How are we to account for the fact that perception and production adapt to different environmental contexts; or for the fact that, in some well-known cases, higher computation is modulated by environmental factors? The challenge is to understand this dependency between computation and the environment in a way that maintains our intuition that there are two kinds of parameters here, deep grammatical dimensions of linguistic form and extra-grammatical environmental factors (the phonetics-phonology problem). I argue that non-linear dynamics provides a powerful mathematical language for formalizing phonetics-phonology. In the core proposal, phonological categories are identified with the attractors of a dynamical system. These attractors are tuned by environmental factors such as speech rate, speaker’s intentions and orthography. Preferred modes of the phonetic output derive as a function of the grammar dynamics and its sensitivity to external conditions. The composite system, grammar plus external conditions, self-organizes to meet varying task requirements. This seems to be the kind of flexibility we seek when faced with the fundamental problem of phonetics-phonology as an aspect of cognition.