Structure Formelle des Textes et Communication

Abstract

From the experience that one has of speech organs, it could seem that it follows that the physics and the physiology of communication have no influence upon the deeper and more general structural properties of language. We propose to prove that, in two respects at the least, this is not in fact the case, and that the properties of speech turn out to be those which would be attributed to it if language were reconstructed from scratch, account being taken only of the pragmatic conditions of communication, and not of its aims. A) The fact that the acoustic or pictural carriers of speech are continuous physical objects, susceptible to disturbance by noise, requires that there exist also another aspect to speech: a discrete message. We shall also derive in this way other properties of language. B) The statistics of words, considered as empty forms, collected by Estoup, Chotlos, Baker, Zipf, Josselson, et al., in widely varied languages and authors, can be explained by assuming that the discrete messages due to each individual author are built out of "concrete units" identical with words, and that the statistics of these units are, intentionally or otherwise, those which transinit the greatest amount of information (in a technical sense used by Claude Shannon), compatible with a given complication of the receiving apparatus, expressed by an average cost of decoding a word, by an ideal decoder. A summary of our mathematical theory has been given by Miller (1954). These two properties seem to be a beginning of confirmation of in axiom of cybernetics.