Linguistic approximation and semantic adjustment in the modeling process

The transcription of data into a discrete representation system (numerical or qualitative) may be inaccurate either because there exist no exact representation or because a concise but inexact description is preferred. This kind of inaccuracy is studied within a general framework using Description Languages, either Qualitative or Numerical. Two principles are stated: 1) the writer always selects a description following a complexity-accuracy tradeoff (Linguistic Approximation Principle); 2) in a Description Language, in every specific context, the Meaning of an expression optimally represents the set of values that it can describe (Semantic Adjustment Principle). Consequently, the Meaning of Qualitative Expressions 1) may change in different contexts; 2) can generally be determined without introducing arbitrary parameters. The corresponding algorithm is presented in the case of Linguistic Modeling, to calculate the fuzzy values associated with Qualitative Expressions.