Knowledge Reuse and Knowledge Validation

Automatic verification tools such as COVER have proven to be valuable aids in the validation process for knowledge-based systems (KBS). COVER checks KBS for logical anomalies. Background domain knowledge can allow COVER to detect errors in knowledge bases that would otherwise go undetected. Ontologies are a necessary component of knowledge sharing: two KBS cannot share knowledge unless they commit to a common ontology. Ontologies also provide a rich source of background domain knowledge for validation. This paper describes a tool, DISCOVER, which verifies KBS against ontologies. DISCOVER verifies heterogeneous sources of knowledge: KBS are represented in CRL (COVER rule language), and ontologies are represented in MOVES (Meta-ontology for the verification of expert systems). The paper describes MOVES and CRL, and discusses a number of anomalies arising between KBS and ontologies. It is shown that DISCOVER can be used to verify that a KBS commits to a given ontology, which is a prerequisite for sharing its knowledge. Introduction and Motivation The reuse and sharing of knowledge bases is a central theme of knowledge engineering in the 1990s (Neches et al. 1991). Whereas, in the 1980s, organisations focussed upon the construction of standalone knowledgebased systems, a significant amount of current interest lies in integrating existing knowledge bases together into enterprise-wide resources. Such resources play a vital role in modern the evolution of organisations, relating to the ideas of enterprise modelling and business process reengineering (Jennings et al. 1996). Knowledge Reuse There are two primary ways in which organisations seek to reuse and integrate existing knowledge bases: * Knowledge fusion: Incorporation of existing knowledge into a new knowledge base, or merging of existing knowledge bases into a combined resource. Data warehousing is an example of this kind of approach (Wiederhold 1992). ̄ Distributed knowledge-based systems: Interoperation of existing knowledge-based systems (or "agents"), distributed as nodes on a network. An example of this approach is the European ARCHON architecture (Cockburn & Jennings 1995). Enabling Technology: Ontologies Early work on enabling technology for knowledge sharing established that three components are needed to allow knowledge to be shared between two knowledge bases (Neches et al. 1991): ̄ a common protocol in which to communicate knowledge; ̄ a common language in which to express knowledge; ̄ a common set of definitions of terminology -an ontology. A great deal of work has been done to define common protocols and languages for the communication and expression of knowledge, the best-known being the KQML protocol (Finin et al. 1994) and the KIF language (Genesereth & Fikes 1992) produced by the Knowledge Sharing Effort (KSE) project (Neches et al. 1991). In many ways, the definition of ontologies is a mote difficult problem, because there are many different domains in which terminology must be defined. These include: ̄ domain terminology for the application domain(s) that the knowledge refers to, for example, medicine, aerospace, or commerce; ̄ task terminology for the operational aspects of the knowledge-based systems, for example, diagnosis, scheduling, or design; ̄ physical ter}ninology describing the nature of reality underpinning the knowledge, including time, space, and part-whole relations. Approaches to building ontologies range from largescale work in defining highly-reusable ontologies of "commonsense" knowledge (Lenat & Guha 1990) more modest efforts in defining terminology in a specific application area (Uschold & Gruninger 1996). From: AAAI Technical Report WS-97-01. Compilation copyright © 1997, AAAI (www.aaai.org). All rights reserved. Although precise definitions of an ontology differ, the most widely held view is that an ontology an in explicit specification of a conceptualisation: "the objects, concepts and entities that are asumed to exist in some area of interest, and the relationships that hold among them" (Gruber 1995). Ontologies may be expressed using informal or semi-formal specification languages, but for our purposes we are interested only in ontologies defined formally in an appropriate knowledge representation language (to permit their manipulation within knowledge-based systems). As a minimum, an ontology will define taxonomic relationships (informal example: "student is a person"); more generally, any constraints may be put on terms (informal example: "all students must take at least one course"). It is worth noting that, although the purpose of an ontology is to define terminology, the form of an ontology is that of a knowledge base or database conceptual schema; any knowledge representation language or database schema definition language may be used to define an ontology. Knowledge Reuse using Ontologies There are two requirements to share knowledge between two knowledge bases: ̄ it must be possible to translate their knowledge representations into a common language; * it must be possible to map their terminologies into a common ontology. The first requirement is accomplished using a set of translation rules; the second is accomplished using a set of mapping rules. Note that there does not have to be a single common language and a single common ontology; however, if multiple common languages and ontologies exist, there will need to be multiple sets of translation and mapping rules. The translation problem is not hard if knowledge bases use a syntacticallysugared version of first-order predicate calculus, which is the approach taken by the KSE project (Gruber 1995), and is assumed for the purposes of this paper. A set of mapping rules between a knowledge base and an ontology defines an ontological commitment of the knowledge base. It is highly desirable that this ontological commitment be consistent: no constraint in the ontology should be in conflict with inferences derivable from the knowledge base, and vice versa. Checking that an ontological commitment is consistent is a validation issue. It is worth noting that there is no completeness requirement on ontological committment: it is not necessary for every term in the knowledge base to have an equivalent term in the ontology, but in that case there will be some unsharable statements. Similarly, there is no need to have an equivalent knowledge base term for every term in the ontology. Verifying Ontological Commitment To support sharing and reuse of knowledge, we want to provide help in verifying the commitment of a knowledge base to an ontology. Furthermore, we want to reuse existing knowledge base verification tools for this task. This paper reports on an initial investigation of the use of the COVER knowledge base verification tool (Preece, Shinghal, & Batarekh 1992) for verifying ontological commitment. COVER is an anomaly checker: it analyses a knowledge base for undesirable properties including conflicting knowledge, redundant knowledge, and deficient knowledge. This paper describes how an ontology containing university terms, ’The University Ontology’ can be used to verify a knowledge base, ’The University Knowledge Base’, that commits to it. These examples are inspired by the knowledge base from (Zlatareva & Preece 1994). The paper: ̄ introduces a simple ontology description language called MOVES (based on a subset of CycL (Lenat Guha 1990), but with a Prolog flavour); ̄ shows how an ontological commitment is defined using mapping rules; ̄ shows how, with extensions, COVER can be used not only to check the ontological commitment, but also to do the terminological translations between ontology and knowledge base (and, hence, to verify the mapping rules also!); ̄ examines what anomalies like conflict, redundancy, and deficiency mean when they involve ontological commitments. The extended version of COVER (which does not change any of COVER’s original functionality) is called DISCOVER (COVER for DIStributed knowledge bases). It is worth observing that DISCOVER can do more than verifying ontological commitment: it can use an ontology as a body of background knowledge against which the original knowledge base can be validated. It can also employ knowledge from other sources to validate the knowledge base; for example, it can use items in a database as test cases, provided that the database commits to a common ontology with the knowledge base.