Procedural theories of linguistic performance
In the modern theory of language it has been found useful to distinguish between questions of ‘competence’ and questions of ‘performance’. The distinction has at least two aspects. First, it recognizes that the description of a language as such is logically distinct from an account of the way in which particular people use that language, but, secondly, it separates questions of grammaticality from questions about naturalness or intelligibility. It is argued that, while the former distinction is valuable, the latter has now outlived its usefulness. A generative grammar can be regarded as an adequate model of the ideal speaker’s competence only if it is accompanied by a specification of processes by which ideas could be encoded in words, and these words subsequently decoded by the hearer. Examples are given of effective procedures, implemented as computer programs, for the performance of specific linguistic tasks; one of these, due to A. C. Davey, is a model of the production of connected English discourse; another, due to R. J. D. Power and myself, is a device that learns, from representative number—numeral pairs, the numeral systems of a variety of natural languages.