UC Berkeley Phonology Lab Annual Report (2011) Retroflex versus bunched [r] in compensation for coarticulation Keith Johnson UC Berkeley This paper presents data from two experiments that are problematic for the auditory spectral contrast theory of compensation for coarticulation, but are perfectly compatible with the gesture recovery theory. The experiments compare compensation for coarticulation effects in the perception of a [dɑ]-[ɡɑ] continuum produced by retroflex [r] context, versus the lack of an effect for a bunched [r] context. The key acoustic feature referenced by the spectral contrast theory (the lowered F3) is present in both bunched and retroflex [r]. In experiment 2, the F4 trajectory (a key acoustic correlate of retroflexion) is seen to modulate compensation for coarticulation. INTRODUCTION “Compensation for coarticulation” is the name of a phenomenon that has two explanations. In one theory (Holt, 1999; Lotto & Kluender, 1998), the effect is due to auditory spectral contrast as a context syllable interacts with a following target syllable in the auditory system. The main competing theory gave the phenomenon its name (Mann, 1980; Fowler, 2006). In this theory, the listener perceptually compensates for the coarticulation that typically occurs between the context syllable and the target syllable. Thus, the interaction between context and target reflects a perceptual gesture recovery process. The particular instance of compensation for coarticulation that is studied in this paper is the one reported by Mann (1980). The perceptual boundary on a stop place of articulation continuum from [dɑ] to [ɡɑ] is shifted by a preceding [ɑl] or [ɑr] context syllable. If the context syllable ends in [l] the boundary is shifted toward [d] - more of the tokens are perceived as [ɡɑ]. And, if the context syllable ends in [r], the boundary is shifted toward [ɡ] - more of the tokens are perceived as [dɑ]. The effect can be quite striking and makes an effective classroom demonstration of context effects in perception. Mann (1980) called this phenomenon “compensation for coarticulation” on the hypothesis that listeners “parse” the effects of coarticulation as they perceptually recover the intended gestures of the speaker. The idea is that because tongue position in [l] is more forward than it is in [r], a following ambiguous syllable on a [dɑ]-[ɡɑ] continuum will sound more like [dɑ] in the [ɑr] context because listeners can attribute some of the “retraction” of the ambiguous syllable to coarticulation with [r]. On the other hand, when the context syllable ends in [l] there is no such coarticulatory cause of the “retraction” of the ambiguous syllable, so it is heard as intentionally back and more like [ɡɑ]. Evidence for the gesture recovery interpretation (e.g. Fowler, Best, McRoberts, 1990; Fowler, Brown & Mann, 2000; Fowler, 2006) has been unconvincing. By this I mean that scholars who are not predisposed by training to see perceptual phenomena in terms of gesture recovery have remained