A Framework for Ontologically-Grounded Probabilistic Matching

In all scientific disciplines there are multiple competing and complementary theories that have been, and are being, developed. There are also observational data about which the the-ories can potentially make predictions. To enable semantic interoperation between the data and the theories, we need ontologies to define the vocabulary used in them. For example, in the domain of minerals exploration, research geologists spend careers developing mod-els of where to find particular minerals. Similarly, geological surveys publish geological descriptions of their jurisdictions as well as instances of mineral occurrences. The commu-nity is starting to develop standardized ontologies to enable consistent use of vocabulary and the semantic interoperation between the model descriptions and the instance descrip-tions. This paper describes a framework for representing these ontologies, instances and theories, and describes ontologically-mediated probabilistic matching between instances and theories. We give an example of our matcher in the geology domain, where the prob-lem is to determine what minerals can be expected at a location, or which locations may be expected to contain particular minerals. This is challenging as models and instances are built asynchronously, and they are described in terms of individuals and properties at varied levels of abstraction and detail. This paper shows, given a model, an instance, and a role assignment that specifies which individuals correspond to each other, how to construct a Bayesian network that can compute the probabilities that the instance matches the model. Key words