Context-explication in conceptual ontologies: the PLIB approach

A number of computer science problems, including heterogeneous database integration, natural language processing, document intelligent retrieval would benefit from the capability to model the absolute meaning of things, independently of any particular use of these things. Such models, termed ontologies, have been heavily investigated over the last ten years, with various purposes and within various contexts. The goal of this paper is to investigate the role of ontologies for data integration and to present an ontology model that was precisely developed to allow neutral exchange and automatic integration of technical data. We first propose a taxonomy of ontologies into linguistic ontologies, based on words and usable for intelligent document processing, and concept ontologies, multilingual and usable with structured data. We then discuss differences between ontologies and usual conceptual models. We claim that the main difference is context-sensitivity, and we identify four requirements for making ontologies less contextual than models and suitable for data integration. Finally we present how these requirements have been fulfilled in the PLIB ontology model developed to give meaning to technical data, and we outline the use of PLIB-based ontologies in various domains including database integration, e-engineering and the semantic Web. relevant for a topics defined by a set of words, even when the same words are not used, and • in a second step, to retrieve which information sources, either unstructured, semi-structured or, structured provide answer elements to a user query. Both kinds of information integration requiring explicit representation of meaning, these last ten years a lot of research has been done to develop ontology models intended to capture the a priori nature of reality, as independently as possible from any particular use of this reality. Once defined, such representations may then be used to reconcile various information sources at the meaning level. The word ontology is now extensively used in a number of computer science domains : knowledge management, natural language processing, database, object oriented modeling, etc. If there seems to be some consensus on what an ontology structure should be – categories (classes), properties, logical relationships – the focus of the various approaches is so different that the same word seems to represent quite different realities, and that ontologies developed, e.g., for natural language processing seems to be nearly useless for e.g., database integration, and conversely. The goal of this paper is to investigate the concept of an ontology in a structured-data perspective. It is also to show how the ontology model that has been developed over the last 10 years in the PLIB standardization project (officially ISO 13584) may be used in the various domains where the meaning of structured data need to be made computerinterpretable, like multidatabase, e-engineering, B2B electronic commerce and web services over the semantic web. The initial goal of PLIB was to allow engineering database integration and neutral exchange of component libraries. The content of this paper is as follows. In the next section we discuss the various kinds of ontology needed for representing semantics. We propose to distinguish between document-oriented linguistic ontology (LO) and structured-data-oriented concept ontology (CO). In the third section we investigate the difference between ontologies and models. We claim that the major difference is explication of the modeling context. We introduce four mechanisms allowing to make ontology much more generic through context explication. In the fourth section we present how these mechanisms are represented in the PLIB ontology model to allow automatic integration of several structured data sources. We discuss in section 5 how such ontologies may be used for database integration, e-engineering and the semantic web. A discussion of related works is presented in section 6. Conclusion is presented in section 7. 2 CONCEPT ONTOLOGIES VERSUS LINGUISTIC ONTOLOGIES Since the term ontology was borrowed from philosophy by John Mc Carthy in the 70’s and introduced in the computer science vocabulary, many definitions have been offered. The most commonly cited definition is one by T. Gruber "An ontology is a formal explicit specification of a shared conceptualization" (Gruber 1993). In all the ontology models, such a conceptualization consist of three parts. primitives items of the ontology, where items are either classes or properties, are those items “for which we are not able to give a complete axiomatic definition. We must rely on textual documentation and a background of knowledge shared with the reader” (Gruber 1993), defined items are those items for which the ontology provides a complete axiomatic definition by means of necessary and sufficient conditions, and logical relationships (or inference rules) provide for reasoning over ontology items, and for solving the problems for which the ontology was designed. The agreed definition and structure description leave open what we consider as the major criteria for classifying ontologies and ontology models: whether their area of interest consists of beings –what there is in the world – or of word– how beings are apprehended and reflected in a particular natural language. We call linguistic ontology (LO) those ontologies whose scope is representing the meaning of the words used in a particular Universe of Discourse (UoD) in a particular language. We call concept ontology (CO) those ontologies whose goal is representing the categories of objects and of objects properties that are in some part of the word. We claim that these two kinds of ontologies address quite different problems and should have quite different content. LO (e.g., Everett et al. 2002) are documentoriented. The typical problem they address may be termed as follows: “find all documents pertinent with respect to a query expressed as a set of words possibly connected by logical operators like AND, OR and NOT, even if these documents don’t contain these words”. Since natural language contain a number of different words for reflecting identical or similar meanings, LO are large in nature. They include a number of conservative definitions, i.e., defined items that only introduce terminology and do not add any knowledge about the world (Gruber 1993). They are language-specific and contain a number of linguistic relationships such that synonym, hypernym, hyponym, overlap, covering, disjoint to capture in a semi-formal way (Walche et al. 2001)