Children typically acquire adult realisations of /r/ after most other adult-like targets (Vihman 1996: 218-219). Impressionistic transcriptions in the literature often equate defective realisations of approximant /r/ with [w], or are labelled as labiodental approximants. For most children, a perceptual miscategorisation of adult /r/ realisations as /w/ does not explain these patterns (Slawinski and Fitzgerald 1998). Instrumental production data are lacking. However, in many cases of apparent [w] realisations, 'covert contrasts' too fine-grained for adult transcribers to detect and represent in transcription may occur (see Scobbie et al. 2000).
The status of labiodental /r/ as a defect has been called into question in recent acoustic work on adult speech (Jones 2004a, b). Principled similarities between the acoustic structure of labiodental /r/ (possibly involving lingual constriction) and the acoustic structure of apical approximant /r/ suggest that labiodental /r/ users may aim to produce a midfrequency resonance at around 1400-1600 Hz. For labiodental /r/ users, this resonance is F2, for apical approximant /r/ users, it is F2+F3 (or F2+FR, Stevens 1998). Labiodental /r/ may persist into adulthood and be hard to correct in clinic because it fulfils a specific intended function, rather than being a 'best attempt' due to immature control of the articulators.
Regardless of its origins, in continuing realisations associated with child-speech into adulthood, the use of labiodental /r/ is evidence for non-plasticity. Those speakers who employ it have 'fossilised' this aspect of their speech, irrespective of the overwhelming majority of non-labiodental tokens of /r/ in the linguistic environment. This latter point raises problems for phonetically-detailed exemplarbased models of speech perception which view frequency of occurrence as an important determinant of target forms (e.g. Pierrehumbert 2001).
The research presented here will begin to address the question of why labiodental /r/ users maintain a realisation of /r/ which differs from the majority of /r/ tokens around them. Following on from the hypothesis that labiodental /r/ users view labiodental /r/ as perceptually equivalent to apical approximant /r/, we test this empirically in perceptual experiments of labiodental /r/ users and nondefective /r/ users. The implications of the results for clinical treatment of labiodental /r/ and phonetically-detailed exemplar-based models of speech perception are discussed.