In the field of knowledge representation, two fundamental philosophies often clash. On one side is the Open-World Assumption (OWA), the bedrock of the Semantic Web and Description Logics (DL). In an open world, if a fact is missing from a database, we simply assume it is unknown. On the other side is the Closed-World Assumption (CWA), used by relational databases and Logic Programming (LP). In a closed world, if a fact is missing, it is assumed to be false.
This research direction focuses on building hybrid systems that can navigate both worlds simultaneously—allowing for the discovery of new knowledge while maintaining strict data integrity.
The Friction Between Discovery and Validation
The tension between these two assumptions is best seen in how they treat "rules." As Boris Motik, Ian Horrocks, and Ulrike Sattler explain in Bridging the gap between OWL and relational databases, axioms in an open-world ontology act as deductive rules. If the ontology says "All Managers are Employees" and we know "Alice is a Manager," the system infers "Alice is an Employee."
In contrast, database schemas act as integrity constraints. If a database requires every Employee to have an ID, and Alice doesn't have one, the system rejects the data. Reconciling these is difficult: how do you allow a system to "fill in the gaps" (OWA) while also "checking for errors" (CWA)? Peter F. Patel-Schneider demonstrated in Using Description Logics for RDF Constraint Checking and Closed-World Recognition that we can force DL axioms to behave like constraints by adopting a closed-world interpretation, but doing so globally sacrifices the power of open-world inference.
Circumscription: Reasoning with Exceptions
One of the most robust ways to bridge this gap is through circumscription. This logical mechanism allows a system to "minimize" certain predicates—essentially saying, "assume these exceptions don't exist unless we explicitly know they do."
Xiaoping Chen introduced $\mathcal{DL}log$ in DL_clog: A Hybrid System Integrating Rules and Description Logics with Circumscription to handle non-monotonic reasoning. This allows for "common sense" rules like "Birds fly, unless they are penguins." In classical logic, a single penguin would make the rule "All birds fly" false. With circumscription, the system can maintain the general rule while gracefully handling the exception. Qian Yang and Jia-Huai You further unified this in Integrating Rules and Description Logics by Circumscription, showing that many different hybrid logic frameworks are actually just different "policies" of the same underlying circumscriptive logic.
DBoxes and Clopen Knowledge Bases
To make these systems practical for real-world data, researchers have moved toward "partitioning" the world. Instead of making the entire system open or closed, they define specific areas of certainty.
- DBoxes: Inanç Seylan, Enrico Franconi, and Jos de Bruijn proposed the concept of a DBox in Effective Query Rewriting with Ontologies over DBoxes. A DBox treats specific parts of an ontology as complete database tables. This allows complex queries to be rewritten into standard SQL, making reasoning over massive datasets feasible.
- Clopen Knowledge Bases: Labinot Bajraktari, Magdalena Ortiz 0001, and Mantas Simkus expanded this in Combining Rules and Ontologies into Clopen Knowledge Bases. In a "clopen" (closed + open) system, predicates defined by rules (CWA) can be embedded directly within the ontology (OWA). This allows the ontology to act as a set of integrity constraints over the outputs of rule-based deductions.
Resilience and Conditional Answers
The ultimate goal of this integration is to enable systems to reason effectively under uncertainty. In Resilient Logic Programs: Answer Set Programs Challenged by Ontologies, Sanja Lukumbuzya and Mantas Simkus describe "resilient" solutions—logic programs that remain valid across every possible scenario the ontology might describe.
Similarly, Medina Andresel and colleagues explored Assumptive Ontology-Mediated Queries in Query Rewriting for Ontology-Mediated Conditional Answers. Instead of failing when data is missing, these systems can provide "conditional answers": "This result is true if you assume these specific facts are also true." By using universal quantification to iterate over possible groundings, these frameworks ensure that assumptions about "closed" data stay within the bounds of what is actually known, while "open" components remain free to infer new possibilities.