Sociophonetic Applications of Speech Perception Experiments

ALTHOUGH STUDIES OF PERCEPTION are still largely assigned to the realms of experimental phonetics or psychology, sociolinguists have been recognizing the importance of perception. Several lines of experimental inquiry have emerged. Nevertheless, perception has been studied far less by sociolinguists than has speech production. One reason is that speech perception is daunting at first. Examining it requires careful attention to experimental design, a considerable amount of preparation, and, in many cases, use of a speech synthesizer. Even so, research on perception can be highly productive. This paper attempts to review the sorts of experiments that have been conducted in the past and to provide guidelines for sociolinguists interested in studying perception, with suggestions for future work.

Although perception has been a neglected stepsister of production in sociolinguistics, it, like Cinderella, may have its day soon. Two important factors could—and should—move perception to the forefront of sociophonetic research. One is simply the huge potential for sociolinguistic perception studies because the area has been neglected for so long. The other reason is a more practical one: although perception experiments require extreme attention to detail in the preparation phase, data analysis is generally less time-consuming than in production studies, and this difference may make it more attractive to researchers.

The aversion of much of sociolinguistics to perception has been, to some extent, more apparent than real. Many sociolinguistic studies over the past generation, especially instrumental studies, have succeeded in divorcing speech production from speech perception. However, perception issues may play a hidden role in studies that ostensibly address production. The reason is that variationists have not always carefully distinguished production from perception. This tendency is an artifact of the reliance of sociolinguistics on impressionistic transcription. The impressionistic tradition, based on the development of the International Phonetic Alphabet and of the Cardinal Vowel system of Daniel Jones, dominated dialect geography, which was largely conceived before modern acoustic equipment was developed. This reliance has continued, for the most part, in sociolinguistics. The shortcoming of this approach is that, although linguistic atlas transcribers and sociolinguists believe that they are recording subjects' production, what they actually record is filtered through their own perceptual labeling abilities, practices, and strategies (Kerswill and Wright 1990).

An example of a hazard inherent in such a mixture involves one of the most frequently studied phenomena of English: the deletion of final stops in consonant clusters, as in lift pronounced [lIf] or desk pronounced [des]. Studies of consonant cluster simplification have customarily involved impressionistic transcription. However, Browman and Goldstein (1990) showed experimentally that, because of the overlapping nature of articulatory gestures, listeners may be unable to hear a stop in a consonant cluster, such as the [t] in perfect memory, even when an articulatory gesture is demonstrably present. Surprenant and Goldstein (1998) furthermore found predictable patterns in the perceptual masking of stops in clusters. These patterns matched many of the results reported in sociolinguistic studies of deletion of consonants in clusters (e.g., Wolfram 1969; Labov 1972; Guy 1980). The implication is that many of the stops tabulated in sociolinguistic studies as "deleted" were, in fact, articulated but were inaudible to the transcribers. Kerswill and Wright (1990) found that the opposite can happen as well: listeners may record an alveolar stop when none exists. Studies of consonant cluster simplification primarily describe the perception of transcribers and only secondarily the production of speakers. The binary interpretation of final stops as either present or deleted may describe perception satisfactorily, but it cannot truly describe production (see Janson 1983, 22-23, for another example).

Marriages of production and perception are certainly not invalid per se; as Lindblom (1980) notes, acoustic measurements of speech are meaningless unless they can be related to perceivable factors. However, variationists need to acknowledge more consistently the fact that production and perception do not always correspond neatly. Directing more studies at perception would help them to do so.

Previous Socioperceptual Approaches

Sociolinguists and other investigators...