ABSTRACT
The authors outline two types of redundancy, tlw.t arising from constraints of the human speecl1 and auditory systems and that nrising from language-specific characteristics. I maintain that there is another type of redundancy, nmncly that arising from our selection of sets of features used to characterize these aspects of language. Decause the fentures are not orthogounl to each other, not all combinations occur, and redundancy is built in: Thus, the choice of the set of features determines some of the redundancies. This is a problem because we don't know or can't agree on how to choose features, i.e., on what counts as evidence for setting up a particular feature. Now, for a diverse set of reasons, people have come up with systems of features that arc not orthogonal. One of the reasons that there is no good agreement on what the features arc is that devising an orthogonal or elegant system for a given language is a quite different problem from devising a system of all possible features in all possible languages. For some tasks, we want features that can mnke distinctions as fine as the speech perception and production systems are capable of making. For other tasks we want features that are capable of uniting a great variety of different sets of sounds that are, for various reasons, linguistically "the same." What is elegant and economical for one task is not necessarily so for others. Thus, assuming that there is one set of features for these various tasks implies excess baggage-i.e., redundancy-and, as such, says more about how we are choosing to look at our data than it says about language.
