Ontology

An ontology gives explicit meanings to classes and relations in a domain. It may say that a clinical trial is a kind of study, define what a sponsor relation connects, or declare that two terms are equivalent under stated conditions. Formal ontologies can support logical inference, but the word also covers lighter conceptual models. A vocabulary becomes useful only when its terms are applied consistently.

Practitioners use ontologies to align sources, constrain data entry and make queries intelligible across datasets. Definitions should distinguish close but consequential concepts, such as sponsor, funder and collaborator. Changes to an ontology can change the meaning of existing assertions, so a record needs to know which version governed interpretation. Formal constraints do not prove that input facts are true.

For defensible action, a reviewer must understand what a field or relationship claims, not merely see its label. An ontology supplies that semantic contract. Schema alignment maps external fields into it; a knowledge graph uses its relation types; open-world and closed-world assumptions govern what absence of a statement permits one to conclude.

A practical method is set out in the Stanford guide Ontology Development 101: define the domain and scope, reuse existing ontologies where possible, list the terms, define classes and their hierarchy, define properties, and create instances. The guide makes a point that is often ignored: there is no single correct model of a domain, and the best test is whether the ontology serves the application it was built for. That test can be made explicit through competency questions, the questions the ontology must be able to answer, which the article on competency questions turns into a traceable contract between business need and ontology.

Ontologies change. Terms are split, merged and retired, and every change affects data already described with them. A responsible release process versions the ontology, publishes the differences, keeps retired terms resolvable with a pointer to their replacements, and checks the result against the data, as discussed in the articles on safe ontology releases beyond SHACL conformance and on deprecating and replacing terms.

The ontology gives a knowledge graph its vocabulary and drives schema alignment when another source is mapped to it. Its inferences are governed by the open-world assumption, and the checks that data are complete in some respect are typically written separately, as closed-world constraints. An ontology is a model and not a source of facts about the world: loading it into a system does not make any statement in the data true.