Ontology and reuse in model synthesis

• Main Author: Biris Brilhante, Virginia
• Published: University of Edinburgh 2003
• Subjects: 306.440285
• Online Access: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.726396

A much pursued ontological capability is knowledge reuse and sharing. One aspect of this is reuse of ontologies across different applications, including construction of other ontologies. Another aspect is the reuse of knowledge that is founded on ontologies, which, in principle, is attainable in that an ontology renders well-defined and accessible the meaning of concepts that underly the knowledge. In this thesis we address both these aspects of reuse, grounding the study in the problem of synthesis of structural ecological models. One basis for the design of ecological models, along with human expertise, is data properties. This motivated the construction of a formal ontology - Ecolingua - where we define concepts of the ecological data domain. The diversity of the domain led us to use the Ontolingua Server as an ontology design tool, since it makes available an extensive library of shareable ontologies. We reused definitions from several of these ontologies to design Ecolingua. However, Ecolingua's specification in the representation language of the Ontolingua Server did not easily translate into a manageable, executable specification in our choice of implementation language, which required us to develop our own complementary tools for scope reduction and translation. Still, a great deal of manual re-specification was necessary to achieve an operational, useful ontology. In sum, reuse of multiple ontologies in ontology construction employing state-of-the-art tools proved impractical. We analyse the practical problems in reuse of heterogeneous ontologies for the purpose of developing large, combined ones. In its operational form, Ecolingua is used to describe ecological data to give what we call metadata, or, said differently, instantiated ontological concepts. Synthesis is achieved by connecting metadata to model structure. Two approaches were developed to do this. The first, realised in the Synthesis-0 system, uses metadata alone as synthesis resource, while the second, realised in the Synthesis-R system, exploits Ecolingua to reuse existing models as an additional synthesis resource. Both Synthesis-0 and Synthesis-R are working systems implemented as logic programs. The systems were empirically evaluated and compared on run time efficiency criteria. Results show a remarkable efficiency gain in SynthesisR, the system with the reuse feature. We generalise the results to systems for synthesis of structural models informed by domain data. Once a data ontology and ontology-constrained synthesis mechanisms are in place, existing models can efficiently be reused to produce new models for new data.