Linguistic segments, acoustic segments, and synthetic speech
Leigh Lisker · Language · 1957
Linguists, like many other people who feel called upon to talk about language, are in the habit of saying that speech activity is continuously variable in nature. Having said this, they proceed to describe particular samples of speech as sequences of events which they isolate on the basis of either articulatory definitions or acoustic definitions, or both. In carrying out this operation they are not seriously inconvenienced by the continuous aspect of speech: the fact that boundaries between any two of the elementary events composing an utterance cannot be fixed with exactness does not constitute a problem for them. They may occasionally use metaphoric expressions such as ‘overlapping phones’ and ‘slurring’, by way of acknowledging the continuously varying character of speech; but the function of such expressions is to justify the keeping of a discrete representation in the face of any demonstration that the physical segmentation of speech is hopeless. (That it is in fact not hopeless at all is beside the point here.) Speech may indeed be continuously varying when looked at in the laboratory, but the basic fact about speech is that human beings can hear it as a sequence of auditory fractions. Linguists may divide an utterance into a larger number of fractions than other listeners do, but even linguists have never been forced to concede the inadequacy of a discrete representation on the ground that it did not account explicitly for the physical continuous character of speech.