Semantische Integration durch Reinterpretation: ein formales Modell

This work presents a formal model for integrating heterogeneous knowledge sources which relies on the principle of reinterpretation. Conflicts occurring in the integration are resolved with a conservative strategy in which symbols involved in a conflict get a new interpretation; the result contains hypothesis on the semantical relatedness of the different readings of the conflict symbols. The formal model focusses on a simple integration scenario according to which two well proven ontologies (receiver’s and sender’s ontology) with similar terminologies have to be integrated. Conflicts can be explained by ambiguities in the public (common) language. This kind of conflict is caused by different readings of the same common symbol in different ontologies. Following the main ideas of belief revision, one can formalize adequacy criteria for the intended integration scenario by postulates. It can be shown that (subclasses of) the adequacy criteria are operationalizable by reinterpretation operators. Reinterpretation operators are based on a conflict diagnosis that is different from the conflict diagnosis for belief-revision functions. But reinterpretation operators can be represented as belief-revision functions if one considers not the ontologies to be the object of revision but the set of hypothesis on the semantical relatedness of different readings of ambiguous symbols. Two incompatible subsets of the integration postulates are operationalized by two reinterpretation operators, type-1 and type-2 operators. Type-1 operators preserve the terminology of the receiver’s ontology, type-2 operators adapt to the terminology of the sender’s ontology. By choosing different sets of initial hypothesis on the semantical relatedness of ambiguous symbols operators of different strength can be defined. Differences between operators of different strength can be characterized by postulates that describe aspects of conservativity and minimality. Iterated applications of reinterpretation operators can be used for a sequentialized integration of ontologies. One-step integration and sequential integration by reinterpretation operators are not reducible to each other. Considering the stability of iterated reinterpretation operators results in additional differences in the set of reinterpretation operators (of type 2) with different strength. Weak operators are stable but may result in an amnesic adaption to the terminology—thereby loosing connections to concepts of the receiver’s ontology. Strong operators realize a less radical adaption but are instable.