Retranslation: A problem in computing with perceptions

In the area of Intelligent Systems, the emerging field of computing with perceptions is a research program that looks into the capability of the human mind of using perceptions to make decisions and perform complex tasks. The primary aim of Intelligent Systems is to build perception-based machines capable of working with statements in natural language.;In dealing with natural language, there are three main types of procedures: (a) Procedures that convert natural language into fuzzy propositions; (b) procedures that manipulate these fuzzy propositions through approximate reasoning and yield other fuzzy sets; (c) procedures that convert these given fuzzy sets again into statements in natural language. These last types of procedures are known as linguistic approximations of the second kind or retranslations.;The problem studied in this dissertation is retranslation or the problem of converting fuzzy propositions obtained by approximate reasoning to statements in natural language. This problem has been little researched.;The methods used in dealing with the retranslation problem are vocabulary-based methods, which have a predefined set of linguistic terms represented by fuzzy intervals; standardization methods, which are construction methods in which a given fuzzy set is expressed as another fuzzy set in a "standard form"; and defuzzification methods, which are special standardization methods for constructing crisp intervals.;The relevant criteria in this dissertation are informativeness, validity and fuzziness. The primary criteria: informativeness, a well established measure of information from Generalized Information Theory; validity, taken as a degree of subsethood of a given fuzzy set into another. Fuzziness is measured as the overlap of the fuzzy set and its complement.;The main results are the standardization method as a powerful method of dealing with the retranslation problem, as well as its computational procedure. Other contributions are distinguishing disambiguation from defuzzification and the overview of methods of defuzzification in the literature to the present.;The dissertation deals only with numerical linguistic variables and only standard fuzzy sets are used.;Keywords. Computing with perceptions; fuzzy sets; fuzzy intervals; natural language; validity; informativeness; retranslation; defuzzification; disambiguation...