What an Ontology Actually Is

Gruber, unpacked

An ontology is a formal, explicit specification of a shared conceptualisation. — Tom Gruber, 1993

Four words carry all the weight:

• Formal — machine-processable. Natural language doesn't qualify.
• Explicit — every concept, relation and constraint is named and stated. Nothing implicit, nothing hand-wavy.
• Shared — agreed by a community of users. A one-person model is a private schema, not an ontology.
• Conceptualisation — an abstract view of what exists in a domain (entities, properties, relations), independent of any particular dataset.

Strip any of those four, and what you have is something less: a glossary, a schema, an ER diagram, a class hierarchy in code.

The four things an ontology gives you

1. Shared vocabulary — your Customer and my Customer are the same thing.
2. Formal semantics — a machine can check consistency and infer new facts.
3. Domain knowledge — the rules of the world, not just the shape of one DB.
4. Interoperability — two systems built independently can exchange data.

An ontology is like a city's zoning code:

• It defines categories (residential, commercial, industrial) — these are the classes.
• It defines how categories relate (commercial may neighbour residential at low density) — these are the properties with their constraints.
• Every plot of land is an individual classified under one or more categories.
• A building inspector is the reasoner: walks the code, checks every plot against the rules, flags violations.

A schema is the blueprint of one building. An ontology is the zoning code the whole city plays by.

The semantic spectrum — every step adds expressivity at the cost of complexity. Pick the cheapest step that still answers your competency questions.