TL;DR
This paper extends the E-Connections framework for combining Description Logics (DLs) to allow for more flexible definition and use of "link properties" that connect different ontologies. This enhanced flexibility enables more expressive modeling of complex relationships, such as transitive relations across ontologies, and provides a tableau-based decision procedure for these extended languages, which are foundational to the Web Ontology Language (OWL).
What Problem Does This Paper Address?
The Semantic Web relies on combining multiple ontologies to represent knowledge in a modular and distributed way. E-Connections offer a robust framework for this, ensuring decidability when merging different logical systems, including DLs. However, the original E-Connections framework imposed strict limitations on how "link properties" (which connect individuals across different ontologies) could be defined and used. Specifically, a link property couldn't be defined in multiple components simultaneously, nor could it point to multiple different target domains within a single definition. This restricted the expressivity and flexibility needed for real-world modeling scenarios, such as representing complex part-whole relationships or transitive relations that span across different knowledge bases.
Methods
The authors introduce an extension to E-Connections for DLs, specifically focusing on languages like SHIQ, SHOQ, and SHIO, which are the basis for OWL. The core innovation lies in allowing link properties to be explicitly annotated with their target domain. For instance, a property loves could be specified as loves(1).Person or loves(2).Pet, indicating that in the context of component 1, it connects to Person, and in component 2, it connects to Pet. This explicit domain specification allows a single property name to be used with different interpretations across components, enhancing modularity and reducing property proliferation.
Furthermore, the paper introduces a mechanism for defining transitive relations across domains. This is achieved through a Trans axiom that specifies which pairs of components a property should be considered transitive over. This allows for fine-grained control over entailments, enabling the simulation of transitive closure without the computational overhead typically associated with it in DLs.
Finally, the paper presents a tableau-based decision procedure for these extended E-Connection languages. This algorithm is designed to check the satisfiability of concepts within these combined knowledge bases, ensuring that the extended framework remains decidable.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Explicit Domain Specification for Link Properties: The ability to specify the target domain of a link property (e.g.,
loves(1).Person) is a crucial step towards more flexible ontology integration. It allows a single property name to represent different relationships in different contexts, mirroring real-world modeling needs. - Controlled Transitivity Across Ontologies: The introduction of
Transaxioms provides a powerful way to manage transitive relations that span across multiple ontologies. This allows for precise control over inferential capabilities, preventing unwanted entailments while still enabling useful reasoning. - Simulating Transitive Closure: The proposed mechanism for handling transitivity effectively simulates the expressivity of the transitive closure operator in many cases, but without the significant computational cost and complexity that direct inclusion of such an operator would entail.
- Foundation for OWL: By extending E-Connections for DLs that underpin OWL, this work directly contributes to more expressive and flexible ways of building and reasoning with OWL ontologies.
Key Results
The paper presents a formal extension to E-Connections for DLs, providing syntax, semantics, and a tableau-based decision procedure. The authors demonstrate how their extensions allow for more flexible definition of link properties and controlled transitivity across ontologies. They also show how these extensions can simulate transitive closure in certain scenarios. The decision procedure is shown to run in 2NExpTime, which is comparable to the complexity of reasoning in the underlying DLs themselves.
Limitations and Future Work
The authors acknowledge that their extension might not fully support all cases when component logics involve inverses and number restrictions on links and nominals, as this can break the separation between domains.
Future work includes exploring the generalization of their extension to Abstract Description Systems, which encompass a broader range of logical formalisms beyond DLs. They also plan to implement these decision procedures as an extension to the Pellet DL reasoner.
Summary updated