Sound judgements: perception of indexical features in children’s speech

Paul Foulkes & Gerry Docherty, Ghada Khattab Malcah Yaeger-Dror(3)

University of York, UK & University of Newcastle, UK, University of Arizona, USA


Several recent studies (e.g. Hawkins 2003, Pierrehumbert 2002) have suggested that rich, substance-based lexical representations, as proposed by the advocates of exemplar models, may provide a plausible framework within which to account for the systematic social-indexical patterns found abundantly within speakers’ performance. One feature of such models is the assumption that, as a result of experience, individuals develop an implicit awareness of the variability associated with a lexical item and of the social meaning associated with that variability. In this paper we describe an experiment which investigates the extent to which individuals’ performance in a perceptual task reflects their awareness of aspects of social-indexical variability.

In Newcastle English (t) is a particularly complex sociolinguistic variable. In word-medial contexts (water) glottalised variants and plain [t] are the dominant forms, but the latter is significantly more frequent in female speech. In pre-pausal context (cat#) the dominant variant is plain [t] in male speech, but pre-aspirated [ht] in women’s speech. Our experiment sought to ascertain whether these gender-correlated variants contribute to listeners’ ability to identify a speaker’s gender. It is likely, of course, that many phonetic cues contribute to gender identification, most notably fundamental frequency (f0). We therefore constructed our test using children’s speech, which is inherently androgynous with respect to f0. 67 single-word items containing medial or final (t) were extracted from recordings of three boys and three girls aged 3;0-4;1. The items were played to a group of 20 listeners from Newcastle, who we predicted might show awareness of the gendered (t) patterns on account of their experience of the dialect. The samples were also played to two control groups who were predicted not to be aware of the patterning (35 listeners from elsewhere in the UK and 114 Americans). The instructions to the listeners were simply to identify the speaker of each stimulus as a boy or a girl.

Overall there was, as anticipated, considerable variability in the responses. The number of correct responses was around 50% for all listener groups, while certain stimuli generated a strong consensus as to the sex of the child. Several factors made a contribution to gender identification, including f0, voice quality, amplitude and tempo. Tokens were coded for these factors as well as sociolinguistic variants. Regression analysis confirmed that all factors were significant in accounting for the responses. For example, breathy voice quality and low amplitude tokens elicited significantly more ‘girl’ responses. Such findings were apparent for all three listener groups.

However, the sociolinguistic variants elicited different responses from the listener groups. With medial (t), Newcastle listeners gave significantly fewer ‘girl’ responses to stimuli containing glottal forms than the other groups did. For final (t) the Newcastle listeners gave significantly more ‘girl’ responses to pre-aspirated tokens than the control groups. Both findings were predicted on the basis of the distribution of these variants in Newcastle English. They suggest that the Newcastle listeners are indeed able to exploit sociolinguistic variables for gender identification, in turn supporting those models of lexical structure which encompass fine phonetic detail.

Session: Paper session
Variation 4 (Phonological)
Thursday, April 3, 2008, 10:30-12:00
room: 15