204 Cynthia G. Clopper
¿ ndings suggest that speech perception involves processing multiple sources
of phonetic detail, including linguistic and talker-speci¿ c information.
The invention of the pattern playback synthesis system in the 1950s allowed
speech scientists to examine the relationship between ¿ ne acoustic details and
the perception of phonological categories using synthetic stimulus materials. In
what has become a classic set of experiments, Liberman et al. (1957) presented
listeners with synthetic CV syllables and asked them to identify the initial con-
sonant. The syllables formed a continuum from /be/ to /de/ to /ge/ and were
created by modifying the transition of the second formant between the con-
sonant and the vowel along a continuum from sharply rising (/be/) to sharply
falling (/ge/). Despite the relatively small acoustic difference between any two
neighboring syllables on the continuum, the listeners identi¿ ed each individual
stimulus item consistently as /b/, /d/, or /g/. That is, the listeners perceived a
continuum of synthetic speech samples as though they were categorical, with
a sharp shift from /b/ to /d/ responses in the early part of the continuum and a
shift from /d/ to /g/ responses in the later part of the continuum.
Liberman et al. (1957) then conducted a paired comparison discrimina-
tion task using an ABX paradigm. On each trial, the listeners were presented
with three stimulus items and were asked to indicate whether the last token
(X) was the same as the ¿ rst (A) or second (B) item. In all cases, the A and B
stimulus items were one-step neighbors on the continuum and the X stimulus
was acoustically identical to either A or B. Liberman et al. (1957) found that
the listeners responded at chance if they had previously identi¿ ed the A and B
stimulus items as the same consonant and well above chance if they had pre-
viously identi¿ ed the A and B stimuli as different consonants. That is, within-
category consonant discrimination was dif¿ cult, whereas between-category
discrimination was not. Taken together, the results of these phoneme identi-
¿ cation and discrimination experiments suggest that phoneme perception is
categorical, at least at some levels of processing, and that even small changes
in the acoustic signal can result in categorical shifts in perception.
However, when Pisoni (1973) replicated these two experiments with a
continuum of vowels from /i/ to /Ԍ/, he found a sharp boundary in the identi¿ -
cation task, but a much weaker decline in performance for between-category
discrimination. Although the primary pattern of results across the two sets of
stimulus materials is similar, Pisoni’s (1973) ¿ ndings suggest that the categor-
ical perception of phonemes has limits. In particular, consonants and vowels
may be processed differently, particularly with respect to the discrimination
of acoustically similar tokens.
By the mid-1970s, speech scientists had also begun to investigate the per-
ception of social sources of variation in the speech signal. For example, Lass et
al. (1976) asked naïve listeners to categorize unfamiliar talkers by gender based