Supporting ontological analysis of taxonomic relationships

https://doi.org/10.1016/S0169-023X(01)00030-1Get rights and content

Abstract

Taxonomies are an important part of conceptual modeling. They provide substantial structural information, and are typically the key elements in integration efforts, however there has been little guidance as to what makes a proper taxonomy. We have adopted several notions from the philosophical practice of formal ontology, and adapted them for use in information systems. These tools, identity, essence, unity, and dependence, provide a solid logical framework within which the properties that form a taxonomy can be analyzed. This analysis helps make intended meaning more explicit, improving human understanding and reducing the cost of integration.

Introduction

Ontology is a discipline of Philosophy that deals with what is, with the kinds and structures of objects, properties, and other aspects of reality. While much of the philosophical practice of ontology dates back to Aristotle and what his students called “metaphysics,” the term ontology (ontologia) was coined in 1613 by Rudolf Gockel and apparently independently by Jacob Lorhard. According to the OED, the first recorded use in English was in 1721. Today's ontology includes questions such as, “what is a castle?” and “what is a hole?” The way we answer these questions reflects the way we perceive and interact with the world.

By the early 1980s, researchers in AI and especially in knowledge representation had realized that work in ontology was relevant to the necessary process of describing the world of intelligent systems to reason about and act in [25]. This awareness and integration grew and spread to other areas until, in the latter half of the final decade of the 20th century, the term “ontology” actually became a buzzword, as enterprise modeling, e-commerce, emerging XML meta-data standards, and knowledge management, among others, reached the top of many businesses strategic plans. In addition, an emphasis on “knowledge sharing” and interchange has made ontology an application area in its own right.

In general, the accepted industrial meaning of “ontology” makes it synonymous with “conceptual model”, and is nearly independent of its philosophical antecedents. We make a slight differentiation between these two terms, however (as shown later in Fig. 1): a conceptual model is an actual implementation of an ontology that has to satisfy the engineering trade-offs of a running application, while the design of an ontology is independent of run-time considerations, and its only goal is to specify the conceptualization of the world underlying such an application. In this paper we describe a well-founded methodology for ontological analysis (called OntoClean) that is strongly based on philosophical underpinnings, and a description-logic-based system that can be used to support this methodology. Although the methodology is not limited to analyzing taxonomies, we focus on that aspect of it here.

Most of the work described here have been published previously in a preliminary form, as will be noted in specific sections. This paper is an extended version of [13] that presents a broader view of the overall methodology and an extended discussion of a system to support it. We note that some valid criticisms of the formalizations we have presented previously have been raised in [19] and [6]. While the basic intuitions underlying these notions remain the same, some of the formalizations that follow have been updated in light of these criticisms. Further work and discussion are needed to fully address these changes and their implications, but it is out of the scope of this paper, whose main purpose is to provide an overview of the methodology we have developed for conceptual modeling.

Section snippets

Background

The notions upon which our methodology is based are subtle, so before describing them in more detail we discuss the basic intuitions behind them and how they are related to some existing notions in conceptual modeling.

The formal tools of ontological analysis

In this section we shall present a formal analysis of the basic notions discussed above, and we shall introduce a set of meta-properties that represent the behavior of a property with respect to these notions. Our goal is to show how these meta-properties impose some constraints on the way subsumption is used to model a domain, and to present a description logic system for checking these constraints.

Methodology

The specific goal of this methodology is to make modeling assumptions clear. One of the most important ways the methodology is used is in analyzing taxonomies to form well-founded taxonomies, which are discussed further in Section 6.

The methodology is made up of a number of formal analysis tools that can be grouped into four distinct layers, such that the notions and techniques within each layer are based on the notions and techniques in the layers below. In Fig. 1, the methodology is depicted

Knowledge-based support

The methodology based on these techniques requires that the assignment of meta-properties to properties in an ontology be performed by hand. This analysis in all cases requires that the modeler be very clear about what each property means. We have developed a support system that can help modelers with this analysis, as it can verify the consistency of a taxonomy based on the constraints described in Section 3.6. A modeler enters information about properties to be used in a conceptual model, and

Example

In this section we provide a brief example of the way our analysis can be used. A complete version of this example is available [14].

We begin with a set of properties arranged in a taxonomy, as shown in Fig. 5. In Table 3 we provide some basic explanations of the intended meaning of these properties. The taxonomy we have chosen makes intuitive sense prima facie, and in most cases the taxonomic pairs were taken from existing ontologies such as Wordnet [24], Pangloss [20], and CYC [21]. See [10]

Conclusion

We have discussed several notions of formal ontology used for ontological analysis in Philosophy: identity, unity, essence, and dependence. We have formalized these notions in a way that makes them useful for conceptual modeling, and introduced a methodology for ontological analysis founded on these formalizations.

Our methodology is supported by a system that helps the conceptual modeler study the deep ontological issues surrounding the representation of properties in a conceptual model, and we

Acknowledgements

This work was supported in part by the Eureka Project (E! 2235) IKF, the Italian National Project TICCA (Tecnologie cognitive per l'interazione e la cooperazione con agenti artificiali), and a Research Committee Grant from Vassar College. We would like to thank Claudio Masolo, Milena Stefanova, Pierdaniele Giaretta, Alessandro Oltramari and Bill Andersen for their useful comments.

Chris Welty is an Associate Professor at Vassar College, and has consulted in the real world at large and small companies including GE, AT&T, and IBM. He holds a Ph.D. in computer science from Rensselaer Polytechnic Institute. He is editor in chief of intelligence Magazine, steering committee chair of the Automated Software Engineering Conferences, an ACM Distinguished Lecturer, and the program chair of the 2001 conference on Formal Ontology in Information Systems (FOIS-2001). His research

References (32)

  • R.C. Goldstein et al.

    Data abstractions: why and how?

    Data and Knowledge Engineering

    (1999)
  • N. Guarino et al.

    An ontology of meta-level categories

  • G. Hirst

    Existence assumptions in knowledge representation

    Artificial Intelligence

    (1991)
  • V. Storey et al.

    An ontology-based expert system for database design

    Data and Knowledge Engineering

    (1998)
  • S. Bergamaschi et al.

    On taxonomic reasoning in conceptual design

    ACM Transactions on Database Systems

    (1992)
  • R. Brachman

    What IS-A is and isn't: an analysis of taxonomic links in semantic networks

    IEEE Computer

    (1983)
  • R.J. Brachman et al.

    Living with CLASSIC: when and how to use a KL-ONE-like language

  • D. Calvanese et al.

    Description logics for conceptual data modeling

  • M. Cararra et al.

    Identity criteria and sortal concepts

  • A. Gangemi et al.

    Understanding top-level ontological distinctions

    Proceedings of the 2001 IJCAI Workshop on Ontologies and Information Sharing

    (2001)
  • N. Guarino

    Formal ontology in information systems

  • N. Guarino

    The role of identity conditions in ontology design

    Proceedings of the IJCAI-99 Workshop on Ontologies and Problem-Solving Methods: Lessons Learned and Future Trends. Stockholm, Sweden, IJCAI, Inc

    (1999)
  • N. Guarino et al.

    Identity, unity, and individuality: towards a formal toolkit for ontological analysis

    Proceedings of ECAI-2000: The European Conference on Artificial Intelligence

    (2000)
  • N. Guarino et al.

    A formal ontology of properties

  • N. Guarino et al.

    Ontological analysis of taxonomic relations

  • N. Guarino et al.

    Ontology-driven conceptual analysis

    AAAI-2000 Tutorial Presentation. Austin, Texas

    (2000)
  • Cited by (256)

    • Sigmoid similarity - a new feature-based similarity measure

      2019, Information Sciences
      Citation Excerpt :

      This section provides a background on the main notions used in this work. A conceptual hierarchy provides a taxonomy (a tree or a lattice) of concepts organised using the partial order IS-A relation, which specialises more general classes into more specific classes [3,33]. The IS-A relation is asymmetric and transitive and defines a hierarchical structure of the ontology, enabling the inheritance of characteristics from parent classes to descendant classes.

    • Concept detection using text exemplars aligned with a specialized ontology

      2019, Data and Knowledge Engineering
      Citation Excerpt :

      We anticipate that this method may be useful in other domains. The concepts of pain medicine were organized into an ontology of classes connected with is-a relationships that form a backbone taxonomy [6,7] with a typical acyclic graph structure composed of various length parent–child paths. Various horizontal inter-class linkages were built into the ontology and were annotated, but were not used in this analysis.

    View all citing articles on Scopus

    Chris Welty is an Associate Professor at Vassar College, and has consulted in the real world at large and small companies including GE, AT&T, and IBM. He holds a Ph.D. in computer science from Rensselaer Polytechnic Institute. He is editor in chief of intelligence Magazine, steering committee chair of the Automated Software Engineering Conferences, an ACM Distinguished Lecturer, and the program chair of the 2001 conference on Formal Ontology in Information Systems (FOIS-2001). His research interests include ontology and ontological analysis, ontologies for information, for digital libraries, and for software understanding, and in general in improving information retrieval by representing knowledge.

    Nicola Guarino is a senior research scientist at the Institute for System Theory and Biomedical Engineering of the Italian National Research Council (LADSEB-CNR). For about ten years now he has been actively promoting the study of the ontological foundations of knowledge representation and knowledge engineering with an interdisciplinary approach centered on logic, philosophy, and linguistics. He was chairman of the First International Conference on Formal Ontology in Information Systems (FOIS'98), is associate editor of the International Journal of Human and Computer Studies, has edited 3 journal special issues on ontology-related topics, and has published more than 30 papers in international journals, books and conferences. His research activities regard ontology design, conceptual modelling, knowledge sharing and integration, logical modeling of physical objects, and ontology-driven information retrieval.

    View full text