Description Logic
Origin. Evolved from semantic networks and frame systems (1970s-80s). KL-ONE (Brachman & Schmolze, 1985) was influential. The "description logic" family formalized with attention to computational complexity. Foundation for OWL (Web Ontology Language) and the Semantic Web. Balances expressiveness with decidability.
Models. Concepts (classes), roles (relations), and individuals. Ontologies defining what kinds of things exist and how they relate. "A parent is a person with at least one child." Knowledge representation with formal semantics and decidable reasoning: subsumption, consistency, instance checking.
Formalism.
Basic elements:
- Concepts: classes/types (Person, Animal, ...)
- Roles: binary relations (hasChild, worksFor, ...)
- Individuals: named objects (john, mary, ...)
Concept constructors (ALC and extensions):
| Syntax | Name | Semantics |
|---|---|---|
| A | Atomic concept | A^I ⊆ Δ^I |
| ⊤ | Top | Δ^I (everything) |
| ⊥ | Bottom | ∅ (nothing) |
| ¬C | Negation | Δ^I \ C^I |
| C ⊓ D | Intersection | C^I ∩ D^I |
| C ⊔ D | Union | C^I ∪ D^I |
| ∀r.C | Value restriction | {x | ∀y. (x,y)∈r^I → y∈C^I} |
| ∃r.C | Existential | {x | ∃y. (x,y)∈r^I ∧ y∈C^I} |
| ≥n r.C | At-least | {x | |{y | (x,y)∈r^I ∧ y∈C^I}| ≥ n} |
| ≤n r.C | At-most | {x | |{y | (x,y)∈r^I ∧ y∈C^I}| ≤ n} |
DL naming convention:
- AL: attributive language (basic)
- C: complement (full negation)
- E: existential quantification
- N: number restrictions
- Q: qualified number restrictions
- H: role hierarchies
- O: nominals (individuals in concepts)
- I: inverse roles
- S: ALC + transitive roles
Example: SHOIN (basis for OWL-DL)
Knowledge base:
- TBox: terminological axioms (C ⊑ D, C ≡ D)
- ABox: assertional axioms (C(a), r(a,b))
- RBox: role axioms (r ⊑ s, transitivity)
Reasoning tasks:
- Concept satisfiability: is C consistent?
- Subsumption: does C ⊑ D hold?
- Instance checking: is a ∈ C?
- Consistency: is the KB satisfiable?
- Classification: compute full subsumption hierarchy
Symbols.
| Symbol | Unicode | Name | Meaning |
|---|---|---|---|
| ⊤ | U+22A4 | Top | Universal concept |
| ⊥ | U+22A5 | Bottom | Empty concept |
| ⊓ | U+2293 | Intersection | Concept and |
| ⊔ | U+2294 | Union | Concept or |
| ¬ | U+00AC | Complement | Concept negation |
| ∀ | U+2200 | Value restriction | All related are |
| ∃ | U+2203 | Existential | Some related is |
| ⊑ | U+2291 | Subsumption | Subclass of |
| ≡ | U+2261 | Equivalence | Same extension |
| ≥n, ≤n | — | Number restriction | Cardinality bounds |
Metatheory. ALC is EXPTIME-complete. Adding features increases complexity: SHOIN is NEXPTIME-complete. Decidability is preserved by careful design; arbitrary first-order features break it. Tableau algorithms are standard: build a model or find contradiction. OWL 2 profiles (EL, RL, QL) trade expressiveness for polynomial reasoning.
Applies to. Ontology engineering (Protégé, OWL). Semantic Web. Biomedical ontologies (SNOMED CT, Gene Ontology). Knowledge graphs. Data integration. Conceptual modeling. Query answering over ontologies.
Limitations. Trade-off: more expressive = harder to reason. No rules, no recursion (extensions like DL-safe rules address this). Closed-world vs open-world: DLs use open-world assumption (absence doesn't mean falsity). Large ABoxes challenge scalability. Unique name assumption often needed. Non-experts find DL syntax opaque.
© 2026 Lingenic LLC