314 Thomas C. Purnell
male subject speaking in three dialects (Purnell et al. 1999): African American
English (AAE), Chicano English (ChE), and Standard American [Northern]
English (SAE). There, it was reported that utterances like “Hello, I’m calling
about the apartment you have advertised in the paper” spoken in these three
dialects were identi¿ ed accurately in a matched guise experiment. When the
word “hello” from this test phrase was played to listeners by itself, the dialect
used in producing that word was identi¿ ed better than chance. The interpreta-
tion made by Purnell et al. (1999) was that the speaker produced some identi-
¿ able acoustic characteristic that acted as a trigger for dialect identi¿ cation.^15
The goal of the present study is to understand this trigger in light of questions
raised previously regarding the relation between acoustic characteristics and
perceptual cues, and the presence of latent factors.
For comparison purposes it is worth noting that Thomas and his col-
leagues ¿ nd differences in acoustics and perception of (Southern) White Eng-
lish and African American English from both source (intonation, creakiness)
and ¿ lter (vowels) properties.16, 17 This ¿ nding con¿ rms what was reported by
Purnell et al. regarding the acoustics of the dialects. We will also see that the
similar acoustic characteristics identi¿ ed by Thomas participate in the cross-
dialectal co-variation when using discriminant analyses. What is unclear from
previous work is the weighting of acoustic characteristics as perceptual cues,
and the relative transparency in Figure 13.1(d) where the focus is primarily
across social group. The results from §3 suggest that in the present data we
should ¿ nd latent factors at work and that source and ¿ lter characteristics may
not be equally weighted.
4.2 Methodology
In the Purnell et al. 1999 study, the solitary “hello” tokens were selected for
measurement because they had been used in a perceptual experiment and pro-
duced results indicating that speakers must be listening to phonetic information
instead of phonological, lexical or grammatical details. In the present study, the
same tokens were re-measured. The measurements in the present study were
selected to characterize our understanding of the interface in the perceptual
dialectology model in Figure 13.1 and of the source and ¿ lter cue differences
across racially af¿ liated dialects. So, although this new set of measurements
includes a number of the same measures used earlier, it does not use all of
them, e.g., harmonic-to-noise ratio, jitter, shimmer and segment durations. The
¿ rst set of acoustic measures was those that are for the most part ¿ lter-oriented,
such as vowel steady state characteristics for both vowels. Speci¿ cally, these
measures included F1, F2 and F3 using both hertz and transformed bark values,