Can We Enforce Full Compositionality in Uncertainty Calculi?

At AAAI'93, Elkan has claimed to have a result trivializing fuzzy logic. This trivialization is based on too strong a view of equivalence in fuzzy logic and relates to a fully compositional treatment of uncertainty. Such a treatment is shown to be impossible in this paper. We emphasize the distinction between i) degrees of partial truth which are allowed to be truth functional and which pertain to gradual (or fuzzy) propositions, and ii) degrees of uncertainty which cannot be compositional with respect to all the connectives when attached to classical propositions. This distinction is exemplified by the difference between fuzzy logic and possibilistic logic. We also investigate an almost compositional uncertainty calculus, but it is shown to lack expressiveness.