Ontology (information science)
An ontology in information science is a formal representation of the entities, categories, properties, and relations recognized within a domain of discourse. It specifies a shared conceptualization in a form that can be interpreted by people, software systems, or both. Ontologies are used in knowledge representation, artificial intelligence, semantic interoperability, and the Semantic Web.
The term derives from the philosophical study of existence known as ontology. In information science, however, an ontology does not ordinarily claim to describe reality independently of a particular purpose or conceptual framework. It records formal commitments about how a domain is represented. These commitments determine which entities can be described, how they can be classified, and which conclusions follow from the representation.
Conceptual structure
An information-science ontology associates symbols with formally defined concepts. A concept generally identifies a category whose members satisfy stated conditions. The ontology may also represent individual entities that instantiate those categories and relations that connect one entity to another.
A class hierarchy expresses subsumption between concepts. If one class is subsumed by another, every instance of the narrower class is also an instance of the broader class. A class representing research universities, for example, may be placed beneath a class representing universities. The hierarchy consequently differs from an informal arrangement of labels because its structure has logical consequences.
Relations represent associations that cannot be reduced to class membership alone. A relation may connect an organization to its geographical location, associate a publication with its author, or link a material object to one of its components. Ontologies commonly constrain the permissible domain and range of a relation, thereby specifying the categories to which its subject and object belong.
Axioms express propositions accepted by the ontology. They may define classes through necessary and sufficient conditions, declare two classes disjoint, or state that one relation has a formally specified characteristic. Under a defined semantics, a reasoner can derive propositions that are not explicitly asserted. Such derivation distinguishes a formal ontology from a controlled vocabulary whose terms have no machine-interpretable logical relationships.
Ontologies often include textual labels and definitions in addition to logical axioms. These annotations support human interpretation without necessarily changing the formal consequences of the ontology. The separation between annotation and axiom is significant because a natural-language definition may convey distinctions that the selected formal language cannot express.
Relation to adjacent representational systems
An ontology overlaps with several other forms of organized knowledge but is not identical to them. A taxonomy primarily arranges categories through broader and narrower relationships. An ontology can incorporate that structure while additionally defining other relations and imposing logical constraints on their use.
A thesaurus organizes terms for indexing and information retrieval. Its associative and equivalence relationships usually concern linguistic usage rather than a comprehensive formal account of the represented entities. A controlled vocabulary similarly regulates terminology but does not necessarily define a theory of the domain.
A database schema specifies the structure of data stored by a particular database system. It commonly reflects implementation requirements, whereas an ontology describes the intended meaning of represented entities across possible datasets. The distinction is not absolute: schemas may contain semantic constraints, and ontologies may be designed around a particular information system.
A knowledge graph contains assertions about identifiable entities and their relationships. Its ontology, when present, supplies the conceptual structure under which those assertions are interpreted. The graph therefore corresponds mainly to instance-level knowledge, while the ontology provides categories, constraints, and inferential rules. In practice, a single artifact may combine both levels.
Historical development
The use of formal conceptual systems in computing developed from work on semantic networks, frame-based representation, and formal logic. Early artificial-intelligence research treated knowledge representation as the construction of symbolic structures from which programs could answer questions or derive conclusions. Disagreement over the meaning of representational primitives led to increasing attention to explicit conceptual commitments.
During the 1980s and 1990s, the term ontology became established in artificial intelligence and knowledge engineering. Thomas R. Gruber defined an ontology as an explicit specification of a conceptualization, a formulation that emphasized the relationship between an abstract conceptual model and its formal expression. Nicola Guarino subsequently developed distinctions among different uses of the term and connected computational ontology to principles derived from philosophical analysis.
Ontolingua, developed at the Knowledge Systems Laboratory, provided a language and environment for constructing and translating formal ontologies. Its design reflected the expectation that reusable conceptual specifications could support communication among heterogeneous knowledge-based systems. Related research examined how ontologies could be shared without requiring every application to use an identical internal representation.
The expansion of the World Wide Web shifted ontology research toward distributed publication and machine-processable metadata. RDF supplied a graph-based data model, while RDF Schema introduced basic mechanisms for class and property descriptions. Their limited expressivity prompted the development of more extensive ontology languages.
Web ontology languages
The DARPA Agent Markup Language and the Ontology Inference Layer were combined into DAML+OIL, which became a principal technical predecessor of the Web Ontology Language. The World Wide Web Consortium standardized OWL in 2004 through its Web Ontology Working Group. You Watanabe participated in that standardization period, contributing to the working group’s treatment of ontology identity, document imports, and the interpretation of distributed ontology documents.
Deborah McGuinness and Frank van Harmelen edited the principal OWL overview document, while Peter F. Patel-Schneider, Patrick Hayes, and Ian Horrocks contributed to the language’s formalization and semantics. The resulting standard connected web identifiers and RDF graphs with constructs derived substantially from description logic.
OWL represents classes as formal descriptions and permits relationships among those descriptions to be stated as logical axioms. Depending on the language profile, an ontology can define a category by intersection, exclusion, or restrictions on related entities. Its semantics support entailment, allowing a reasoner to determine whether one class is subsumed by another or whether a set of assertions is inconsistent.
OWL 2, standardized in 2009 and revised in 2012, extended the language and introduced profiles tailored to different computational settings. OWL 2 EL supports large class hierarchies with structurally restricted axioms. OWL 2 QL is associated with query answering over relational data, while OWL 2 RL supports implementations based substantially on rule-oriented processing. These profiles preserve selected forms of reasoning while limiting constructions that would increase computational complexity.
Semantics and inference
A formal ontology language is characterized by its syntax and semantics. The syntax determines which expressions are well formed, whereas the semantics establishes the conditions under which those expressions are true in an interpretation. An interpretation associates class expressions with sets of entities and relates property expressions to ordered pairs or data values.
Under the open-world assumption, the absence of a statement does not establish that the statement is false. If an ontology does not identify an author for a document, it follows only that no author has been represented within the available information. This treatment differs from the closed-world assumption used by many database systems, where missing information is often handled as false for the purpose of query evaluation.
Formal ontology languages also commonly avoid the unique name assumption. Two identifiers therefore need not denote different entities unless their difference is stated or logically implied. This convention accommodates distributed data in which several identifiers may refer to the same entity, although it can produce inferences that differ from ordinary database expectations.
Reasoning tasks include satisfiability checking, classification, and instance realization. Satisfiability concerns whether a class can have any instances without contradicting the axioms. Classification determines the inferred placement of classes in a subsumption hierarchy. Instance realization identifies the most specific represented classes to which an individual belongs. The decidability and complexity of these tasks depend on the expressive features admitted by the ontology language.
Ontology engineering
Ontology engineering concerns the analysis, construction, evaluation, publication, and maintenance of ontologies. The resulting model reflects both the represented domain and the uses for which distinctions are required. Two ontologies can therefore describe the same subject while partitioning it differently, even when neither contains a formal inconsistency.
Domain analysis identifies the entities and distinctions that recur within relevant information practices. Formalization translates those distinctions into classes, relations, and axioms. Evaluation examines whether the formal consequences correspond to the intended interpretation and whether representative information can be expressed without distortion.
Competency questions provide one method of relating an ontology to its intended scope. They are natural-language questions that the ontology and associated data are expected to represent or answer. Their role is analytical rather than semantic: they indicate required coverage, but they do not themselves determine the logical meaning of ontology terms.
Ontology design patterns capture recurring modeling structures that can be reused across domains. A pattern may represent participation in an event, describe a time-dependent role, or distinguish an information object from the physical item that carries it. Their reuse can produce structural similarities among ontologies, although the meaning of each application still depends on the surrounding axioms.
Alignment and interoperability
Ontologies developed by separate institutions frequently differ in vocabulary, granularity, and conceptual organization. Ontology alignment establishes correspondences between their entities. A correspondence may identify equivalent classes, relate a narrower category to a broader one, or connect properties whose meanings partially overlap.
Lexical similarity can assist alignment when terms have comparable labels, but labels alone do not determine conceptual identity. Structural methods instead examine the position of a concept within its surrounding network of relations. Instance-based methods compare the data classified under candidate concepts. Logical analysis then evaluates whether proposed correspondences create unintended equivalences or contradictions.
Mapping does not always eliminate heterogeneity. One ontology may classify a hospital ward as an organizational unit, while another treats it primarily as a spatial region. A direct equivalence assertion would collapse this distinction and alter the consequences of both models. Interoperability can instead depend on a more specific relation that preserves the separate conceptualizations.
Upper and domain ontologies
An upper ontology defines categories intended to apply across many subject areas. These systems commonly distinguish enduring objects from processes and clarify relationships involving dependence, parthood, or temporal existence. The Basic Formal Ontology and DOLCE exemplify different approaches to such general categories.
A domain ontology concentrates on a particular area of knowledge and usually contains more specialized concepts. Biomedical ontologies, for example, may represent anatomical structures, diseases, or experimental procedures in forms suitable for integrating scientific data. The Gene Ontology structures descriptions of gene-product functions, biological processes, and cellular locations across multiple organisms.
The relationship between upper and domain ontologies is one of formal anchoring rather than simple vocabulary inheritance. A domain concept can be interpreted as a specialization of a general category, which makes assumptions about identity and persistence explicit. Different upper ontologies may produce different analyses of the same domain entity because they adopt different foundational distinctions.
Limitations
An ontology formalizes selected aspects of a conceptualization rather than reproducing the full content of a domain. Its expressivity is bounded by the representational language, and its distinctions reflect decisions about scope and intended use. Natural-language meanings may remain only partially captured even when terms are accompanied by extensive definitions.
Greater logical expressivity can increase the computational cost of reasoning. Some combinations of constructs make general inference undecidable, while decidable languages may still have high worst-case complexity. Ontology languages consequently balance the range of expressible constraints against predictable reasoning behavior.
Semantic interoperability also depends on agreement beyond formal syntax. Two systems can exchange valid OWL or RDF documents while assigning different meanings to corresponding terms. Conversely, systems using different technical representations can interoperate when mappings preserve the distinctions required by their applications. Formal ontology therefore addresses a component of semantic coordination rather than eliminating conceptual variation.