Modular Reuse of Ontologies: Theory and Practice

Bernardo Cuenca Grau, Ian Horrocks, Yevgeny Kazakov, Ulrike Sattler

Introduction

In the Semantic Web, ontologies serve as the foundational blueprints of knowledge. However, prominent ontologies like the SNOMED medical terminology are gargantuan, often containing hundreds of thousands of concepts. When researchers attempt to reuse specific terms—such as "Cystic Fibrosis"—from these massive sources, the standard practice is to import the entire ontology. This approach introduces significant computational overhead and the risk of "logical side effects," where new axioms in a local ontology unintentionally alter the meaning of terms in the imported source.

The Problem: Logical Side Effects

The authors formalize this challenge through the lens of Conservative Extensions. The goal is to ensure that combining a new ontology with an existing one does not change the logical implications of the original terms. Without formal safeguards, adding local axioms can create "leaks" that imply unintended facts in the source domain.

Example ontologies for medical research projects and medical terms, illustrating how axioms interact across different domains.

As shown in the example above, an ontology of medical research projects ($\mathcal{P}$) might reuse terms from an ontology of medical terms ($\mathcal{Q}$). If the definitions in $\mathcal{P}$ are not carefully constrained, they could inadvertently lead to new inferences within the medical domain of $\mathcal{Q}$, violating the principle of modularity.

Methods: From Undecidability to Locality

The paper first establishes a "No Free Lunch" theorem: for the complex Description Logics (DLs) underlying OWL DL, determining if an extension is "safe" or extracting a minimal module is mathematically undecidable.

A summary of Description Logics (DLs), their constructors, and axiom types, representing the logical framework addressed by the authors.

To bypass this theoretical barrier, the authors introduce the concept of Locality. Instead of exhaustive logical checking, they propose a syntactic approximation. An axiom is considered "local" if it becomes a tautology when terms outside the target signature are ignored.

  • Semantic Locality: A model-theoretic approach ensuring any model of the reused part can be extended to a model of the whole.
  • Syntactic Locality: A grammar-based shortcut that can be computed in polynomial time, making it practical for massive ontologies.

The researchers developed a fixpoint algorithm that starts with a set of target symbols and iteratively extracts axioms until all remaining axioms in the source are local, ensuring a logically sound module.

Evaluation and Datasets

The algorithm was tested against a diverse suite of real-world ontologies, categorized by their complexity and size:

  • Large-Scale Taxonomies: NCI (National Cancer Institute), SNOMED, and the Gene Ontology (GO).
  • Complex Logical Structures: GALEN (Full and Small), SWEET (NASA’s Earth/Environment terminology), and DOLCE-Lite.

Key Results

The locality-based approach demonstrated significant practical advantages over previous heuristic-based tools:

  • Drastic Size Reduction: For ontologies like NCI and SNOMED, the extracted modules were frequently less than 1% of the original size.
  • Logical Soundness: Unlike tools like Prompt-Factor, which often lacked a formal stopping condition and pulled in excessive data, the locality algorithm consistently produced minimal, logically sound fragments.
  • The GALEN Challenge: The study revealed that complex ontologies like GALEN-Full contain "cycles of dependencies." This means modules are either very small or very large (nearly 30% of the total), with little middle ground—a structural insight previously hidden from ontology engineers.

Limitations and Future Work

While highly effective, locality is a sufficient but not necessary condition. This means the algorithm might occasionally include an axiom that isn't strictly required, resulting in a module slightly larger than the theoretical minimum. Additionally, the presence of "nominals" (enumerated individuals) in OWL DL presents challenges for current locality definitions, suggesting a need for further refinement. Future work may also explore using these modules to optimize standard reasoning tasks by allowing reasoners to process small, independent chunks of knowledge rather than monolithic structures.

Summary updated

Abstract

In this paper, we propose a set of tasks that are relevant for the modular reuse of ontologies. In order to formalize these tasks as reasoning problems, we introduce the notions of conservative extension, safety and module for a very general class of logic-based ontology languages. We investigate the general properties of and relationships between these notions and study the relationships between the relevant reasoning problems we have previously identified. To study the computability of these problems, we consider, in particular, Description Logics (DLs), which provide the formal underpinning of the W3C Web Ontology Language (OWL), and show that all the problems we consider are undecidable or algorithmically unsolvable for the description logic underlying OWL DL. In order to achieve a practical solution, we identify conditions sufficient for an ontology to reuse a set of symbols ``safely''---that is, without changing their meaning. We provide the notion of a safety class, which characterizes any sufficient condition for safety, and identify a family of safety classes--called locality---which enjoys a collection of desirable properties. We use the notion of a safety class to extract modules from ontologies, and we provide various modularization algorithms that are appropriate to the properties of the particular safety class in use. Finally, we show practical benefits of our safety checking and module extraction algorithms.
BibTeX
@article{grau2008modular,
  author = {Bernardo Cuenca Grau and Ian Horrocks and Yevgeny Kazakov and Ulrike Sattler},
  title = {{Modular Reuse of Ontologies: Theory and Practice}},
  journal = {J. Artif. Intell. Res.},
  year = {2008},
  doi = {10.1613/jair.2375},
}
Figure