Classi¿ cation of Regional Language Varieties 213also revealed to be an important dimension of perceptual similarity (Clop-
per and Pisoni 2007). This ¿ nding suggests a strong interaction between dia-
lect and gender in perception, similar to the perceptual interference effects
reported for linguistic categories and talker gender (e.g., Mullennix and
Pisoni 1990; Strand 1999). Additional research is needed to explore the role
of talker gender in perception and the apparent inability of listeners to ignore
this important source of talker variability in speech processing.
The results of the free classi¿ cation experiment were consistent across
two different sets of stimulus materials and for listeners with different regions
of origin and degrees of geographic mobility. However, as in the previous
forced-choice categorization experiments (Preston 1993; Clopper and Pisoni
2004), the region of origin of the listeners had a signi¿ cant effect on their per-
formance in the free classi¿ cation task. Listeners tended to perceive greater
similarity between their own dialect and neighboring dialects than between
more geographically distant dialects. Geographic mobility attenuated this
effect and resulted in greater discrimination overall.
In a second experiment, we used a paired comparison similarity ratings
task to examine the perceptual similarity of dialects in more detail (Clopper,
Levi, and Pisoni 2006). In this experiment, the listeners were presented with
pairs of talkers and after listening to one sentence produced by each talker,
were asked to rate the similarity of the talkers’ dialects on a seven-point simi-
larity scale. Only four dialect regions were represented in the paired compari-
son similarity ratings task. We reduced the set of talkers in this experiment to
make the number of pairwise comparisons manageable and used a subset of
the male and female talkers from our earlier forced-choice categorization and
free classi¿ cation tasks (Clopper and Pisoni 2006; Clopper and Pisoni 2007).
While the listeners consistently assigned higher ratings to talkers from
the same dialect region than to talkers from different dialect regions, the mean
ratings for both same-dialect and different-dialect pairs were near the middle
of the range, again suggesting relatively poor performance overall. The mean
ratings for the same-dialect and different-dialect pairs are shown in Figure
8.2. In addition, talker gender was found to be an important factor in this
experiment; same-gender pairs were consistently rated as more similar than
different-gender pairs, despite instructions to ignore talker gender in making
the dialect similarity judgments (Clopper et al. 2006). This ¿ nding again sug-
gests interference between two different sources of social information in an
explicit perceptual task.
Multidimensional scaling analyses of the aggregate similarity data
revealed the same two dimensions of perceptual similarity that were
obtained in the earlier free classi¿ cation experiment (Clopper and Pisoni