Linking a distributed view of cognition to an integrational perspective on language, learning to talk is presented as an ontogenctic achievement. Examining this as an epigenetic process permits an innovative sketch of how infants come to be heard as producing grammatical utterances. Appealing to 'shallow' or content-free cognition, I show how adjustments by contextualizing bodies allow adult overinterpretations to shape infant doings. Far from needing,representations', the baby uses joint activity, affect, and self-directed anticipative learning. Humans, then, use affective co-ordination to develop neurophysiological biases for speaking/hearing vocalizations around syllabic structures. This promotes a kind of agency that allows a 2 year old human, like an encultured bonobo, to act in ways that appear to be self-implicating, self-directing, self-regulating and self-serving. Both species can use (what we hear as) abstracta in novel and coherent behaviour. Unlike its wild counterpart, however, a human needs no external computational hardware. Rather, her achievement derives from strategic use of contextualizing bodies to gradually discover the rewards that accrue from taking part in utterance-activity. (C) 2004 Elsevier Ltd. All rights reserved.
