TL;DR
This paper introduces "concept reification" to Description Logics (DLs), allowing concepts to be treated as individuals. This bridges a critical gap between DLs and object-oriented databases, enabling more reusable and domain-independent knowledge representation by unifying terminological and assertional knowledge levels.
What Problem Does This Paper Address?
Traditional Description Logics (DLs) maintain a strict separation between the terminological level (concepts and their definitions) and the assertional level (individuals and their properties). This separation, while simplifying reasoning, limits expressiveness. For instance, you can't easily represent a situation where a concept itself is an instance of a higher-level concept (a "meta-concept"). This is crucial for building reusable knowledge models, as it would allow a generic terminology to be instantiated for various specific domains. The paper argues this limitation hinders the unification of DLs with object-oriented databases, where meta-data (like classes) are often treated as data.
Methods
The core contribution is the introduction of concept reification. This means:
- Concepts as Individuals: A concept
Ccan be treated as an individualC'. - Membership Role
∈: A new role∈is introduced.X ∈ Csignifies that individualXis an instance of conceptC. This role effectively merges the terminological and assertional levels. - First-Order Semantics: To maintain decidability and avoid the complexities of higher-order logic, the authors propose a modified first-order semantics. Concept-valued variables are interpreted over "intensions" (reified individuals) rather than all possible concepts.
- Tableaux-like Calculus: A sound and complete inference algorithm is presented, based on a tableaux-like calculus operating on constraint systems. This algorithm extends existing DL reasoning techniques to handle the new reification construct.
Datasets
Not discussed in the paper.
Key Insights and Ideas
- Unifying Knowledge Levels: Concept reification is presented as the key to bridging the gap between DLs and object-oriented databases by allowing concepts to be treated as data.
- Reusability: The ability to reify concepts is essential for creating domain-independent and reusable knowledge models. For example, a generic "position" concept could be reified and then instantiated for specific roles like "manager" or "secretary."
- Expressiveness Boost: Reification allows DLs to express constructs like "the set of all concepts that an individual is an instance of" (e.g.,
∃∈.{X}). - Decidability Maintained: By carefully defining a first-order semantics for reification, the authors ensure that key inference services remain decidable, a critical aspect for practical DL applications.
Key Results
The paper proposes a formalization of concept reification within the ALCOE Description Logic. It provides a modified semantics and a sound and complete inference algorithm for checking the consistency of knowledge bases that utilize this new feature. The authors demonstrate how this extension makes DLs more expressive and suitable for applications requiring meta-reasoning and knowledge reuse, particularly in the context of object-oriented databases.
Limitations and Future Work
- Instability with Role Inverses: The paper notes that the presence of role inverses (like the inverse of
∈) can lead to "instability" in the reasoning process, meaning new constraints might be discovered indefinitely. This issue is acknowledged as orthogonal to reification itself and is not fully tackled in this paper. - General Inclusions: The authors restrict their discussion to concept names and avoid general concept inclusions or equations to prevent potential paradoxes and maintain decidability.
- Future Work: The authors suggest that the presented algorithms can be extended to more expressive DL languages. The core idea of reification is presented as a fundamental step towards a unified framework for DLs and object-oriented databases.
Summary updated