Automated Support for Composition of Transformational Components in Knowledge Engineering

The knowledge engineering world provides a rich source of software components for transforming formally ex- pressed knowledge on a large scale, such as induction systems, knowledge base refiners and ontology merging tools. Although most of these systems have been designed as stand-alone components, there is interest in making them ac- cessible on the Web, with the ultimate goal in mind that a knowledge engineer should be able, with a small amount of intellectual effort, to locate and assemble sequences of these components to perform complex transformations on large repositories of knowledge. The sorts of transformations used in knowledge engineering are not always trustwor- thy: some may not preserve the semantics of the knowledge transformed; some may not be able to perform a given transformation reliably under all circumstances. Therefore, it is crucial to have ways of inspecting the key properties we expect to be preserved by each transformational component and of describing how these properties change as new transformations are applied. We present initial experiments on a large-scale knowledge engineering problem and show how an abstract char- acterisation of knowledge-transformation steps, accompanied by a customisable editor, can allow a high degree of automation in this task. With such an editor we can analyse and represent sequences of general transformation steps and check if properties such as subsumption, completeness and soundness are preserved during different stages of the transformation, by analysing the structure of these sequences.