Towards A Realism-Based Metric for Quality Assurance in Ontology Matching
Werner Ceusters
International Conference on Formal Ontology in Information Systems (FOIS 2006)
Baltimore, Maryland (USA), November 9-11, 2006
Abstract
Ontology matching is commonly defined as a matter of finding semantic correspondences between terms in ontologies as a first step towards their merger, mapping or alignment. However, it has been pointed out that there still prevails no common understanding of what such ‘semantic correspondences’ are supposed to be, and that in consequence “human experts do not agree on how ontologies should be merged, and we do not yet have a good enough metric for comparing ontologies.” In what follows we define such a metric, starting from the thesis that if two or more ontologies are to be considered for matching, then however much they may reflect distinct views of reality on the part of their authors, the portions of reality to which they refer must still be such as to overlap. Our approach takes account of the fact that both authors and users of ontologies may make mistakes (the former in their interpretation of reality and in the formulation of their views, the latter in misinterpreting the former’s intentions). To do justice to such factors, we need to draw a distinction between the three levels of: (1) reality; (2) cognitive representations; and (3) publicly accessible concretizations of these representations. We can then define ‘semantic correspondence’ not, as is usual, in terms of (horizontal) relations of assocation between the terms within the ontologies to be matched, but rather in terms of the (vertical) relation of reference (so that terms correspond semantically if they refer to the same entities in reality). One conclusion of our argument is that, when ontology matching has been used as the first step towards ontology merging, then the merged ontology can contain inconsistencies only if there are already inconsistencies in at least one of the source ontologies.