Comments on Leydesdorff's'a validation study of Leximappe'

Abstract

We thank Loet Leydesdo~ l for sending us the text of his meticulous research on coword analysis. I take this opportunity to reply to him, and thus to contribute to a dialogue that I hope will prove fruitful. Leydesdooe f has for some time been using information theory with success for smoothing curves. However, is that in itself sufficient to prove that information theory is a universal tool for evaluating others? It is above all a mathematical tool that does not concern itself with the meaning of the codes that it studies (what is called the second articulation of language in linguistics). Leydesdorff uses information theory in order to 'evaluate' the Leximappe method of eoword analysis of scientific texts. The question we pose is whether it can be used as an evaluation tool. First of all, we should spell out the premises of the two approaches: information theory and coword analysis. Information theory analyses messages made up of a combination of codes taken from a finite list. It is a theory used by engineers to measure the capacity of their communication channels. A code's 'information content' which it arbitrarily calls quite simply 'information' is a function of its unpredictability. It follows that the information transmitted by a series of messages (for example messages in a language like French) is maximum if the codes (for example the letters of the alphabet and punctuation marks) are equiprobable. However, information is structured by various more or less predictable regularities such as the doubling of consonants, vowels at the end of words etc. This repetition serves to attenuate the quantity of new information carried by language in order to make things easier for the interlocutor. When one takes the word as a unit in itself, these regularities can be regarded as the expression of regularities inherent in the organization of situations being described: a postman is more likely to talk about letters than vegetables. In other words, letters