Domain semantics typically allow to describe things very precisely and with a precise meaning. In many cases, however, not everything that can be said is either true or false, but often true or known only in a limited domain of validity: a certain time span (e.g. the temperature of an object), a spatial domain (e.g. a law), a certain simplification of a theory (e.g. a proportionality coefficient), or even with a certain claim (which still might be unvetted or even wrong). Without general ways of expressing such conditional validity, a multitude of practical challenges arise: Triplets are generated to express facts that are valid only in a specific context and therefore can no longer be used meaningfully together with other triplets from other contexts in automated reasoning, as they would lead to false conclusions. Also, it often requires overcomplications of concepts, like the concept of identity, when a process description needs different identifiers for the same object at different times. While some domain ontologies have tools to express such cases, there seems to be no general solution. Natural languages, however, typically have general grammatical constructs like “According to A. et al. (2024) the Pinatubo was active from April 2, 1991, to June 15, 22:30, ...” to express the conditionality of claims. The poster exemplifies these challenges and discusses possible general solutions that mimic these natural language capabilities while still enabling complete formal reasoning. The idea is based on the use of identifiers that represent claims, which can then be declared valid under certain conditions. The proposed approach allows to extend deduction rules of the domain ontologies to deduction rules for such conditional claims.
Ballani et al. (Tue,) studied this question.