the aggregate free classi¿ cation data
revealed two primary dimensions of perceptual dialect similarity: geography
(northern vs. southern varieties) and distinctiveness (many vs. few character-
istic properties). When both male and female talkers were used, gender was