We view the development of an adequate account of phonetic capacity as the chief goal of experimental phonetics. There appear to be two important tasks. First, it it is necessary to determine the membership of the set of universal phonetic features, since these are the basis of phonological capacity and the elements of phonological competence. Moreover, we want to understand the role of the various stages in the speech chain in the psychological definition of each feature. These stages include (at least), the activation of the muscles of the vocal tract by neuromotor commands, the gestures made in response to these commands by the various articulators, the resulting dynamic changes, not only in the shape of the vocal tract but also in air pressure and airflow at different points in the tract, and finally the cues in the acoustic output. It may well be that not all these stages are pertinent to phonetic capacity; on the other hand, other stages, as yet poorly understood, may be involved.

The second task is to characterize psychologically the translation from the discrete, essentially timeless phonetic level to the continuous, time-bound activity characteristic of lower levels. If at any of these levels speech could be consistently separated into stretches corresponding to phonetic segments, the problem would be fairly simple; but we know that this cannot be done. At any level in speech which we can observe, there are no true boundaries corresponding to any phonetic unit shorter than the breath-group, though in the acoustic signal there are many apparent boundaries reflecting articulatory events, e.g. stop closures and releases, spectral discontinuities in liquid and nasal sounds, onset and offset of voicing and the like. Yet the psychological reality of phonetic segments can hardly be doubted; and at any rate, without them, phonology would collapse.

The feature-specified, dynamic vocal tract model by which we would represent phonetic capacity is on just the same level of theoretical explanation as the phonological and syntactic models of linguistics. It does not have, as yet, any but the most general sort of neurophysiological basis; it does not account in itself for productive or perceptual performance; in particular, it does not conflict with an analysis-by-synthesis account of speech perception. It is simply a way of stating some properties which the neural mechanisms for speech must incorporate to account for the observed behavior of speaker-hearers.

By virtue of his phonetic capacity the speaker-hearer acquires certain phonetic skills, just as he acquires the phonological rules of his language by virtue of his phonological capacity. He must 'calibrate' his perceptions to allow for the idiosyncrasies of the vocal tracts of the other speakers to whom he listens, even before he understands his native language. (If there were no other reason for postulating phonetic capacity, we would want to do so to account for the fact that an infant learns to interpret the output of the vocal tract of each of the individuals around him, though these vocal tracts differ radically from one another and from his own in size and shape; and he does so with sufficient accuracy to permit the collection of the 'primary linguistic data' (Chomsky 1965: 25) essential for language