1. I want in this paper to draw attention again to the theoretical basis of the use of distinctive features in generative phonology, and with that in mind, to consider what improvements can be made. One function of distinctive features is to provide a formal means of expressing the notion of a NATURAL CLASS (cf. Harms, 1968: 26), such that a phonological rule which applies to a natural class of segments may be expressed in a simpler way than a rule applying to some other class of segments. Distinctive features, and notational devices and conventions, are intended to capture this notion of simplicity so that by examining a formalized rule one may discover, counting the symbols in it according to a set of values provided, at least whether the rule is simpler than a comparable rule, if not its simplicity in any absolute sense. It is also intended that the set of features should be a substantive universal of language, though, as Harms' examples (1968: 23–38), drawn from various people's analyses of various languages, show, it cannot yet be seen that there is a tendency towards agreement on which these might be, as Harms himself (38) is aware. He sees the problem, however, in these terms:

The basic set of features can be viewed as a hypothesis about language, subject to empirical validation. Arguments for adding new features to the list or for altering the basic features must demonstrate the inadequacy of the basic hypothesis. Such arguments, however, cannot be based upon an appeal to the simplicity metric, for the features themselves are elements in the simplicity metric. Phonemic solutions that do not assume the same set of features cannot be compared in terms of simplicity. The claim must be that one solution provides a more reasonable hypothesis of the phonological structure of the language in question.