218 Cynthia G. Clopper
to capture this important role of the lexicon in the construction of represen-
tations of phonological variation.
Dialect representations are also connected to talker-speci¿ c represen-
tations. In order to assign speci¿ c phonological phenomena to a given dia-
lect category, the listener must know something about the background of
the talker. That is, an individual talker may exhibit the pen~pin merger, but
until the listener learns that the talker is from the southern United States,
a link between the southern dialect category and the pen~pin merger can-
not be established. Once connections have been made between dialect and
phonological representations, however, these connections can then be used in
explicit dialect categorization tasks. Thus, a listener’s experience with dialect
variation is crucial for the formation of robust dialect representations. When a
listener encounters an unfamiliar talker with the pen~pin merger, the listener
can access his or her dialect representations in order to make a guess as to the
talker’s regional background. However, the results of the dialect classi¿ cation
experiments (Clopper and Pisoni 2004, 2006) suggest that this particular part
of the system is not well-speci¿ ed for most listeners. This is likely the result
of failures in the acquisition process. Given that we do not ever learn the
regional background of many of the talkers we encounter, it is likely that the
connections between phonological and dialect representations are somewhat
sparse, making explicit dialect categorization of unfamiliar talkers dif¿ cult.
Individual talker representations are also connected to phonetic repre-
sentations. Most listeners can accurately identify friends and family over the
telephone based on very short utterances and listeners can also be trained to
identify unfamiliar talkers in the laboratory (Nygaard, Sommers, and Pisoni
1994). The identi¿ cation of an individual talker is based on idiosyncratic
aspects of the talker’s voice which are reÀ ected in the phonetic representa-
tions of that talker’s speech, including voice quality, pitch, and talker-speci¿ c
instantiations of phonological categories (Pittam 1994). Thus, individual
talker representations must be connected to the ¿ ne phonetic details of the
talker’s speech that are available at the level of phonetic representation.
Finally, non-linguistic social representations account for listeners’ per-
formance in attitude judgment tasks. Naïve listeners make highly consistent
judgments about the intelligence or kindness of unfamiliar talkers, and their
judgments correlate with dialect and/or language prestige (Ryan and Giles
1982). The interpretation of these results typically assumes that when listen-
ers are asked to assess the intelligence of an unfamiliar talker based on a short
sample of speech, they ¿ rst activate a dialect representation and then respond
based on the connections between that representation and stereotypes about
that dialect that are stored as non-linguistic social information. Since these