The mechanisms involved in the production of a speech sound vary from one speaker or iteration to the next, resulting in considerable variation in the phonetic realisation of speech sounds. The exact settings of the articulators in the production of a speech sound will vary from one speaker to the next, but, even within the same speaker, no two repetitions of a sound are quite the same. Furthermore, the phonetic realisation of a common form may vary according to an individual’s dialect, speaking style, or even in the way sounds coarticulate or overlap with each other in connected speech. Yet, despite the intrinsically gradient nature of phonetic variation, listeners are usually successful in working out the intended message of the speaker by extracting meaningful categories from the speech signal. It goes without saying, therefore, that interpreting the intention of the speaker along with the speech form he/she produces in a given context is central to the process of mapping between speech segments and abstract representations. Put another way, the communicative use of language necessarily involves extracting ‘constancy’, in the sense of category membership, from ‘variation’ in any given context.