Is this a large language model?

No. Here model means a structured representation of domain concepts and meanings.

No. SKOS is one relevant representation for concept schemes; the required detail may call for another approach.

Can two terms name the same concept?

Yes, but apparent synonyms should be checked within the intended context rather than assumed universally interchangeable.

AI Glossary / Data engineering and architecture / Advanced

Terminology model

Definition

A terminology model is a structured description of the concepts used in a subject area, their meanings, characteristics and relationships. It helps subject specialists and system builders agree on what terms refer to before those meanings are embedded in data or software. The emphasis is on representing domain knowledge, not merely collecting words or training a language model to predict text.

What it means

A shared word can conceal different concepts. In a fictional education business, course might mean a reusable curriculum or a particular scheduled delivery of that curriculum. A terminology model makes that difference explicit and records how the concepts relate. Conversely, two labels might designate the same concept. The model therefore separates the name people use from the meaning they intend. For AI applications, that distinction can support clearer retrieval, questions and data mappings. It does not ensure that a model will interpret every sentence correctly, especially when context is missing or terminology changes.

How it works

Subject specialists identify relevant concepts, write definitions and review relationships and characteristics. R. Karge's terminology-model paper presents this as a bridge between expert knowledge and information-system development, with structured concepts and potentially rules. An implementation may assign identifiers and maintain labels separately. W3C's SKOS provides one vocabulary for representing concept schemes, labels and semantic relationships; it is not the only possible representation or a complete substitute for every detailed domain model. In a practical review, the team tests definitions against examples and counterexamples, resolves disagreements and records the scope in which the agreed meanings apply.

Three practical examples

An education team distinguishes a course definition from a scheduled course offering before designing enrollment records.
A search team records preferred and alternative labels so users can locate the same concept through different familiar expressions.
A data-integration team checks whether two departments' uses of active customer describe the same population before mapping their fields together.

Think of it this way

Imagine preparing the legend for a complex map before several teams add routes. The legend explains what each symbol means and how different kinds of places relate. Simply spelling every symbol's name correctly would not prevent someone confusing a station with an entire railway line. A terminology model similarly makes distinctions explicit so that people can discuss the same subject without silently changing its meaning.

Worked example

Consider an original miniature model with three concepts: Course, Offering and Enrollment. Define Course as a reusable learning specification, Offering as a scheduled delivery of one Course, and Enrollment as a learner's registration for one Offering. Suppose Course C1 has offerings O1 and O2. Learner L1 registers for both, producing enrollments E1 and E2. Under these definitions, the records describe one course, two offerings and two enrollments for one learner. Counting two enrollment records as two distinct courses would be wrong. Add the alternative label class for Offering only within this fictional organization's terminology. A document using class elsewhere still needs contextual interpretation. The model clarifies the counting units; it does not by itself implement a database constraint or prove that all imported records satisfy the definitions.

Illustrative case study

In a fictional AI reporting project, managers ask how many courses ran last quarter. One department counts curriculum titles, while another counts scheduled deliveries. Both reports appear plausible but answer different questions. A terminology workshop introduces separate Course and Offering concepts and documents the relationship. The assistant's revised interface asks which measure is intended when the wording is ambiguous. The team evaluates it using questions that deliberately mix the old labels and records cases requiring clarification. This scenario illustrates how agreed terminology can expose ambiguity before it becomes a misleading answer. It does not claim that vocabulary changes alone repair every reporting or model-quality problem.

Common misconceptions and mistakes

A terminology model is more than an alphabetical list with brief descriptions: relationships and conceptual distinctions matter. It is also not automatically an executable database schema. Another mistake is treating matching labels as proof that concepts are equivalent. Different departments may use the same phrase under different inclusion rules. Conversely, different words can refer to the same concept. Avoid making every association a broader-or-narrower relationship; being part of something, being a kind of something and being associated with something are different ideas. Finally, a well-presented diagram can still encode disputed definitions that require domain review.

Business use

For a proposed glossary or AI knowledge service, choose a small domain and test whether its definitions support real questions. In the fictional course example, reviewers could compare counts of courses, offerings, enrollments and learners and explain each result. Record who can approve a changed definition and how dependent reports learn about it. Preserve previous meanings when historical interpretation matters. These activities make terminology useful in everyday decisions rather than merely adding another reference document. Evaluate improvement through fewer ambiguous mappings or clearer answers on representative cases, without assuming that a larger vocabulary necessarily produces better results.

Check your understanding

How many courses appear in the miniature example?

Answer: One.

How many offerings and enrollments appear?

Answer: Two of each.

Why can the same word cause integration problems?

Answer: Its intended concept or inclusion rule may differ across contexts.

Why did the fictional assistant ask a clarification question?

Answer: The user's wording did not distinguish curriculum titles from scheduled deliveries.

Related terms

Controlled vocabulary
Ontology
Concept model
semantic relationship
data dictionary
SKOS

Learn next

Choose five terms from a familiar business process. Write definitions and identify one pair that is easily confused. Add a relationship between two concepts and explain whether it expresses a kind, a part or another association. Test the definitions against three concrete examples, then revise any wording that leaves the examples ambiguous.

Sources

R. Karge, A terminology model approach for defining and managing statistical metadata: https://www.odbms.org/wp-content/uploads/2013/11/P2_TerminologyModel_v1.pdf W3C, SKOS Simple Knowledge Organization System Primer: https://www.w3.org/TR/skos-primer/ The course model, map analogy and AI-reporting scenario are original educational illustrations. SKOS is presented as a related representation, not as the mandatory implementation of every terminology model.

Previous: Termination condition Next: Ternary search