Formal and Applied Ontology

The Philosophical Task of Formal and Applied Ontology

Formal and applied ontology investigates how the most general distinctions needed to describe a domain can be articulated precisely and used responsibly. A hospital distinguishes a biological disorder from its diagnosis, a patient from a clinical record, and a treatment procedure from the therapeutic effect it may produce. A museum distinguishes a physical object from a work of art, an original from a reproduction, and ownership from custody. A scientific database distinguishes a measurement, its unit, the measured phenomenon, and the instrument that produced it. These distinctions condition which statements can be made, which questions can be answered, and which errors can be detected. Their stability matters before a computer executes a query.

A formally represented ontology makes commitments explicit through a vocabulary and axioms interpreted within a stated semantics. Its philosophical significance arises from the question whether these commitments adequately distinguish the entities and relations of the intended domain. Formal validity provides exact consequences of premises. The justification of the premises requires further philosophical analysis, empirical evidence, and attention to the social practices that establish some entities. A perfectly consistent database can classify every medical appointment as an illness. Its consistency would leave its conceptual error untouched.

The term ontology connects a philosophical inquiry into being with a family of engineered representational artifacts. The historical connection is substantive: both demand answers concerning kinds, identity, dependence, and relation. Their methods and success conditions diverge. Philosophical ontology may ask whether universals exist independently of instances; an information ontology may introduce a class expression for a kind without deciding that metaphysical controversy. Conversely, computational interoperability may demand a precise identity criterion where ordinary philosophical analysis permits unsettled alternatives. Gruber's (1993) portable specification program and Smith's (1998) account of the basic tools of formal ontology identify different but complementary aspects of this problem.

The central claim of this monograph is that formalization is an accountable transition from conceptual distinctions to inferentially governed representation. The transition requires a sequence of decisions: identify the domain, determine the intended entities and relations, distinguish asserted facts from modeling conventions, select semantic commitments, test inferential consequences, evaluate real cases, and maintain the ontology as classifications and evidence change. Every step bears philosophical consequences because changing a relation from parthood to subclasshood, or a record from an observation to a disease, changes what the system is allowed to infer.

The historical inquiry therefore cannot be separated from technical detail. Aristotle's distinction between predication and inherence matters when engineers confuse a kind with its bearer. Husserl's dependence distinctions matter when an attribute is modeled as an independent object. Quine's logic of quantification matters when a model's vocabulary is mistaken for an inventory of what exists. Modern ontologies make these classical questions practically testable, yet each implementation remains answerable to the world it proposes to describe.

Definition, Scope, and Three Levels of Commitment

Formal ontology is the systematic study and explicit characterization of domain-independent structures of entities and relations, including category, identity, part, whole, dependence, and related forms of organization. In philosophical use, the term may identify a theory of the formal structures shared across material domains; in knowledge engineering, it may designate an axiomatized representational system built to make such structures usable in software. Applied ontology is the disciplined adaptation of ontological analysis to specific fields, practices, and systems. Its products include domain ontologies, reference ontologies, controlled classifications with formal semantics, and linked conceptual models.

An engineering ontology is commonly understood as an explicit specification of a conceptualization, following Gruber (1993). The formulation deserves precision. A conceptualization selects the intended distinctions and relationships of a domain; a specification articulates them sufficiently clearly for interpretation and use. A formal specification makes at least some interpretations and inferential consequences mathematically tractable. It does not follow that every specification expresses a single universal conceptualization or that every language offering a taxonomy possesses a fully developed formal ontology.

Three analytically different commitments are involved. The first is metaphysical: whether the proposed categories and relations genuinely characterize what exists, and in what sense. The second is semantic: what structures count as interpretations of the formal language, and which claims hold in every model of the axioms. The third is operational: how symbols, records, measurements, identifiers, and organizational procedures are connected to phenomena and maintained across contexts. An ontology can be semantically consistent yet metaphysically disputed; metaphysically plausible yet too expressive for efficient reasoning; or logically rigorous yet operationally disconnected from reliable data.

Consider the sentence 'Every biopsy specimen is part of an organism.' Its conceptual content depends on the intended temporal interval, whether detached specimens are included, and what counts as the relevant part relation. A formal axiom may express a universal restriction. Under ordinary first-order semantics its consequences follow from the chosen interpretation; under a description logic, expressibility and reasoning depend on the language fragment. In an operational system, specimen identifiers, handling dates, and chain-of-custody records determine what users will infer. Confusing these commitments produces false certainty.

Formal ontology overlaps with metaphysics, logic, epistemology, philosophy of science, and philosophy of technology. Its own research object is the relation between ontological analysis and disciplined formal representation. This focus distinguishes it from independent investigations of existence, universals, identity, grounding, and logical consequence, which it engages only where they determine the adequacy of modeling. The goal is a principled connection between formal meaning and subject-matter accountability.

Aristotle and the Historical Origins of Categorical Analysis

Aristotle's Categories begins from distinctions whose importance reaches far beyond the later technical vocabulary of ontology engineering. In Categories 2, at 1a20–1b9, Aristotle classifies beings by whether something is said of a subject and whether it is present in a subject. In Categories 4, at 1b25–2a4, he enumerates categories customarily rendered as substance, quantity, quality, relative, place, time, position, state, action, and affection. The classifications answer distinct questions. The first analyzes relations of predication and inherence; the second articulates broad ways of signifying what can be said of things. Reading them as a database schema obscures their philosophical structure (Aristotle, Categories, trans. Edghill; primary translation).

In Categories 5, at 2a11–19, primary substance is exemplified by an individual person or horse. Species and genera appear as secondary substances. This distinction forces a recurring ontological decision: does a classificatory term name the individual bearer, a kind instantiated by individuals, or a predicate that applies to them? A computer language can represent all three with different syntactic entities, but the choice among them concerns what its statements mean. Calling the class Patient an individual patient does not merely change notation; it changes the permissible inferences concerning illness, death, and identity.

The Aristotelian account also shows why relations are not reducible to a hierarchy of nouns. An individual may possess a quality, occupy a location, act upon something else, or stand in a quantitative relation. Some facts characterize the subject's kind, while others depend on time or circumstance. A robust ontology must therefore distinguish kind membership from contingent predication and from relations involving multiple participants. In formal systems, these distinctions correspond to different roles for constants, predicates, classes, and property assertions, though the historical Aristotle did not use those modern devices.

Metaphysics Δ, especially book V, and Metaphysics Ζ expand the inquiry into substance, essence, and explanatory priority. These texts do not yield a mechanically unified modern ontology. Their value lies in showing that classification depends on explanatory commitments: how entities endure, what makes them what they are, and which aspects are basic. Modern engineers who reproduce the surface of the category list without asking these questions repeat terms while abandoning their rationale.

A historically disciplined use of Aristotle recognizes both the continuity and the transformation. Categorical ontology retains the demand to articulate kinds and relations. Formal ontology adds explicit logical interpretation, computational consequences, and institutional maintenance. That addition makes the adequacy of the initial distinctions more consequential, because a mistaken categorical division may propagate automatically through very large information systems.

Medieval Logic, Universals, and Modes of Predication

The medieval reception of Aristotelian logic made the problem of categorization inseparable from questions about universal terms, concepts, and the things to which they apply. Porphyry's Isagoge, traditionally transmitted with Aristotle's Categories, raises the status of genera and species, asking whether they subsist, whether they are corporeal, and whether they exist apart from sensible particulars. Medieval philosophers developed different responses to these questions, producing sophisticated accounts of predication, abstraction, common natures, and linguistic signification. The issue for formal ontology is the degree to which an organized vocabulary must settle a metaphysics of universals.

Thomas Aquinas distinguishes the nature considered in itself from the modes under which it exists in individuals and in intellect. In De ente et essentia, especially chapters 2–4, the relation between essence, individuation, and abstraction helps explain why the same conceptual content can be used in general definitions while being exemplified concretely. William of Ockham, by contrast, develops a nominalist analysis according to which general terms function through signification and mental representation without requiring corresponding universal things. The contrast concerns the ontological interpretation of predicables, not merely the structure of taxonomic trees (Aquinas, De ente et essentia; Ockham, Summa logicae, I).

A contemporary ontology of laboratory samples can contain a class for blood specimens while remaining neutral between realism and nominalism about universals. It must nevertheless determine whether blood specimens are individual material entities, whether the class denotes a set of such entities or a reusable kind, and whether properties such as specimen type, origin, and condition are intrinsic, relational, or temporally variable. Thus a model may bracket a metaphysical dispute but cannot avoid all philosophical interpretation.

Medieval theories of supposition and semantic context provide another enduring lesson. A term may stand for different kinds of items depending on the propositional environment. If a sentence concerns the word 'animal' rather than animals, its subject has shifted from creatures to linguistic expressions. The modern distinction between use and mention continues this problem. Knowledge-graph errors frequently arise when a label, concept scheme entry, document, individual, and real-world object are represented as if they were the same referent.

This historical material supports a general principle: a classification must specify what its units designate and which predications are licensed. An ontology is damaged when a taxonomic name is treated as an object of the kind it names, or when a statement about language is presented as a statement about things. The medieval elaboration of logic supplies a conceptual discipline whose relevance survives technological change.

Husserl: Formal Ontology, Material Regions, and Dependence

Edmund Husserl gave 'formal ontology' a distinctive philosophical role in the Logical Investigations, first published in 1900–1901, with the Third Investigation especially important for its analysis of parts, wholes, and foundation. His inquiry concerns structures that can apply across materially different domains: the relations of parts to wholes, of moments to independent pieces, and of dependent contents to what grounds their possibility. Formality here means independence from the specific content of a regional science, not merely the use of mathematical symbols. A formal ontology must therefore be distinguished from formal logic even when both share general structural aspirations (Husserl, Logical Investigations, Third Investigation; Smith, 1998).

The distinction between independent parts and dependent moments has direct explanatory force. A wheel may be detached from a vehicle and remain an object; the particular coloration of an apple cannot persist as that coloration without the bearer whose coloration it is. Exactly how dependence should be formalized is disputable, but a domain ontology that models an illness's severity, a patient's age, or a transaction's status as freely transferable standalone objects must confront the corresponding dependence questions. Their identifier records may persist when their bearers disappear; persistence of a record is not persistence of the represented quality.

Husserl's distinction between formal and material ontologies also helps place foundational and domain ontologies. A material ontology articulates the kinds and structures proper to a region of objects. A formal ontology seeks structures applicable across such regions. The distinction anticipates the engineering separation between upper-level categories and domain-specific terminologies, while the analogy remains imperfect. A deployed top-level ontology contains adopted modeling choices and explicit axioms. Husserl's philosophical investigation was broader than any contemporary file format or standardized taxonomy.

The relation of a dependent moment to its foundation is stronger than ordinary association. If the particular shape of a metal artifact necessarily requires some spatially extended bearer, an ontology should represent the dependence constraint rather than merely state that the shape 'is associated with' the artifact. Relational vocabulary matters because its inferential behavior embodies the metaphysical distinction. The same consideration applies to dependent social roles, information qualities, and institutional statuses, although their dependence conditions vary.

Husserl's contribution therefore extends beyond a historical label. It establishes a demand for conceptual discrimination among types of parts, wholes, and dependence. It also reveals a central difficulty: how to preserve subtle ontological structure when practical representation languages trade expressiveness for tractability. The solution must begin by determining which differences are philosophically and operationally indispensable.

Frege, Leśniewski, Carnap, and Quine

Gottlob Frege's Begriffsschrift (1879) transformed the logical analysis of generality and relational judgment. His later distinction between a concept and an object, elaborated in 'Über Begriff und Gegenstand' (1892), challenged the straightforward assimilation of predicates to names. In 'Über Sinn und Bedeutung' (1892), the distinction between sense and reference clarified why meaningful expressions and their referents must be distinguished. Ontology engineering inherits the basic challenge: names, class predicates, and assertions of identity occupy different logical roles. An identifier's equality to another identifier cannot be assumed solely because two descriptions are similar (Frege, 1879, 1892a, 1892b).

Stanisław Leśniewski developed a formal system explicitly called Ontology as part of a three-system foundation including Protothetic and Mereology. His Ontology is a rigorously articulated logic of names, distinct in its technical aim from contemporary metaphysical ontology. This terminological coincidence illuminates a genuine philosophical tension: one can formalize the logic of naming and predication without endorsing an unrestricted inventory of abstract entities. His work on parthood and nominalism also demonstrates that formal systems can differ sharply in the kinds of beings they permit (Leśniewski, 1927–1931; see the historical analysis in the Stanford Encyclopedia of Philosophy).

Rudolf Carnap's Der logische Aufbau der Welt (1928) sought a logically organized constitution system in which different kinds of scientific objects could be constructed through relations of definability. Later, 'Empiricism, Semantics, and Ontology' (1950) distinguished questions internal to a linguistic framework from choices about adopting a framework. Whatever one thinks of Carnap's philosophical conclusions, his analysis places modeling frameworks themselves within philosophical scrutiny. Choosing to represent diseases, institutions, or numerical quantities in a certain language alters which claims the framework renders straightforward.

W. V. O. Quine's 'On What There Is' (1948) made ontological commitment an explicit issue of regimented quantified theory. His familiar criterion examines which entities must be values of bound variables for accepted theoretical claims to be true. The criterion does not make every software class a metaphysical entity. Rather, it invites inquiry into the interpreted domain of a theory and the commitments incurred by its existential assertions. The distinction between 'the ontology contains a class Person' and 'the theory entails that at least one person exists' is indispensable. A vocabulary can be empty of instances while its schema remains expressive (Quine, 1948; Ontological Commitment).

Together these thinkers shift attention from labels to logical form, meaning, and theory. Frege clarifies predicates and reference; Leśniewski offers an alternative logic of names and parthood; Carnap exposes the constitutive role of frameworks; Quine requires explicit accounting of existential commitments. Their positions differ on crucial philosophical questions, but each makes an ontology more answerable for what its propositions actually assert.

Formal Languages, Models, and Ontological Interpretation

A formal ontological theory is intelligible through an interpreted language. Its signature specifies the nonlogical vocabulary: perhaps predicate symbols Person(x), Specimen(x), and partOf(x,y), alongside logical connectives, quantifiers, equality, and any permitted functions. Its axioms state constraints such as 'every specimen is a material entity' or 'no event is a kind of enduring object.' An interpretation assigns a domain and extensions to nonlogical symbols. Truth in a model depends on those assignments. An axiom is satisfiable when some admissible model makes it true; a theory entails a proposition when every model of the theory makes that proposition true.

An important example is subclasshood. The sentence ∀x(Surgeon(x) → Physician(x)) declares that every surgeon is a physician in the intended interpretation. The inverse does not follow. Nor does the sentence entail that surgeons exist: universal statements may be satisfied even when the extension of Surgeon is empty. When a domain application requires actual existence, the additional existential claim ∃x Surgeon(x) must be warranted separately. This is a small logical distinction with large consequences for data quality, automated reporting, and political or scientific claims derived from structured information.

Other distinctions are equally consequential. Consistency means an interpretation can satisfy the theory; completeness, in a relevant logical sense, concerns the relation between provability and semantic validity, while completeness of coverage concerns whether the vocabulary adequately describes a field. A logically consistent ontology may omit crucial categories. A domain-complete taxonomy may be encoded in a logic whose reasoning tasks are computationally difficult. Sound reasoning from erroneous classifications can preserve logical soundness while producing false empirical conclusions. Evaluation must therefore identify the exact sense in which an ontology is claimed to work.

Model theory also draws attention to multiple interpretations. An axiom system can have different models that agree on its explicit assertions. When two classes have the same extension in a dataset, that does not automatically make them identical in conceptual intension. Likewise, a model using Person and LegalPerson may deliberately distinguish kinds even if the present data happen to classify exactly the same entities under both. A domain model must be scrutinized for unintended equivalence, unintended disjointness, and implicit assumptions hidden in its use of relation symbols.

The correct philosophical question is what the formalism preserves, what it leaves open, and what additional conditions bind it to a domain. Model-theoretic semantics confers inspectable inferential discipline; it does not select the correct ontology by itself. The selection requires comparison with alternative conceptualizations, case evidence, and explicit purposes (Gruber, 1993; W3C, 2012, OWL 2 Structural Specification).

Categories, Universals, Instances, and Types

The most visible structure of many applied ontologies is a hierarchy of classes. Yet a hierarchy becomes philosophically intelligible only when it distinguishes the nature of its relata. A universal, on a realist interpretation, is a repeatable kind or property instantiated by numerically distinct individuals. A class in set-theoretic semantics denotes a collection of objects satisfying a condition. A type in a programming language constrains the values assigned to expressions or objects. A category in philosophical ontology articulates a maximally general way of being or classification. These notions can be connected through explicit mappings, but they perform different explanatory tasks.

A physician is an individual bearer of a professional role; Physician is a predicate of certain individuals; a hospital's staff directory is a collection of records; a text string 'physician' is a lexical item. A model that treats these as equivalent will misrepresent changes of role, employment, and record continuity. Distinguishing the kind, its instances, the bearer's status, and the document that identifies it enables correct inferences across time. The need arises even when a system uses only one namespace of identifiers, for a uniform storage syntax does not establish an ontological unity.

The realism–nominalism dispute shapes how classes are interpreted. Realists may view a scientifically established kind as a real structure discovered in the domain. Nominalists may treat classification as a disciplined relation among particular things and linguistic practices. Constructivist approaches emphasize the procedures through which classifications stabilize. The engineering requirement is to state which interpretation the ontology adopts and which conclusions require that interpretation. A disease taxonomy designed for epidemiological reporting might use operational criteria without claiming every classification captures a unique natural essence.

A subtype or subclass relation has demanding inferential consequences. If every bacterial infection is an infection, placing BacterialInfection under Infection is justified. Placing BacterialInfection under Bacterium would confuse a pathological process or condition with its causal organism. A subclass statement ordinarily represents an all-instance inclusion; it is not a causal, participatory, temporal, or part–whole relation. Such discipline was central to the ontological analysis pursued by Guarino and Welty (2002).

Formal ontology should preserve distinctions even where data labels are conventional. Scientific classification is historically revisable, and the apparent stability of a term may conceal changes in its diagnostic or operational criteria. Competent ontological modeling therefore tracks definitions, provenance, and revisions. A vocabulary that is fixed in syntax but unstable in referential practice will yield apparently comparable data whose empirical meanings diverge.

Identity, Individuation, and Criteria of Unity

Identity is the relation under which one and the same entity is recognized across descriptions, contexts, and times. The logical symbol = expresses numerical identity within a given interpretation, and substitutivity holds in classical extensional contexts. Yet an information system must decide when two database entries, files, observations, organisms, legal persons, or artworks designate the same entity. The inference from matching attributes to identity is a substantive assumption; two people may share every recorded characteristic and remain numerically distinct.

Criteria of identity differ with kinds. A material object may be tracked through a history of physical continuity, a software artifact through its maintained versioning relation, and a legal organization through institutionally specified conditions of continuation. These criteria are not interchangeable. A company can retain legal identity despite changes in shareholders and personnel; a medical specimen may lose identity under division or processing depending on the adopted scientific protocol. Ontology design must make those policies explicit rather than hiding them under a general 'same as' relation.

The risk is acute in OWL, where owl:sameAs denotes strong individual identity. If two distinct persons are joined by this relation because they have the same name, each property assertion about either person can propagate as if it applied to the other. Conversely, distinct identifiers do not in themselves prove distinct individuals, since OWL does not adopt a general unique-name assumption. Equality and inequality assertions therefore demand reliable evidential support (W3C, 2012, OWL 2 Primer).

Identity is related to unity: what makes an object one thing rather than a collection of parts? For an organism, organization and physiological integration matter; for a committee, legal constitution and membership rules matter; for a dataset, an editorial or technical policy may establish the object of reference. OntoClean employs criteria of identity and unity to assess taxonomic relations, thereby turning philosophical analysis into a practical design check (Guarino & Welty, 2002).

A general lesson follows. Identity should be modeled at the granularity appropriate to the question, with explicit provenance for equivalence claims. Matching records, shared features, overlapping spatial location, and causal continuity can provide evidence of identity; none substitutes automatically for a kind-sensitive criterion. A formal identity axiom is powerful precisely because it cannot judge the quality of the evidence that licenses it.

Mereology, Dependence, Constitution, and Boundaries

Mereology investigates parthood and wholes. A familiar classical principle makes parthood reflexive, antisymmetric, and transitive when the relation is interpreted as an order on entities of a relevant domain. Proper parthood is irreflexive and excludes identity. Additional principles such as supplementation, fusion, and extensionality remain philosophically contested. A theory accepting unrestricted fusion permits arbitrary sums; a domain ontology for anatomy, geography, or architecture may require more selective principles. It is therefore inaccurate to treat a generic partOf relation as logically self-explanatory (Leśniewski; Simons, 1987; Smith, 1998).

The difference between a part and a member is significant. An engine is part of a vehicle, while a driver may be a member of a team operating vehicles. A person is a member of a committee but does not ordinarily constitute a spatial component of that committee. A biological cell is part of an organism in a different sense from a dataset record's belonging to a collection. Some domains require several part relations, distinguished by temporality, granularity, and entity kinds. Conflating them can yield false transitive inferences across heterogeneous relations.

Ontological dependence also admits distinct forms. A quality depends specifically on its bearer; a reproducible text may depend generically on some bearer or encoding, while surviving the loss of any one physical copy. A role can depend on an institutional structure while being realized by a person. An event may depend on its participants without being identical to them. These differences matter for persistence, responsibility, database deletion, and explanatory reconstruction. Husserl's Third Investigation supplies a historical foundation; modern discussions distinguish existential, identity, rigid, and generic dependence (Ontological Dependence).

Constitution poses a further difficulty. The material constituting a sculpture can change under restoration, while the sculpture's identity may be tied to artistic and historical continuity. Material constitution should not be equated with numerical identity merely because two descriptions occupy the same region. Boundaries are likewise diverse: a coastline may be physically irregular and scale-dependent, a property boundary legally instituted, and an anatomical boundary partly conventional. Barry Smith's analysis of formal ontology emphasizes boundary, continuity, and contact alongside part and dependence (Smith, 1998).

Formalizing these structures requires decisions about the intended subject matter. Theorems proved for an abstract parthood relation apply to that relation under its axioms; whether a laboratory's relation 'containedIn' satisfies transitivity is a domain question. An ontology succeeds when its formal relations are both mathematically disciplined and evidentially appropriate to their intended applications.

Temporal Entities, Processes, and Change

Temporal ontology concerns the persistence of entities, the articulation of events, and the dependence of truth on time. A material object can be numerically the same object at two times while having different masses, components, or roles. An episode of surgery unfolds through temporal parts and cannot be treated as an enduring bearer in exactly the same sense as a surgical instrument. A medical encounter, a diagnosis, a treatment plan, and an outcome each possess a distinct temporal organization. Formal representation must therefore specify when assertions hold and whether the relation itself changes.

One modeling approach distinguishes continuants, which endure through time, from occurrents, which unfold in time. Basic Formal Ontology builds important categories around this division, including material entities, qualities, roles, processes, and temporal regions (Arp, Smith, & Spear, 2015; ISO/IEC 21838-2:2021). Another approach emphasizes perdurance, temporal parts, or explicit temporal indexing of properties. These alternatives encode different metaphysical assumptions and support different tasks. A system mapping them should preserve the information that their concepts treat as essential, rather than equating categories solely because they have similar natural-language labels.

Consider a patient who is febrile at one observation and afebrile at another. The statements Fever(p) and NotFever(p) are inconsistent if both assert the same temporally unrestricted property of the same individual in classical logic. They may be jointly true when indexed to different observation times. A sound ontology must represent the observation event, its temporal scope, its provenance, and the underlying condition. Merely adding timestamps as untyped strings does not guarantee appropriate temporal interpretation.

Temporal reasoning also interacts with event identity. Two clinical records may describe the same operation, two sequential episodes, or a broader hospital stay containing multiple events. Relationships such as before, overlaps, during, and participatesIn answer different questions. The choice of temporal granularity—seconds, days, episodes, historical periods—determines which comparisons are meaningful. International event-based data integration depends on handling these choices explicitly.

This leads to an important constraint on interoperability. A mapping between temporal ontologies is adequate only when it preserves the distinctions required by the intended queries. If one system represents roles as qualities of enduring bearers and another represents role episodes as temporally bounded events, a naive one-to-one class equivalence can destroy evidence about change. More complex correspondences may be needed to retain meaning.

Modality, Essentialism, and Context

Many ontological claims are modal: they concern what an entity must be, may be, or could not be. Essentialism asks which properties belong to a thing in virtue of what it is. Ordinary formal ontologies implement some modal information indirectly, through constraints on classification, participation, dependence, and disjointness. Such axioms have consequences relative to their chosen semantics; they do not automatically settle contested metaphysical accounts of necessity. A modeler who interprets a datatype restriction as a fact about every possible organism has made a philosophical inference that the technical language alone does not establish.

Necessary and contingent properties must be distinguished in time-sensitive domains. A particular employee is contingently employed by a firm; being a human organism may be treated as essential to a particular person in one philosophical framework and disputed in another. A legal entity exists through constitutive rules that may change. Formal ontology can accommodate multiple perspectives by stating the criteria and scope of its categorization, but it cannot dissolve a metaphysical conflict through an arbitrary label.

Context introduces a second dimension. 'Resident' may mean a person living in a location, a physician in postgraduate training, or a legal status within an institution. A shared vocabulary must assign a definite interpretation to each use. Context is not merely an annotation that can be deleted after data integration. It can determine the truth conditions of an assertion. Similarly, the expression 'has parent' differs across biological descent, adoption, genealogy, and data structures.

Modal claims also raise questions about counterfactual applications. A safety ontology may need to describe conditions that must hold for a procedure to count as compliant. A clinical ontology may describe expected or typical sequences without treating them as universal necessities. A taxonomy that converts defeasible defaults into strict universal axioms can become inconsistent when confronted with legitimate exceptions. The difference between norms, statistical regularities, and constitutive conditions must therefore be part of the model's design.

A mature ontology records where necessity arises: from logic, adopted definition, domain theory, or institutional rule. It also distinguishes concept invariants from implementation-specific assumptions. This discipline prevents a useful local standard from being mistaken for an exhaustive metaphysics of its subject matter.

Knowledge Representation and the Engineering Turn

The late twentieth century transformed ontology into a central topic of knowledge representation and information systems. Knowledge engineers faced a practical problem: distinct applications needed to share meanings rather than exchange opaque strings. Thomas Gruber's 'A Translation Approach to Portable Ontology Specifications' (1993) proposed a common representational vocabulary of classes, relations, functions, and other elements, together with methods for translating declarative content among systems. His account of an ontology as a specification of a conceptualization became a durable reference point because it tied reuse to explicit meaning.

This engineering turn should be understood historically. Logic programming, semantic networks, frame systems, database schema design, and artificial intelligence each contributed methods of organizing concepts. They also brought different assumptions concerning inheritance, default reasoning, constraints, and object identity. The point of ontology engineering was to make shared commitments sufficiently stable for communication across these representational cultures. The 1998 volume Formal Ontology in Information Systems, edited by Nicola Guarino, documented a field in which philosophical analysis, AI, database design, and information integration were increasingly in direct contact.

Guarino's program required greater attention to the intended meaning of representational choices. A vocabulary that describes both objects and roles must distinguish what is essential to an entity from what it temporarily does. A hierarchy that combines physical components, taxonomic kinds, and organizational units requires separate relations rather than an overloaded parent link. This is why ontology engineering involves philosophical criticism of data models rather than the mere translation of database schemas into RDF.

The distinction between a conceptual schema and an ontology is a matter of function and semantics, not a magical file extension. A database schema may constrain storage and keys without declaring an intended theory of material parthood, biological identity, or organizational dependence. A formally axiomatized ontology may support reasoning across systems while remaining incomplete as an account of operational data integrity. Many mature architectures need both mechanisms, each with an explicit contract concerning what it guarantees.

The engineering turn also made collaboration indispensable. Terms gain long-term value when their definitions, identifiers, provenance, and revision histories can be inspected by communities beyond their original developers. What began as a problem of representation becomes a problem of scholarly and institutional governance: who can amend the ontology, on what evidence, and with which duties to users of earlier versions?

Foundational Ontologies: DOLCE and BFO

A foundational or top-level ontology seeks general categories and relations reusable across many domains. Two influential examples reveal why foundations are philosophically substantive. DOLCE—Descriptive Ontology for Linguistic and Cognitive Engineering—originated in the WonderWeb work of the early 2000s and was designed around categories informed by cognition, linguistic practice, and commonsense description. Its formal axiomatization articulates entities such as endurants, perdurants, qualities, and their relations, serving as a basis for analyzing complex domains (Borgo et al., 2022).

Basic Formal Ontology, developed in a realist tradition closely associated with Barry Smith and collaborators, distinguishes continuants and occurrents and provides formal categories intended to support domain ontologies representing scientific reality. The BFO account treats roles, functions, dispositions, objects, qualities, and processes in ways meant to assist the construction of interoperable knowledge. ISO/IEC 21838-2:2021 codifies BFO-2020 terms and relations and their axiomatizations in OWL 2 and Common Logic. The existence of this standard makes precise technical conformance claims possible but does not settle the philosophical choice between alternative foundational frameworks.

DOLCE and BFO offer a productive comparison. DOLCE explicitly characterizes itself as descriptive of a commonsense or cognitively accessible worldview; BFO emphasizes a realist orientation suited to sciences of entities and processes. These emphases influence how an application represents socially constituted objects, mental contents, roles, or material events. A mapping based only on superficial category-name similarity may miss the divergent commitments behind the descriptions. The research question is which distinctions and inferences each foundation preserves for a given scientific or institutional purpose.

Consider a disease and its diagnosis. A disease process, an underlying pathological disposition, a physician's diagnostic act, a proposition in a medical record, and a patient's legal status can be located differently depending on the framework and modeling objective. The diagnostic term is not itself an occurrence in the patient's body. A successful ontology makes these differences explicit and establishes carefully typed relations among them. It must also allow revisions when medical knowledge changes.

The strongest reason to use a foundational ontology is coherent cross-domain reuse of distinctions whose interpretation has been examined and maintained. The strongest objection is that a universal categorization can impose inappropriate metaphysical commitments on specialized fields. The response is principled selection, documented scope, and rigorous mapping, not automatic allegiance to one supposedly complete hierarchy (Borgo et al., 2022; Arp et al., 2015).

OntoClean and the Discipline of Taxonomic Analysis

OntoClean, developed by Nicola Guarino and Christopher Welty, translates aspects of philosophical analysis into a method for assessing taxonomies. Its central insight is that a hierarchy can be syntactically well formed while violating ontological constraints. The method attends to metaproperties such as rigidity, identity, unity, and dependence in order to diagnose category confusions. Their 2002 account in Communications of the ACM explains how supposedly innocuous taxonomic decisions can license incorrect subsumption inferences (Guarino & Welty, 2002).

Rigidity concerns whether membership in a kind is essential to its instances. Being a person may function as a rigid sortal in a given ontology; being a student or a patient is normally a contingent role. A role can be acquired and lost without destroying the bearer. If Student is modeled as a superclass of Person, the subclass direction would suggest that all persons are students. If Person is modeled under Student, it would suggest that every person is necessarily a student in models whose hierarchy is so interpreted. Either move may misconstrue the intended dependence of a role on a bearer.

Identity and unity criteria help distinguish kinds of aggregates and continuants. 'Organization,' 'committee,' 'crowd,' and 'building' may all classify entities with multiple components, but they imply different principles for determining their members and persistence. Treating all aggregates as one class without relation-specific qualification may make a coherent simple taxonomy impossible. OntoClean forces the designer to ask what the kind means before choosing the inheritance structure.

The method's power also reveals its limits. Whether a kind is rigid may itself be disputed. Social kinds change with institutions, science revises biological classifications, and a role may be legally constitutive of an entity that otherwise would not exist. Metaproperty analysis is therefore a disciplined argument subject to evidence and conceptual scrutiny, not a mechanical test that automatically discovers true ontology.

Its enduring contribution is methodological accountability. A taxonomy is evaluated according to the implications of its classifications, especially where one generalization seems to subsume another. The method demonstrates that applied ontology needs philosophical judgments that can be stated, criticized, and revised.

RDF, Description Logics, and OWL 2

RDF provides a graph-based model for representing statements using subject–predicate–object triples. RDF 1.1 defines graphs as sets of such triples and distinguishes IRIs, blank nodes, and literals in its abstract data model (W3C, 2014). RDF is an infrastructure for linked representation; an arbitrary RDF graph is not automatically an adequate philosophical ontology. Meaning arises from vocabularies, semantic conventions, and explicit constraints. RDF Schema extends the vocabulary with class and property mechanisms, while OWL provides richer language constructs for ontological reasoning.

Description logics offer a family of logical formalisms for representing concepts, roles, and individuals with well-defined inference tasks. Subsumption asks whether one concept is necessarily included in another; satisfiability asks whether a concept could have an instance under a theory; consistency asks whether a theory has a model. These tests are crucial because complex sets of axioms may generate contradictions that are difficult to see by reading labels. Their results concern the axioms and semantics actually supplied, rather than observational truth.

OWL 2, standardized by W3C with a second edition in 2012, provides classes, object and data properties, individuals, and a rich vocabulary of axioms. It supports distinct semantic perspectives, including Direct Semantics for OWL 2 DL and RDF-Based Semantics. The profile families OWL 2 EL, QL, and RL impose different syntactic restrictions to obtain useful computational tradeoffs. EL targets efficient reasoning for large terminological ontologies, QL is designed for query answering with large data collections, and RL supports rule-oriented implementations. Language selection changes the reasoning a system can practically perform (W3C, 2012, OWL 2 Overview; OWL 2 Profiles).

Consider two axioms: every CardiacSurgery is a Surgery; every Surgery is a MedicalProcedure. A reasoner may infer that every CardiacSurgery is a MedicalProcedure. If another axiom declares MedicalProcedure disjoint with Document while incorrectly placing CardiacSurgery under Document, an instance may make the ontology inconsistent, depending on the precise axioms. Detecting such inconsistency is valuable but does not decide whether the underlying clinical concept is correctly described.

OWL's usual open-world reasoning has another major consequence: lack of an assertion is generally not evidence of its negation. A hospital record lacking an allergy entry cannot establish that the patient has no allergies. This differs from the integrity expectations often imposed on closed operational databases. The difference is philosophical as well as computational because it concerns what can be concluded from incomplete knowledge. Correct use of OWL therefore requires explicit policies governing the interpretation of missing information, identity, and domain completeness.

Common Logic, Rules, Constraints, and Validation

Different representational requirements demand different logical instruments. OWL 2 is deliberately constrained to support specified reasoning tasks. Common Logic, standardized in ISO/IEC 24707:2018, provides a framework for sharing information expressed in first-order logic and permits forms of quantified expression that go beyond many description-logic applications. The relative freedom of an expressive logical language enables careful axiomatization of complex relations, but practical inference may become less predictable or computationally demanding. Formal ontology thus confronts a design choice: which properties must be expressible, which entailments must be decidable, and how much implementation complexity can be justified by a domain's purposes.

Rules offer another representational route. A rule might derive the classification of a laboratory case from a pattern of observations, subject to the semantics and assumptions of its rule system. Derivation rules should be distinguished from definitional axioms and from institutional commands. A clinical classification rule can encode an evidence-based procedure for identifying cases without declaring that every entity satisfying the procedure is metaphysically identical to a disease. The exact treatment of negation, exceptions, uncertainty, and temporal change determines whether the rule represents strict consequence or defeasible judgment.

SHACL, the Shapes Constraint Language, became a W3C Recommendation in 2017. It describes conditions against which RDF data graphs can be validated. A shape may require that a Patient record contain an identifier, that a date property have an expected datatype, or that a relation occur with an allowed number of values. These are operational constraints on data graph conformance. They should not be confused with OWL's open-world ontological assertions, whose semantics allow some absent information to remain unknown (W3C, 2017, SHACL).

Suppose a hospital exchange standard requires each clinical specimen record to have exactly one collection date. A validation shape can report records with no recorded collection date. That failure establishes nonconformance with a data obligation; it does not establish that the specimen had no collection event. An OWL restriction describing specimens as having some collection event may be satisfied by an interpretation with an unnamed event. Different results follow because data completeness and model-theoretic satisfiability answer different questions.

This distinction also clarifies the relation to SPARQL queries. A query can retrieve specified graph patterns under its formal semantics; it cannot by itself guarantee that a returned match accurately refers to a domain object. Query answers may require provenance checks, entailment regimes, or additional validation. A system that reports complete answers must state the relevant closed-world and data coverage assumptions (W3C, 2013, SPARQL 1.1 Query Language).

The philosophical conclusion is that representation, inference, and validation are related practices with different norms. Ontological axioms articulate intended structures; inference evaluates their consequences; validation tests a graph against specified obligations; empirical and institutional review evaluates whether the graph describes the intended world. A mature architecture assigns each task to the appropriate mechanism and discloses the assumptions on which its outputs depend.

Ontology Development, Evaluation, and Governance

Ontology development requires a methodology because the most important failures arise before formal axioms are tested. The first phase is conceptual investigation: identify the community, the domain, the questions the representation must answer, and the distinctions that are likely to affect those answers. 'Competency questions' turn expectations into concrete tests: Can the system distinguish the diagnosed disease from its diagnostic act? Can it retrieve every specimen collected during a specified interval? Can it identify which legal institution issued an authorization? Questions of this kind prevent an attractive hierarchy from being mistaken for a successful research artifact.

METHONTOLOGY, developed in the 1990s by Fernández-López, Gómez-Pérez, and collaborators, organized ontology development into life-cycle activities involving specification, conceptualization, formalization, implementation, and maintenance. The life-cycle approach remains useful because ontologies are revised as a domain and its evidence change. Project initiation should also review existing ontologies: reuse can improve interoperability, while unexamined reuse can import inappropriate assumptions into an application (Fernández, Gómez-Pérez, & Juristo, 1997; Fernández-López & Gómez-Pérez, 2002).

Evaluation has several distinct objects. Logical evaluation assesses consistency, satisfiability, unintended equivalences, and inferential behavior. Structural evaluation examines documentation, naming practices, category patterns, and the integrity of intended relationships. Empirical evaluation compares assertions with actual domain evidence and representative cases. Usability evaluation asks whether researchers, clinicians, curators, or developers can employ the model to perform the required tasks. Organizational evaluation asks whether the ontology has a legitimate procedure for correcting errors and recording version changes. A success claim should specify which tests were performed and what their coverage was.

Ontology testing should include counterexamples. A role classification may appear correct when every sample instance happens to possess the role. It should also be tested against an individual that loses the role while remaining the same bearer. A material-part relation should be tested against detached components. A taxonomic hierarchy should be tested with borderline classifications and historical records. This method exposes hidden assumptions that happy-path examples conceal.

Governance is constitutive of long-lived meaning. Stable identifiers help consumers recognize concepts across releases. Definitions should distinguish the label used by a community from the intended referent of the category. Deprecation should record why a concept changed and which successor, if any, preserves its meaning. Mappings should carry dates, evidential support, and confidence where equivalence is uncertain. Provenance should identify the authority responsible for substantive claims. An ontology without a revision procedure gradually becomes an archive of accumulated, partially incompatible decisions.

The FAIR principles—findability, accessibility, interoperability, and reusability—provide a useful research-data framework, while their satisfaction demands context-sensitive implementation. An ontology may aid FAIR data through explicit vocabularies and resolvable identifiers, but publishing an RDF file alone does not establish that a dataset can be meaningfully reused. Reuse requires licensing, definitions, provenance, access conditions, versioning, and verified mappings (Wilkinson et al., 2016).

An adequate development method therefore connects conceptual analysis, formal testing, applied evaluation, and governance in a continuing cycle. The philosophical question remains present throughout: whether the distinctions that a system institutionalizes are warranted by the entities and practices it describes.

Biomedical and Scientific Ontology

Biomedicine displays both the promise and the difficulty of applied ontology. Researchers need to integrate clinical observations, anatomy, molecular processes, diseases, phenotypes, experimental interventions, and evidence produced by different instruments. If two databases use the same word 'gene' for different kinds of informational or material entity, exchanging records does not yield semantic interoperability. Likewise, a drug's chemical substance, its branded preparation, its prescription, and its administration event must remain distinct even when everyday discourse uses one name for several of them.

The Open Biological and Biomedical Ontologies Foundry, established through community collaboration, promotes coordinated development and shared principles for interoperable ontologies. The 2007 account by Smith and collaborators emphasizes coordinated biomedical resources and common formal organization (Smith et al., 2007). The project's significance lies in its combination of domain expertise, reference vocabularies, relations, and editorial procedures. No universal ontology alone can replace the empirical labor required to describe rapidly changing biological knowledge.

Anatomical parthood illustrates a recurring problem. A tissue may be part of an organ, and an organ part of a body, but the same relation cannot be applied indiscriminately across developmental phases, samples removed from an organism, and cellular substructures whose status changes over time. An ontology may need separate relations for proper part, located in, derived from, and participating in a process. Using a general relation in place of an exact one can generate transitive chains that misclassify cells, specimens, or procedures.

Disease representation exposes an even deeper distinction. A pathological process, disposition to disease, diagnostic category, diagnosis event, and clinical record each has a different identity and temporal structure. A patient can receive an incorrect diagnosis; the record exists even when its claim is false. Ontological modeling that merges diagnosis and disease loses the ability to represent error, disagreement, and revision. That loss is scientifically dangerous because it creates the appearance of certainty where the records provide evidence of judgments.

Scientific ontologies also encounter revisions in natural-kind classifications. Biological species concepts, molecular functions, and clinical syndromes can change through new research. An ontology must record historical versions and the evidential reasons for changed definitions. Its categories should support new observations without making earlier datasets silently appear to have adopted later meanings.

The success of biomedical ontology is therefore measured through both reasoner checks and scientific consequences: does an integrated query preserve specimen identity, temporal status, experimental method, and evidential provenance? Do the model's definitions support appropriate discoveries while preventing invalid inferences? These tests connect philosophical rigor to clinical and research accountability.

Social Institutions, Cultural Heritage, and Digital Artifacts

Applied ontology must also represent entities whose existence depends on social and normative practices. A university, citizenship status, contract, ownership right, artistic work, and official appointment are individuated partly through institutional rules and historically situated acts. Their properties may not be captured by physical composition alone. A piece of paper can be destroyed while a legal obligation persists; a digital file can be copied without multiplying the authored work that its copies instantiate. Such cases demand accounts of constitution, provenance, identity, and institutional authority.

Cultural heritage provides particularly demanding examples. A museum catalog may relate an artwork to an artist, production event, current physical carrier, ownership history, exhibition, restoration, and digital reproduction. The CIDOC Conceptual Reference Model, standardized as ISO 21127:2023, provides an established semantic framework for integrating cultural heritage information (CIDOC CRM). Its value lies in modeling events and relationships that connect diverse records, not simply in supplying a vocabulary of object types.

An artwork's ontology may involve several distinguishable entities: a creative work, a historically produced object, a representation of that object, and records concerning it. A digitized photograph of a painting is not numerically identical to the painting, even where both are discovered under a common catalog identifier. A conservation event changes a material object's condition and possibly its components; the identity question depends on established historical and curatorial criteria. Formalization must make those criteria explicit where inferences require them.

Legal classifications also expose normative dependence. A person may possess a legal status through a valid institutional act. Its recognition can vary across jurisdictions; a historical registry may record a status that a later authority contests. This makes context and provenance indispensable. An ontology should distinguish 'recognized by institution X at time t' from a context-free assertion whose truth every system must accept. Forcing plural jurisdictions into one undifferentiated classification would manufacture a consensus that the records do not support.

Digital artifacts add further difficulties. A software package may have versions, forks, dependencies, and releases. A document can have numerous byte-identical copies and remain one work in a chosen bibliographic theory, while a forensic system may need to individuate each storage instance. A hash identifies content under particular technical assumptions; it does not establish authorship or the metaphysical identity of an artifact across all purposes.

The general principle is that ontological commitments may be physically grounded, socially constituted, and technologically mediated. Effective applied ontology represents these relations explicitly, permits historical disagreement to be documented, and treats institutional assertions as attributable claims rather than timeless properties detached from authority.

Interoperability, Alignment, and Ontological Pluralism

Interoperability is often described as the capacity of different systems to exchange usable information. The challenge becomes philosophical when two systems partition the domain differently. One ontology may model a university as a legally constituted organization; another may model it primarily as a collection of campuses, departments, or educational activities. A direct equivalence assertion between these class expressions could erase their different criteria of identity. Interoperability requires an account of how the conceptualizations relate, not simply a mechanical alignment of names.

Three kinds of relation should be distinguished. Exact equivalence asserts that expressions have the same relevant denotation or logical interpretation. Subsumption asserts that every instance of one kind also belongs to a more general kind. A mediated correspondence can translate one perspective into another through relations, transformation rules, or contextual qualification. The third is often the most accurate solution because categories constructed for distinct purposes may overlap without matching. Mappings must be justified by the formal axioms and the domain evidence available.

Suppose a public-health registry classifies a case by reporting criteria, while a clinical ontology classifies disease processes by biological mechanism. Equating ReportableCase with Disease would conflate the act or status of reporting with the condition being reported. A better model distinguishes the disease, the patient's episode, the administrative reporting act, and the legal criteria for reportability. It then connects these entities through explicit relations. This mapping may be more complex than matching labels but preserves the meaning required for sound inference.

Ontological pluralism is sometimes interpreted as a concession that multiple incompatible conceptualizations are equally useful. Its stronger form is more precise: a domain may support several legitimate representational perspectives because their questions differ. The perspectives can be evaluated against their purposes, evidence, and declared semantics. Pluralism therefore demands comparison and translation rules. It cannot be reduced to the claim that any categorization is as good as any other.

A fundamental difficulty is that automated alignment systems are usually trained on incomplete lexical, structural, or statistical evidence. High label similarity may suggest a candidate mapping, not a proof of equivalence. False equivalences can have greater consequences than missing mappings, because they propagate assertions between systems. In biomedical and legal contexts, mapping governance should require review of high-impact correspondences, preservation of evidence, and rollback procedures.

Interoperability is achieved when multiple representational systems can preserve the distinctions necessary for the questions they jointly answer. Exact uniformity of categories is one possible means, but documented translation between appropriately distinct ontologies may better serve the same goal.

Knowledge Graphs, Artificial Intelligence, and Epistemic Accountability

Knowledge graphs make ontological relationships operational through networks of entities, identifiers, and assertions. Their architectures range from weakly typed property graphs to RDF resources governed by explicit ontology languages. Hogan and collaborators (2021) survey the field's data models, extraction methods, reasoning, and applications. The term knowledge graph does not guarantee logical consistency, scientific validation, or a particular account of what its nodes represent. Those attributes depend on the graph's construction and governance.

The connection to artificial intelligence intensifies the problem of meaning. Systems may extract relations from texts, align heterogeneous schemas, or generate candidate ontology classes and axioms. Language models can help identify recurring terminological patterns and formulate questions for expert review. But the textual frequency of a classification provides no independent warrant for its metaphysical truth. A model that reads a mistaken report may produce an elegantly structured false assertion. Inference over the resulting graph then preserves and amplifies the source error.

A useful architecture distinguishes claims from the things claimed about. A proposition that a certain artifact was created by a particular author should carry a source, date, confidence or evidential characterization, and method of attribution where relevant. Contradictory claims may coexist as recorded statements while the underlying event remains disputed. This avoids treating the contents of every imported document as equal in epistemic status. Provenance becomes an ontological and epistemological condition of responsible representation.

Neurosymbolic research connects learned representations with symbolic constraints and graph reasoning. Surveys identify opportunities to combine explicit relations with predictive pattern recognition while also documenting challenges of rule quality, scalability, explainability, and uncertain or incomplete information (DeLong, Fernández Mir, & Fleuriot, 2025; Hogan et al., 2021). In 2026, research on knowledge-graph-augmented language-model reasoning continues to address these problems, including inconsistencies between generated explanations and graph-supported conclusions (Wu et al., 2026).

An Artificial Era perspective may extend the investigation to the provenance of concepts and representations produced by nonbiological systems. Here the specific philosophical issue is responsibility for an artifact's asserted meaning: who or what selected its kinds, validated its relations, accepted its revisions, and placed it into institutional use? Answers require documentary history and governance procedures. They do not follow merely from using a generative model.

The appropriate epistemic standard is therefore traceable inference. An assertion should be traceable to its source or authority; a formal consequence to its axioms; an empirical classification to its observations and methods; and a revision to its documented decision. Ontologies can support new reasoning agents most reliably when they expose these layers rather than compress them into one opaque claim of 'knowledge.'

Critical Arguments and Open Research Problems

The first major philosophical challenge is the underdetermination of categories. The same observations can be described by multiple class systems. No inference engine can derive a unique, complete taxonomy from observations alone without assumptions about explanatory purpose, identity, and what differences matter. This underdetermination does not make ontology arbitrary. It places the justification of category selection in explicit philosophical arguments and domain testing. The task is to identify which distinctions support stable explanations and responsible decisions.

A second challenge concerns ontological imperialism: the imposition of one top-level ontology as the mandatory structure for every domain. Shared foundations can improve interoperability, but their categories are themselves philosophical commitments. When social, biological, and computational systems are forced into an unsuitable common scheme, conceptual information is lost. The response is to make foundational assumptions inspectable, specify the domain of their applicability, and use mediated correspondences where exact subsumption is unwarranted. This preserves the advantages of reuse while keeping conceptual difference visible.

A third challenge is the expressiveness–tractability tension. Some philosophical distinctions require relations or quantification that widely deployed description logics cannot conveniently express. Enriching the language may make certain reasoning tasks expensive or undecidable; restricting the language may conceal important distinctions. A principled architecture allocates different tasks to compatible instruments: description logic for suitable classification reasoning, more expressive logic for carefully delimited axiomatization, shapes for validation, rules for operational derivations, and application code for specialized processes. The essential condition is to disclose transitions between these instruments.

Fourth, the open-world–closed-world tension affects what absence means. Open-world semantic systems generally permit unknown facts; operational record systems may require mandatory fields and definitive completeness policies. Treating missing information as a negative fact can produce grave errors, while treating every missing value as permanently unknowable may prevent necessary operations. The answer lies in explicit completeness declarations with defined scopes, times, and institutional responsibility. Whether a conclusion is justified depends on the informational conditions under which it is drawn.

Fifth, ontology maintenance raises a problem of diachronic reference. Scientific theories change; social institutions revise statuses; terms are replaced or divided. When the meaning of a class changes, should its identifier remain the same? Stability supports data reuse, but unqualified stability can mislead users into treating past records as instances of a newly defined category. An adequate policy distinguishes changes of label, changes of definition, splits, mergers, and replacement concepts. Each carries a different claim about continuity of meaning.

Sixth, ontological engineering must confront normative power. A classification can determine access to healthcare, eligibility for services, legal recognition, or visibility in public archives. A system may be internally consistent and still encode biased or insufficiently supported categories. Its designers and maintainers should preserve the provenance of normative choices, test consequences for affected populations, and provide procedures for contesting and correcting classifications. The philosophical issue concerns what warrants the authority of formalized distinctions over real-world decisions.

These challenges expose an unresolved research frontier. We need methods that combine philosophical criticism of categories, formal proof, uncertainty representation, temporal change, cross-system translation, and accountable institutional governance without pretending that any single procedure establishes all dimensions of truth. Progress should be assessed against explicit examples where different assumptions lead to different conclusions and where the consequences can be traced through the system.

Philosophical Synthesis

Formal and applied ontology can be understood through four connected tests. The conceptual test asks whether the intended entities, categories, relations, and identity conditions have been adequately distinguished. The semantic test asks which interpretations satisfy the axioms and what follows under the chosen logic. The evidential test asks whether claims about a domain are supported by observation, authoritative records, or other relevant grounds. The governance test asks how those claims and categories can be corrected, maintained, and responsibly used. These tests arise from the research developed in this monograph; they constitute an analytic synthesis rather than a newly instituted canonical theory.

The four tests are ordered by dependence but are continually revisable. A conceptual distinction guides formalization; formal consequences can reveal a conceptual contradiction; empirical evidence can require the distinction to be revised; governance makes that revision legible to future users. An ontology that passes only the semantic test may be consistent and irrelevant. One that passes only the evidential test may contain accurate observations without a reliable framework for combining them. One that passes the conceptual and semantic tests but lacks governance may become obsolete while remaining apparently authoritative.

The historical development supports this synthesis. Aristotle's categories established the demand to analyze what predicates are said of and what kinds of beings are involved. Medieval philosophy investigated universals and signification. Husserl articulated formal structures of dependence and parts. Frege and twentieth-century logic supplied techniques for clarifying reference and inferential form. Quine required explicit accounting of theoretical commitments. Gruber and ontology engineering brought those questions into the practical problem of shared representation. Modern standards make the resulting commitments machine interpretable, yet the domain's legitimacy and truth remain additional questions.

A decisive consequence concerns the relation of ontology to data. Ontologies should distinguish things from assertions about things, universals from instances, roles from their bearers, part–whole relations from subclass inclusion, observations from phenomena, and interpretations from the evidence supporting them. When these distinctions are formally explicit, integration and inference become more trustworthy. When they are erased, scalable computation accelerates the consequences of conceptual mistakes.

The framework also clarifies pluralism. A single foundational ontology can coordinate a domain when its categories match the questions asked. Multiple ontologies may be necessary when different disciplinary perspectives legitimately classify the same world in different ways. The choice between these approaches should be made through semantic comparison, empirical adequacy, and the preservation of relevant differences. Interoperability requires sufficiently shared meaning to enable the intended inference; it does not require the elimination of every distinct perspective.

The philosophical result is a conception of formal ontology as public conceptual accountability. A formalized vocabulary becomes intellectually valuable when its definitions, axioms, evidence, permissible inferences, and revision history can be examined. Applied ontology realizes this value through systems that continue to distinguish the world, representations of the world, and the institutions responsible for interpreting those representations.

Conclusion

Formal and applied ontology occupies a durable intersection of metaphysics, logic, scientific classification, and the engineering of information. Its central problem is how to represent entity kinds and relations precisely enough for systematic inference while remaining answerable to the reality, evidence, and institutional practices of a particular domain. The history from Aristotelian categorical analysis through medieval semantics, Husserlian formal ontology, twentieth-century logic, and contemporary knowledge representation shows that this problem has developed through successive forms of the same underlying demand for conceptual discipline.

The research has established the philosophical significance of identity conditions, universals and instances, dependence, parthood, temporal organization, modal assumptions, interpretation, and the logical consequences of classification. Comparison of DOLCE, BFO, OntoClean, RDF, OWL, Common Logic, and SHACL demonstrates why no single representational device can perform the tasks of metaphysical analysis, logical reasoning, empirical verification, data validation, and governance. Each has a legitimate contribution, and its success requires an explicit account of its scope.

The enduring research questions concern disputed categories, changing scientific definitions, temporally qualified relations, meaningful alignment between heterogeneous systems, the evidential status of machine-produced assertions, and the authority of institutional classifications. These are active philosophical problems because ontology is now a medium through which knowledge is organized and decisions are made. A rigorous ontology makes its conceptual commitments inspectable, its inferential consequences testable, its empirical claims attributable, and its revisions historically intelligible.