From Controlled Vocabularies to Ontologies: Tracing the Evolution of Knowledge Representation

"Finally hearing about controlled vocabularies—this is the true starting point of ontology."

Controlled Vocabulary

A controlled vocabulary is a standardized term set maintained by a registration agency. It assigns a single preferred label to each concept and maps synonyms, near‑synonyms, and spelling variants to that label. Core features:

Term deduplication : one official label per concept.

No structure : no formal hierarchy or relationships.

Human‑centric : ensures linguistic consistency for human indexers and retrievers.

Under the NISO Z39.19 standard the vocabulary solves the problem of “everyone uses the same word”, but it does not address the question of how the words relate to each other.

Taxonomy

Building on a controlled vocabulary, a taxonomy adds a hierarchical structure (broader/narrower, BT/NT). Concepts are linked in a tree or directed‑acyclic graph, enabling vertical navigation such as “Mammal → Dog” or “Computer Science → Artificial Intelligence”. Taxonomies answer “A is a B” but leave cross‑branch relationships undefined (e.g., “Computer” and “Printer” may belong to different branches with no explicit link).

Thesaurus

Mid‑20th‑century information growth required richer connections, leading to the thesaurus. It introduces three standard relations:

Hierarchical (BT/NT) : broader/narrower, e.g., "Mammal" BT "Dog".

Equivalence (USE/UF) : synonym mapping, e.g., "Canine" USE "Dog".

Associative (RT) : non‑hierarchical semantic relatedness, e.g., "Dog" RT "Pet Food".

These relations weave a semantic network, allowing free navigation via rules such as “use, replace, belong, part, reference”. Thesauri have been successful in library and information science (e.g., AAT, TGN, Chinese subject heading lists) but remain human‑readable; the meaning of an RT link is ambiguous for machines.

Ontology

Gruber (1993) defines an ontology as “a formal, explicit specification of a shared conceptualization”. The definition emphasizes three attributes:

Formal : expressed with strict mathematical language (e.g., description logic).

Explicit : all assumptions and constraints are declared.

Shared : achieved through community consensus.

Compared with a thesaurus, ontology adds:

Formal semantics : each relation has a precise logical definition (e.g., part_of is transitive and antisymmetric).

Arbitrary‑granularity relations : domain‑specific predicates such as commander_of, causes, located_in.

Class‑instance distinction : separates concepts (classes) from individuals (instances); "Dog" is a class, "my golden retriever" is an instance.

Constraints and axioms : property domains, cardinality limits, mutual exclusivity (e.g., a person can have only one biological father).

Machine reasoning : inference over defined relations (e.g., if engine part_of car and piston part_of engine, then piston part_of car).

Open‑world assumption : facts not asserted are not assumed false.

Knowledge Graph

On top of an ontology, a knowledge graph populates the schema with massive instances and facts. It enables:

Complex SPARQL queries.

Linking to external Linked Data sources.

Cross‑domain knowledge fusion.

Dynamic updates and evolution.

Re‑engineering from SKOS to OWL

Many assume that converting a thesaurus (SKOS) to an ontology (OWL) is a simple extension. The W3C clarifies that SKOS is not a formal knowledge‑representation language; skos:broader lacks the strict logical semantics of owl:subClassOf. Therefore a full re‑engineering process is required:

Semantic clarification : replace vague RT links with precise predicates such as causes or located_in.

Logical constraint injection : add characteristics like transitivity, symmetry, functionality to relations.

Class‑instance separation : distinguish conceptual hierarchies from instance data.

Axiomatization : make implicit domain assumptions explicit logical axioms.

This redesign goes beyond format conversion; it restructures the knowledge model.

Pragmatic Starting Point

Building a high‑quality controlled vocabulary already demands extensive domain‑expert effort. Formalizing an ontology adds the need for logicians and deeper expert collaboration. Consequently, starting from an existing controlled vocabulary or professional thesaurus is the most cost‑effective entry point, leveraging decades of domain consensus and the W3C technology stack (SKOS → OWL mapping).

Semantic Maturity Pipeline

Controlled Vocabulary → Taxonomy → Thesaurus → Ontology → Knowledge Graph

Each layer adds expressive power while increasing construction cost and formalization requirements.

Appendix: Overview of the Maturity Levels

Controlled Vocabulary : flat list, no relationships, no formal semantics, target users are human indexers, standard NISO Z39.19.

Taxonomy : hierarchical tree/DAG, only broader/narrower relations, low formalization, human browsers.

Thesaurus : semantic network with hierarchical, equivalence, and associative relations, medium formalization, assisted human retrieval, standards ISO 25964, SKOS.

Ontology : formal graph with arbitrary relations, high formalization (OWL 2, RDF(S)), supports automatic machine reasoning, used by both machines and humans.

Knowledge Graph : large‑scale instance graph based on an ontology, high formalization (RDF, SPARQL), enables machine reasoning and query, serves machines and humans.