Applied and Formal Ontology

A reader arriving from computing will have met the word "ontology" in a quite different sense: a machine-readable specification of the concepts in a domain and the relations among them, used to make data interoperable. Gene Ontology, SNOMED CT, schema.org, and the OWL files behind a knowledge graph are all "ontologies" in this sense.

The two usages are related — the engineering practice took its name from the philosophical one deliberately — but they are not the same activity, and confusing them wastes a good deal of time in both directions. This page exists to prevent that.

The engineering sense

Gruber's standard definition: an ontology is a formal, explicit specification of a shared conceptualisation. Each word is load-bearing.

• Formal — machine-processable, with an inference procedure.
• Explicit — the categories and constraints are written down rather than tacit in a schema.
• Shared — the point is agreement across systems and institutions.
• Conceptualisation — an abstract model of a domain, for some purpose.

An ontology in this sense specifies classes (Patient, Diagnosis, Gene), properties relating them, hierarchies of subsumption, and constraints (a Diagnosis has exactly one subject; Gene and Protein are disjoint). Reasoners then derive subsumption and consistency automatically.

Upper ontologies try to supply the top-level categories that domain ontologies specialise:

• BFO (Basic Formal Ontology) — the most philosophically self-conscious. Its top-level split between continuants (things that persist through time, wholly present at each moment) and occurrents (processes, which have temporal parts) is the endurance/perdurance distinction adopted as a design decision. Widely used in biomedicine.
• DOLCE — Descriptive Ontology for Linguistic and Cognitive Engineering; explicitly descriptive rather than revisionary, aiming to capture ordinary categories rather than fundamental ones. The Strawsonian ambition, in software.
• SUMO — Suggested Upper Merged Ontology; broad coverage, first-order axiomatisation.

The formal machinery

Applied ontologies are written in description logics, which are decidable fragments of first-order logic. The trade is deliberate: expressive power is sacrificed to keep subsumption and consistency checking decidable and, ideally, tractable. OWL 2 and its profiles (EL, QL, RL) are standardised description logics with exactly this motivation, each profile drawing the line at a different point on the expressiveness/complexity curve.

The model theory is standard: classes are interpreted as subsets of a domain, properties as binary relations, and subsumption is containment across all models. Nothing exotic is happening semantically — which is itself worth noting, since it means the applied ontologist has already made the choice of an objectual, classical, first-order framework that the existence and quantification page treats as substantive.

What the two share

The overlap is real and not merely verbal.

• Categories and taxonomy. Both organise a domain into most-general kinds and species. BFO's continuant/occurrent split is a philosophical thesis, whatever its engineering role.
• Criteria of identity. A well-built ontology must say when two records denote the same patient, gene, or event, and "no entity without identity" is as much an engineering constraint as a philosophical slogan.
• Parthood. Mereology is unavoidable in any domain with structured objects, and applied ontologies routinely need the distinctions the mereology page draws — proper parthood, transitivity, whether a part of a part of an organ is a part of the organ.
• Dependence. Roles, dispositions, and qualities in BFO are dependent continuants: they cannot exist without a bearer, which is exactly ontological dependence.
• Occurrents. Whether processes have temporal parts, and how a process relates to its participants, is the events debate under another name.

Practitioners of applied ontology are frequently well read in the philosophical literature — BFO's designers explicitly draw on it — and the traffic has been genuinely two-way, with the applied work sharpening questions about identity criteria that philosophy had left informal.

Where they diverge

Four differences matter, and each explains a characteristic misunderstanding.

Purpose versus truth. An applied ontology is built for a purpose — data integration, query answering, clinical coding — and is assessed by whether it serves it. It is a designed artifact. Philosophical ontology asks what there is, and correctness is not relative to an application. Asking whether BFO is true is close to a category mistake; asking whether it is useful for biomedicine is the right question.

Pluralism by design. Many ontologies coexist, deliberately, for different domains and purposes, with mappings between them. In philosophical ontology, competing schemes are rivals — though note that Dupré's promiscuous realism and quantifier variance are exactly the positions on which the engineering situation is the metaphysically correct one.

Tractability constrains content. Description logics omit expressive resources — full negation, arbitrary quantifier nesting, most of second-order logic — because reasoning must terminate. So an applied ontology's categories are shaped by computational cost, which is not a consideration in the philosophical case.

Fundamentality is not the question. Applied ontologies are frankly built at the level of the domain: patients, diagnoses, and genes, with no suggestion that these are fundamental. The section's questions about grounding and levels simply do not arise, and there is no expectation that a biomedical ontology reduce to physics.

The terminological warning

Three points to keep in view:

1. In the applied sense, "an ontology" is a count noun — one builds an ontology, compares two ontologies, merges them. In the philosophical sense, ontology is a subject, and has no plural.
2. Applied ontologies are artifacts; philosophical ontology is an inquiry. Neither reduces to the other.
3. Nothing about the existence of many good ontologies in the engineering sense shows that ontological pluralism is true in the philosophical sense. That inference is tempting and invalid — the engineering plurality follows from the diversity of purposes, which the philosophical question is not relative to. Whether it is nevertheless relative to something is the deflationist's claim, and it needs its own argument.

Where this sits

This page is a boundary marker rather than a contribution. Its one substantive observation is the third point above: the applied field is a working demonstration that many mutually inconsistent classification schemes can each be perfectly good, and that fact is evidence for nothing about the philosophical question, because the two enterprises answer to different standards. Seeing clearly why it is evidence for nothing is a useful exercise in what the philosophical question is actually asking.