Uniform Interpolation and Forgetting for ALC Ontologies with ABoxes

Patrick Koopmann, Renate A. SchmidtAAAI 2015 · DOI

Introduction

In Knowledge Representation, the ability to selectively remove information from an ontology while preserving logical entailments—a process known as uniform interpolation or "forgetting"—is essential for privacy, ontology reuse, and versioning. While techniques for forgetting in TBoxes (terminological axioms) are established, extending these methods to ABoxes (assertional facts) introduces significant complexity. This paper presents the first practical method for uniform interpolation in $\mathcal{ALC}$ ontologies containing both TBoxes and ABoxes.

Methodology

The proposed approach utilizes a resolution-based algorithm designed to handle the interaction between general concepts and specific individuals. The process begins with normalization, where complex expressions are simplified using "definers."

Central to the method is a specialized resolution calculus that incorporates unification to match variables with constants across the ABox. This allows for the propagation of logical constraints. To maintain completeness, the algorithm initially operates within an extended logic, $\mathcal{ALC}\nu$, which supports greatest fixpoints and disjunctive ABoxes.

Resolution calculus rules for ABox forgetting

Finally, a back-approximation step transforms the results into standard $\mathcal{ALC}$ or $\mathcal{ALCO}$ (adding nominals), ensuring the output remains usable within standard Description Logic frameworks.

Experimental Evaluation

The authors evaluated a prototype implementation using "ABox-heavy" ontologies from the NCBO BioPortal and the Oxford Ontology Library. These datasets provided a rigorous test for the algorithm's ability to manage large sets of factual assertions alongside terminological rules.

Results and Discussion

Empirical results demonstrate that the approach is remarkably efficient. Despite the theoretical triple-exponential complexity, most interpolation tasks were completed in under 15 seconds, as shown in the performance metrics for forgetting 10% of the signature.

Experimental results for forgetting 10% of symbols

Key findings include:

  • Representability: Although theory suggests nominals might be necessary, real-world ontologies were often representable in pure $\mathcal{ALC}$ after forgetting.
  • Scalability: The algorithm maintained manageable ABox sizes even when forgetting a significant portion of the signature. When forgetting 90%, ABoxes occasionally grew to make "hidden" relations explicit, but remained within practical limits.
  • Practicality: The use of helper concepts (definers) effectively "simulates" complex fixpoint logic, keeping the output human-readable and compatible with standard reasoners.

Conclusion

This work bridges a gap in ontology engineering by providing a practical framework for ABox forgetting. Future research will focus on extending the method to more expressive logics (e.g., those including role hierarchies or number restrictions) and optimizing the output to reduce logical redundancy and improve readability for massive datasets.

Summary updated

BibTeX
@inproceedings{koopmann2015uniform,
  author = {Patrick Koopmann and Renate A. Schmidt},
  title = {{Uniform Interpolation and Forgetting for ALC Ontologies with ABoxes}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2015},
  doi = {10.1609/aaai.v29i1.9206},
}
Figure