AI Glossary / AI and machine learning / Advanced
Description logic
Definition
Description logic is a family of formal knowledge-representation languages for describing classes of things, their relationships and individual instances. It supports automated reasoning about what follows from those descriptions, such as class membership and consistency.
What it means
A business may need a precise account of concepts such as supplier, approved supplier and component. Description logic provides a way to state relationships among these concepts so software can derive consequences systematically.
For example, a class can be defined through properties and restrictions rather than only a manually maintained list. A reasoner can then identify instances that satisfy the definition or detect incompatible assertions.
Description logics underpin important ontology technologies, including description-logic-based parts of OWL. They are not ordinary prose descriptions or a statistical learning method. The system reasons from the knowledge and semantics supplied to it; it does not automatically discover whether the starting assertions are true.
How it works
An ontology defines classes, properties and axioms. Individual assertions record facts about particular entities. A reasoner checks consequences under the selected logic.
Different description logics offer different expressive features and computational properties. Choosing a more expressive language can make reasoning more demanding, so practical ontology design considers the tasks and available reasoner.
Open-world semantics are particularly important in OWL-style use: a missing assertion is not automatically false. Also, different names do not necessarily denote different entities unless that distinction is stated or follows from the ontology. These conventions differ from common database expectations.
Three practical examples
- A product catalog infers broader category membership from formally defined component classes.
- An enterprise knowledge graph checks whether new assertions conflict with declared class relationships.
- A data-integration project maps terminology across systems using explicit concept definitions and equivalences.
Think of it this way
Think of a carefully written classification handbook whose rules can be applied mechanically. If every rechargeable device is a device, identifying an item as rechargeable also places it in the broader class.
The analogy has limits because formal logic follows exact semantics. Everyday readers often assume unstated facts or exceptions that the ontology has not encoded.
Worked example
Understanding missing information
Imagine a fictional ontology stating that every approved supplier is a supplier. It also states that Northstar is an approved supplier. A reasoner can conclude that Northstar is a supplier.
Now consider another entity, Lakeside, listed only as a supplier. Under open-world reasoning, the absence of an approved-supplier assertion does not establish that Lakeside is unapproved. Its approval status may simply be unknown.
If an operational workflow requires proof of approval before ordering, that decision rule must be designed explicitly. The ontology's inability to infer approval is not the same statement as a formal proof of non-approval, even though both may lead the workflow to request more information.
The example shows why inference and completeness validation should be distinguished. A system can reason correctly from incomplete knowledge while still lacking the evidence required for a business action.
Illustrative case study
A fictional engineering company combines catalogs from several subsidiaries. One system uses “module” for a complete assembly, while another uses it for a replaceable subcomponent.
The knowledge team writes explicit definitions and relationships rather than merging labels by spelling alone. A reasoner helps identify contradictory classifications after integration.
Domain specialists review the underlying assertions and decide which concept mappings are justified. The ontology improves consistency, but the team does not assume that a logically consistent catalog is factually complete or correct. The case illustrates the joint roles of formal reasoning and subject-matter review.
Common misconceptions and mistakes
- Treating missing facts as false under open-world semantics: absence can mean unknown.
- Assuming different names always identify different entities: identity semantics need attention.
- Equating consistency with truth: a coherent set of wrong assertions can still be wrong.
- Adding expressive features without considering reasoning cost: language choice affects practical performance.
Business use
Description logic is useful when precise concept definitions and machine-checkable relationships matter, particularly in knowledge graphs and data integration. It can support reuse of a shared vocabulary across teams.
Ask which reasoning tasks the ontology must support and how unknown information is handled. Separate inferred facts from source assertions and preserve provenance. If the workflow requires mandatory fields or closed-world validation, choose an appropriate validation mechanism rather than assuming logical inference alone checks every operational requirement.
People also ask
Is description logic machine learning?
It is primarily a formal representation and reasoning framework, though it can be combined with learning.
Is OWL identical to one description logic?
OWL is a family of ontology languages with profiles and semantics; description logics provide key foundations for parts of it.
Can a reasoner add useful facts?
It can derive entailed consequences, not invent unsupported factual observations.
Check your understanding
An ontology records a person as an employee but says nothing about managerial status. Under open-world semantics, is the person necessarily not a manager?
Answer: No. The status may be unknown. A negative conclusion needs appropriate supporting assertions or axioms.
Related terms
- Ontology
- Knowledge representation
- Automated reasoning
- OWL
- open-world assumption
Learn next
Read ontology and open-world assumption. Write a small set of class relationships and distinguish what follows logically from what remains unknown.
More in AI and machine learning
- Dependability state model - A dependability state model represents a system through states such as operational, degraded and failed, together with possible transitions between them.
- Dependency network (graphical model) - A dependency network is a probabilistic graphical model that describes each variable through a conditional distribution given selected other variables.
- Depth - Depth is how many layers or levels a model has, counting from input to output, and more depth lets a model build more complicated ideas from simpler ones.
- Determining the number of clusters in a data set - Determining the number of clusters means choosing how many groups a clustering analysis should use or report.
- Deterministic annealing - Deterministic annealing is an optimization method that starts by treating the problem as soft and blurred, then gradually sharpens it, so the search avoids...
- Detrended correspondence analysis - Detrended correspondence analysis, or DCA, is an ordination method developed to summarize ecological community data while modifying certain patterns...
Sources
W3C, OWL Web Ontology Language Reference: https://www.w3.org/TR/owl-ref/, earlier standard excerpts consulted. W3C, OWL Web Ontology Language Guide: https://www.w3.org/TR/owl-guide/, introduction and sections 2 and 4.2 checked for description-logic foundations, entailment, open-world and identity semantics. Examples are original.
Previous: Derived Relation Next: Descriptive analytics