Approximate Reasoning using Linguistic Valuations

The knowledge given by an expert in order to construct a Knowledge Based System is often not very precise and contains uncertainty. In general, the expert prefers to express the uncertainty in a qualitative form rather than in a quantitative on . In this paper, we present one model of the uncertainty in qualitative form. A linguistic approach in a context of a many-valued logic is proposed.We also present, all necessary tools to represent and to manage the uncertain knowledge in an Expert System. Contrary to machines, Man perfectly manages uncertain and imprecise knowledge. The following sentences are fully understandable by them: they are about 200 young persons. the temperature is very high. a very large suite. cough is quite frequent. One of the most important problems encounted during the realization of expert systems is that the knowledge representation is often chosen independently from its management. But, the representation has not any signification that allows us to know if it represents correctly or not the related knowledge. So, we have to study the doublet (use of the knowledge, use of its representation) specially when we manage uncertainty and imprecision. There are several mathematical formalisms and tools to represent such expert knowledges (probabilistic logic, default logic, fuzzy logic, etc...). We have introduced [1, 2] fundamental concepts of a many-valued logic to represent the knowledge, using linguistic valuations to translate both uncertainty and imprecision. A mechanism to manage the given knowledge in a symbolic way, avoiding computations is also proposed in [1, 2].