The Role of Socioindexical Expectation in Speech Perception.

Listeners extract cues to speaker identity from the speech stream. Recent evidence suggests that listeners will also perceive these socioindexical cues, even if absent, when primed to expect them. Most researchers interpret these findings as evidence for exemplar models of speech perception (Niedzielski, 1999; Hay et al., 2006b; Staum Casasanto, 2009a). At least one early line of research, however, attributes the influence of socioindexical knowledge on speech perception to listeners’ negative bias (Rubin, 1992). A series of three experiments with experienced and inexperienced listeners investigates the use of socioindexical expectation during speech perception. The first experiment, a yes/no accent identification task, reveals that listeners, whether experienced or inexperienced with Chinese-accented English, are capable of judging the authenticity of a non-native accent. Experienced listeners are significantly more accurate and inexperienced listeners are significantly more likely to rate an imitated accent as authentic –suggesting they depend more heavily on stereotypical features. Experiment 2 addresses whether listeners can use socioindexical expectations to enhance speech perception. Both experienced and inexperienced listeners were significantly better at transcribing Chinese-accented sentences in noise when presented with an Asian face than when presented either with a silhouette or a Caucasian face. This result suggests that the negative bias hypothesis can not be correct; listeners can use socioindexical cues to enhance speech perception. Experiment 3 used eye-tracking to investigate the time course of the influence of socioindexical expectation. Inexperienced listeners hearing a Standard American English voice showed no significant difference in fixation latencies when presented either with an Asian or Caucasian face. Listeners shown an Asian face showed significantly longer dwell times to the target image late in the trial, however. This result reinforces the finding that the negative bias hypothesis can not be correct but is also not consistent with an exemplar based account in which socioindexical expectation pre-activates groups of socially labeled exemplars (Johnson, 2006). These results point to the need for more natural –more linguistic– tasks in the investigation of socioindexical speech perception and the need to look closely at time course to better understand the role of socioindexical expectation in speech perception.Ph.D.LinguisticsUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/89684/1/clunis_1.pd