A methodology for computing with words

Abstract An alternative interpretation of linguistic variable is introduced together with the notion of a linguistic description of a value or set of values. The latter is taken to be a fuzzy set on words where the membership values quantify the suitability of a particular word as a label for the value or values being considered. This concept is then applied to reasoning with linguistic quantifiers these being defined as linguistic descriptions of probability values. From this viewpoint linguistic quantifiers are constraints on probability values and hence using the voting model and Bayesian methods infer second order densities. In this respect such quantifiers can be view as an alternative form of imprecise probability. These ideas are then used in our proposed methodology for converting probabilistic inference rules into linguistic inference rules and a computationally cheap approximation algorithm for such rules is then introduced. The approach is illustrated in a number of worked examples using various types of rules including linguistic syllogisms and a linguistic version of Jeffrey's rule. Finally a number of methods for information fusion with linguistically quantified statements are discussed.