ABSTRACT
The systematic analysis of variation has always been a concern of phonetics, but until recently it has not been pursued at more abstract levels of linguistic structure. As long as the theory of speech perception and production looked to a one-to-one, or one-to-many, relationship between linguistic elements and their physical realizations, this was not a major problem. But the weight of the evidence points to a many-to-one relation between linguistic and physical representations; a recognition system that employs probabilistic weighting of variable features in a multivariate model; parallel processing of the signal at many levels; and interrupts that inform the processing at one level of results obtained at another.
