228 Bartáomiej Plichta and Brad Rakerd
2.2 Extrinsic normalization
Proponents of extrinsic normalization argue that vowel perception requires
a frame of reference that is established based on information obtained from
sources beyond the vowel itself. Ladefoged and Broadbent (1957) showed that
this information can come from an earlier portion of an utterance that is in
progress. They synthesized multiple versions of a lead-in sentence (“Please
say what this word is.”) by modifying its formant structure to simulate speech
samples from talkers with differing vocal tract characteristics. Then each of
the alternative forms was presented to listeners followed immediately by a
set of common target words. The notable ¿ nding was that variations in the
lead-in systematically inÀ uenced perception of target word vowels, strongly
suggesting that listeners normalize for talker differences in real time.
- Perception data collected “in the ¿ eld”
The present study, like that of Ladefoged and Broadbent, asked whether per-
ception of a vowel occurring at the end of an utterance may be inÀ uenced by
information conveyed earlier in that utterance. But in this case the informa-
tion of particular interest was sociophonetic, speci¿ cally information regarding
NCCS. To learn whether and how this information might inÀ uence a listener’s
perception of vowels, we needed to gather data about a rather large inventory of
speech samples. We also needed to test listeners who had diverse sociophonetic
histories, and who resided in distant parts of Michigan. A decision was therefore
made to conduct data collection for this study in the ¿ eld, so much as possible.
Variationist sociolinguistic research has been dominated by ¿ eldwork
(Chambers 1995). Traditionally, sociolinguists prefer to collect their data at a
place of the subject’s own choosing, such as the home or place of employment.
The argument of observer’s paradox (Labov 2001) demands that ¿ eldworkers
not create an atmosphere in which the authenticity and naturalness of the lan-
guage sample is compromised. Studies of speech perception, on the other hand,
have generally been conducted in laboratory settings, where the environmental
acoustics are well controlled and where the ¿ delity of stimulus presentation can
be assured.
In designing the current sociophonetic speech perception study, a con-
scious effort was made to observe the main principles of sociolinguistic ¿ eld
research while at the same time providing controlled and rigorous experimen-
tal conditions to the extent possible. Therefore, the experiment was run at a
quiet place of each participant’s own choosing, with the stimulus presentation
and response collection achieved by means of a portable (laptop) computer.