The Web changes knowledge representation in a number of surprising ways, and decentralized knowledge representation systems such as the Semantic Web will require a theory of identity for the Web beyond current use of same As links between various data-sets. We propose that one problem, as posed by the semantics of same As in OWL, is enshrining a metaphysical notion of individual identity into the formal semantics of the logic that prevents universal identity linking across semantic roles. However, as shown by languages such as RDF and KIF, a metaphysics-free semantics is possible that can handle both classes and properties as first-class citizens of the universe of discourse that can have identity statements and so produce valid inferences. We empirically analyze the behavior of identity and inference on the Semantic Web currently in order to analyze the size of the problem, and show the non-normal distribution of equivalence classes inference produces. We also show via a human experiment how inference leads to more difficulty in judgments about identity by individual experts, yet adding experts leads to fairly reasonable results, even over inferred triples.
展开▼
