210 Cynthia G. Clopper
effect on their responses. The Michigan listeners perceived the north-south
boundary at a more northern location along the geographic continuum than
the Indiana listeners. Thus, region of origin affects the perception of social
categories, such as regional dialect, as well as the perception of local vowel
systems (e.g., Labov and Ash 1998).
More recently, Clopper and Pisoni (2004, 2006; Clopper, Conrey, and
Pisoni 2005) have used forced-choice perceptual categorization tasks to
explore the role of listener background in dialect perception. In the ¿ rst of a
series of studies, we asked undergraduate listeners to categorize a set of male
talkers by regional dialect based on sentence-length utterances (Clopper and
Pisoni 2004). The listeners were shown a map of the United States with six
dialect regions drawn and labeled on it and were asked to listen to each short
sample of speech and select the region that they thought the talker was from.
Overall performance was quite poor in this six-alternative forced-choice cat-
egorization task; average accuracy was only 30%. However, performance was
statistically above chance, which is 17% in a six-alternative task, con¿ rm-
ing that while the task was dif¿ cult for the listeners, their responses were
not entirely random. Unlike the perception of talker gender, which is robust
to degradation (Lass et al. 1976), dialect categorization is quite dif¿ cult for
naïve listeners, even under ideal listening conditions.
The large number of errors produced by the listeners allowed us to quan-
titatively investigate their patterns of confusions. A clustering analysis of the
listeners’ responses revealed a perceptual similarity structure that broadly
corresponded to the phonological variation present in the speech signals.
The listeners tended to confuse northern varieties with one another, southern
varieties with one another, and western varieties with one another, but made
fewer errors between these three broader dialect categories. We replicated
these results with several different sets of sentence materials, as well as with
a group of female talkers (Clopper et al. 2005), and with two groups of mixed
male and female talkers from different corpora (Clopper et al. 2005; Clopper
and Pisoni 2006). Figure 8.1 shows the overall accuracy on the six-alternative
dialect categorization task in the four different experiments.
We also examined the effects of the listener’s region of origin and geo-
graphic mobility on forced-choice categorization performance. Like Preston
(1993), we found that region of origin was an important factor in the percep-
tion of the regional dialect of unfamiliar talkers. The listeners in our ¿ rst
experiment (Clopper and Pisoni 2004) correctly categorized more talkers
from regions that they had lived in than from regions that they had not lived
in. In addition, those listeners who had lived in multiple different dialect
regions prior to attending college performed better overall than the listeners