In knowledge representation, two fundamentally different paradigms have traditionally existed in isolation. On one side are Description Logics (DLs) and ontologies, which govern the Semantic Web. They operate under the Open-World Assumption (OWA): if a piece of information is missing, it is treated as unknown. This is ideal for web-scale data where no single source has complete knowledge.
On the other side is Logic Programming (LP) and rule-based systems. They operate under the Closed-World Assumption (CWA) and rely on Negation-as-Failure (NAF): if a fact cannot be proven true, it is assumed false. This is essential for database management, default reasoning, and handling exceptions.
Integrating these two paradigms is notoriously difficult. If you naively combine them, their underlying logics clash. For instance, if an ontology does not explicitly state whether a patient is allergic to a drug, an open-world system leaves it as unknown, while a closed-world rule might assume the absence of an allergy record means it is safe to administer the drug. Bridging this semantic mismatch requires frameworks that allow rules to reason about what an ontology does not know, without destroying the open-world properties of the ontology itself.
The Modal Interface: MKNF Knowledge Bases
One of the most successful strategies for unifying these worlds is to introduce an epistemic modal logic that acts as an interface between them. In A Faithful Integration of Description Logics with Logic Programming, Boris Motik et al. proposed using the logic of Minimal Knowledge and Negation as Failure (MKNF).
Instead of allowing rules to query the ontology directly, the MKNF framework wraps predicates in modal operators:
- $\mathbf{K}P$ means "$P$ is known to be true" (i.e., it is logically entailed by the ontology).
- $\mathbf{not}P$ means "$P$ is not known to be true" (i.e., the ontology cannot prove it).
By using these operators, a rule engine can safely inspect the knowledge state of the ontology. For example, a rule can trigger based on the explicit absence of information ($\mathbf{not}\text{Allergy}$) without forcing the ontology itself to assume the patient is definitively non-allergic. This allows for a "faithful" integration where both components retain their original semantics. To make this approach computationally viable for practical applications, Jianmin Ji et al. developed well-founded operators for hybrid MKNF knowledge bases in Well-Founded Operators for Normal Hybrid MKNF Knowledge Bases, significantly lowering the complexity of computing these combined states.
Drawing Boundaries: Circumscriptive Hybrid Reasoning
Another way to bridge the gap is to selectively "close" parts of the open world using John McCarthy's concept of circumscription. Circumscription works by minimizing the extensions of specific predicates—effectively stating that no objects satisfy a predicate unless they are explicitly forced to by the system's rules or axioms.
In DL_clog: A Hybrid System Integrating Rules and Description Logics with Circumscription and A Hybrid System DLclog Integrating Description Logics and Rules, Xiaoping Chen (and later under Xiaoping Chen) introduced $\mathcal{DL}log$. This framework allows negative predicates from an open-world ontology to appear directly in rule bodies. By applying parallel circumscription, the system can handle negation-as-failure over structured ontological concepts.
This means you can write rules like "if a city is not known to be a seaside city, apply a default inland tax rate." The circumscriptive engine minimizes the "seaside city" concept, allowing the rule to safely execute under a closed-world assumption for that specific predicate. Demonstrating the underlying unity of these approaches, Qian Yang et al. showed in Integrating Rules and Description Logics by Circumscription that major competing hybrid semantics can actually be viewed as variations of a single underlying logic, differentiated only by their specific "circumscription policies."
Embedding into Generalized Logics
Rather than building an interface layer or minimizing specific predicates, a third school of thought embeds both the ontology and the rules into a single, overarching non-monotonic formalism.
In Embedding Non-Ground Logic Programs into Autoepistemic Logic for Knowledge-Base Combination, Jos de Bruijn et al. demonstrated that non-ground logic programs can be faithfully embedded into First-Order Autoepistemic Logic (FO-AEL). Because FO-AEL naturally accommodates both classical first-order logic (the basis of ontologies) and epistemic theories (the basis of rules), it serves as a neutral mathematical ground where the interactions between open- and closed-world assertions can be formally analyzed and resolved.
To ensure these unified systems remain decidable and practical, researchers have adapted well-founded semantics to hybrid architectures. For instance, Wlodzimierz Drabent et al. in Hybrid Reasoning with Rules and Ontologies proposed extending rule bodies with external ontological constraints under a 3-valued semantics (true, false, or undefined). Similarly, Georg Gottlob et al. introduced an "equality-friendly" well-founded semantics in Equality-Friendly Well-Founded Semantics and Applications to Description Logics. Their approach allows non-monotonic negation to coexist with existential rules without requiring the strict Unique Name Assumption (the database-centric belief that different names always imply different entities), bridging the gap between database rules and web-standard ontologies like OWL.
Practical Applications and Resilience
The drive to integrate these paradigms is not merely theoretical; it is forced by real-world engineering constraints. Enterprise systems frequently require open-world reasoning for discovering new logical insights alongside closed-world constraints to ensure data integrity.
+-----------------------------------------------------------------------+
| Hybrid Knowledge Base |
| |
| +-----------------------------+ +-----------------------------+ |
| | Ontology Layer (OWA) | | Rule Engine (CWA) | |
| | "Every patient has a blood | | "If NO allergy is proven, | |
| | type, even if unrecorded" | | administer medication" | |
| +--------------+--------------+ +--------------+--------------+ |
| | | |
| +-----------------+-----------------+ |
| | |
| v |
| +-----------------------------+ |
| | Hybrid Integration Layer | |
| | (MKNF / Circumscription) | |
| | Maps OWA unknowns to safe | |
| | CWA negation-as-failure | |
| +-----------------------------+ |
+-----------------------------------------------------------------------+
As Peter F. Patel-Schneider demonstrated in Using Description Logics for RDF Constraint Checking and Closed-World Recognition, standard open-world description logic axioms can be repurposed for data validation and constraint checking simply by enforcing a closed-world interpretation over the target dataset. In large-scale enterprise environments, systems like SHER, developed by Kavitha Srinivas in OWL Reasoning in the Real World: Searching for Godot, use hybrid reasoning modules to partition massive datasets, applying closed-world data management techniques where performance matters and open-world inference where logical flexibility is required.
This integration also extends to procedural and behavioral systems. Martín Rezk tackled the integration of production rule systems with description logics in Combining Production Systems and Ontologies, providing a formal fixed-point semantics to reconcile operational rules with static domain models.
Furthermore, when systems must operate in unpredictable environments, they require solutions that are logically sound across multiple possibilities. To address this, Sanja Lukumbuzya et al. introduced Resilient Logic Programs (RLPs) in Resilient Logic Programs: Answer Set Programs Challenged by Ontologies. RLPs combine Answer Set Programming (ASP) with ontologies to ensure that a generated plan or solution remains valid ("resilient") across every potential scenario allowed by the open-world ontology.
Even in natural language processing, as explored early on by Leonardo Lesmo et al. in Interpretation of Noun Phrases in Intensional Contexts, hybrid architectures are essential to separate real-world assertions from purely conceptual or hypothetical objects (such as beliefs or desires) created within specific context spaces.
Ultimately, hybrid open- and closed-world reasoning succeeds because it avoids forcing a false choice between logical certainty and practical assumption. By creating formal interfaces—whether through modal operators, circumscription, or autoepistemic embeddings—modern knowledge representation systems can safely exploit the power of both paradigms simultaneously.
Key Foundational Work
DL_clog: A Hybrid System Integrating Rules and Description Logics with Circumscription (2007) — contributed 3 concept elements to this direction.
Interpretation of Noun Phrases in Intensional Contexts (1988) — contributed 2 concept elements to this direction.
A Faithful Integration of Description Logics with Logic Programming (2007) — contributed 1 concept elements to this direction.