Using Description Logics for RDF Constraint Checking and Closed-World Recognition

Peter F. Patel-SchneiderAAAI 2015 · DOI

Introduction

The Semantic Web's formal foundation, Description Logics (DL), typically operates under the Open-World Assumption (OWA). In an open world, "absence of evidence is not evidence of absence"—if a database lacks a specific record, it is assumed to be unknown rather than non-existent. While this facilitates decentralized data integration, it complicates data validation. For instance, a requirement that "every person must have exactly one name" cannot be enforced as a constraint in OWA; instead, the system simply infers the existence of a name if one is missing.

This mismatch led to the creation of separate validation languages like Shape Expressions (ShEx) and SPIN. In this paper, Peter Patel-Schneider argues that a new language is not required. Instead, by adopting a "closed-world" interpretation of standard DL axioms, the existing DL framework can be repurposed for robust RDF constraint checking and closed-world recognition.

Methodology: Shifting to Closed-World Semantics

The author proposes a framework that maintains DL syntax but modifies its semantics to support validation:

  1. Canonical Interpretations: Any RDF graph is transformed into a single, specific DL model where the only individuals, properties, and types that exist are those explicitly stated in the graph. This effectively "closes" the world.
  2. Constraint Checking: In this canonical model, if a triple is missing, it is false. This allows standard DL constructs, such as cardinality restrictions (e.g., = 1 foaf:name), to act as triggers for validation errors rather than triggers for inference.
  3. Unique Name Assumption (UNA): Unlike standard OWL, this approach assumes that different IRIs always denote different individuals. This is essential for accurate counting and cardinality checks.
  4. Handling Recursion: For complex, recursive definitions (e.g., "a student whose friends are all students"), the paper utilizes maximal extensions. This ensures the system identifies the largest possible set of individuals satisfying the recursive criteria, providing a clear mathematical path for handling recursive shapes.

Illustrative Example

The paper demonstrates these concepts using a constructed university domain involving students, faculty, and enrollments. Figure 1 illustrates the sample RDF data graph.

RDF Example Graph

To validate this data, the author applies DL axioms that serve as integrity constraints. As shown in Figure 2, these axioms define rules such as disjointness between persons and organizations, and specific cardinality requirements for names and enrollments.

Description Logic Constraints

By applying these axioms under closed-world semantics, the system can identify inconsistencies that standard RDFS reasoning would miss. For example, if an individual of type exo:UniStudent is not linked to at least one exo:Uni via the exo:enrolled property, the constraint defined in axiom 3 of Figure 2 would be violated, signaling invalid data.

Key Insights and Results

The research highlights that the perceived "failure" of OWL for validation is a matter of semantic choice, not syntactic limitation. The same axiom (Person ⊑ =1 name) can mean "infer a name" in an open world or "check for a name" in a closed world.

Furthermore, because RDF and RDFS are relatively inexpressive compared to full First-Order Logic, many theoretical hurdles of closed-world reasoning—such as undecidability or extreme computational complexity—are mitigated. The paper shows that checking these axioms against a graph is polynomial in time, provided no new vocabulary is introduced. Additionally, the "maximal extension" approach provides a robust solution for recursive shapes, a historically contentious topic in RDF validation.

Conclusion and Future Work

This work demonstrates that Description Logics can provide a unifying framework for both data recognition and constraint checking. While effective for most scenarios, the paper acknowledges limitations regarding "non-monotone" recursion, where adding data might invalidate previously valid nodes. Future work will explore "hybrid" reasoning—integrating open-world discovery with closed-world integrity—and optimizing the translation of complex DL constraints into efficient SPARQL queries for practical implementation in existing database systems.

Summary updated

BibTeX
@inproceedings{patelschneider2015using,
  author = {Peter F. Patel-Schneider},
  title = {{Using Description Logics for RDF Constraint Checking and Closed-World Recognition}},
  booktitle = {AAAI Conference on Artificial Intelligence},
  year = {2015},
  doi = {10.1609/aaai.v29i1.9177},
}
Figure