AN INFORMATIONAL THEORY OF THE STATISTICAL STRUCTURE OF LANGUAGE 36

Unfortunately, none of the elements is given in a form directly suitab~e for use in a mathematical theory, but must first be extracted by qualitative discussion and analysis from all the obtainable data, historical and structural, on languages, and eventually in other fields. Such a qualitative discussion cannot give us more than a reasonable degree of confidence in the results to which -it leads. Therefore the aim of a mathematical theory cannot be, strictly speaking, to deduce one of the elements from the remaining two, but only to add (very considerably) to the confidence we put in each of these elements individually, by showing that they are fully compatible. However, frolp. a technical (as opposed to conceptual) point of view, the proofofthe compatibility cannot avoid the use ofone ofthe above-mentioned deductive methods.