Thesaurus in Indexing: A Guide to Controlled Vocabulary and Concept Mapping

When you search a library catalogue or a research database for “heart attack” and the system also retrieves documents indexed under “myocardial infarction,” something invisible is working behind the scenes. That something is often a thesaurus. In Library and Information Science, a thesaurus is far more than a list of synonyms like the one bundled with a word processor. It is a carefully engineered tool that controls vocabulary, maps relationships between concepts, and ensures that indexers and searchers speak the same language. This guide explains what a thesaurus is, what it does, and how one is actually built.

What is a thesaurus? 🔗

In information retrieval, a thesaurus is a controlled vocabulary in which concepts are represented by terms that are organized so the relationships between them are made explicit. The international standard ISO 25964 defines it as a controlled and structured vocabulary where concepts are represented by terms, arranged so that relationships between concepts are clear, and where preferred terms are accompanied by entries that point from synonyms toward the chosen term.

The key word here is controlled. A controlled vocabulary mandates the use of predefined, preselected terms, unlike natural language, which allows anyone to describe a topic however they wish. Without this control, two indexers, or even the same indexer on different days, might describe the same subject using different words. This causes a mismatch between how a document is stored and how a user searches for it, which directly damages retrieval.

It helps to distinguish the two meanings of the word. The familiar thesaurus, such as Peter Mark Roget’s Thesaurus of English Words and Phrases from 1852, arranges words so writers can find synonyms. An information retrieval thesaurus has a different focus: it arranges concepts rather than mere words, and every authorized term, called a descriptor, carries a clearly marked meaning. The term itself was first used in this technical sense by Hans Peter Luhn in the 1950s, a period when several new approaches to subject indexing were emerging.

The building blocks: descriptors and non-descriptors 🔗

A thesaurus contains two kinds of terms. Preferred terms, also called descriptors, are the authorized terms actually used to index and search documents. Non-preferred terms, also called entry terms or lead-in terms, are synonyms or near-synonyms that are not used for indexing but instead point the user toward the correct descriptor.

Consider a vocabulary that chooses “frequency” as its descriptor. The synonym “pitch” would not be used for indexing. Instead, its entry would read “USE Frequency,” guiding both the indexer and the searcher to the single authorized term. This solves two classic problems of natural language. Synonyms scatter related material under different words, while homonyms (a word like “cancer,” which can mean a disease or a crab) collapse unrelated material under one word. Vocabulary control resolves both by assigning exactly one term to each concept and clearly separating concepts that happen to share a spelling.

Functions of a thesaurus 🔗

The central function of a thesaurus is to improve information retrieval by making the relationships among concepts explicit. A flat list of approved words is useful, but the power of a thesaurus comes from the structured links it draws between those words. These relationships fall into three standard types, each marked by a conventional notation.

Equivalence relationships 🔗

The equivalence relationship connects a preferred term to its synonyms and near-synonyms. The notation USE points from a non-preferred term to the descriptor, while UF (Used For) marks the reverse direction on the descriptor’s own record. For example, the descriptor “Frequency” would carry “UF Pitch,” and the entry for “Pitch” would carry “USE Frequency.” This relationship gathers all the different ways people might name a single concept and channels them into one access point, so that no relevant document is lost simply because the searcher chose a different word.

Hierarchical relationships 🔗

The hierarchical relationship organizes terms by levels of generality, and it is the primary feature that separates a thesaurus from simpler controlled vocabularies. A Broader Term (BT) sits one level above a descriptor and has a wider, less specific meaning, while a Narrower Term (NT) sits one level below and is more specific. The two are reciprocal: if “Computers” has the broader term “Apparatus,” then “Apparatus” must list “Computers” as a narrower term.

This hierarchy is not limited to one kind of relationship. ANSI/NISO Z39.19 recognizes three varieties: the generic relationship (a class and its species, such as “Bird NT Magpie”), the whole-part relationship (such as “Foot NT Toe”), and the instance relationship (a general category and a specific named example, such as “Seas NT Mediterranean Sea”). These links let a searcher who starts with a broad query move smoothly toward more specific material, or generalize a narrow search to capture more results.

Associative relationships 🔗

The associative relationship, marked RT (Related Term), connects two terms that are meaningfully related but are neither equivalent nor in a hierarchy. The relationship between “Cybernetics” and “Computers,” or between a disease and the virus that causes it, illustrates this. Related terms are often presented through “See Also” references that draw a searcher’s attention to material they might otherwise miss.

Defining these connections is a notably subjective task, and standards advise restraint. Because excessive related-term links reduce the precision of a search by pulling in loosely connected material, the relationship should be reserved for connections that are strongly implied. Beyond these three relationship types, a thesaurus also offers scope notes, which are short explanations clarifying how a particular term should be used, helping indexers apply terms consistently.

Constructing a thesaurus 🔗

Building a thesaurus is a disciplined process governed by established standards. The international standard ISO 25964, published in two parts in 2011 and 2013, is the current benchmark, and it builds on a long lineage including ISO 2788 and the American standard ANSI/NISO Z39.19. These standards ensure that thesauri are consistent enough to be shared, compared, and even mapped to one another. The work generally moves through the stages described below.

Defining scope and gathering terms 🔗

The first step is deciding the subject domain and the intended users, because both shape every later decision. A thesaurus for veterinarians and one for the general public might cover the same animals but choose very different preferred terms, such as “Felis catus” versus “Cats.” Terms are then gathered from the literature of the field, from existing indexes, from subject experts, and from the actual queries users submit. A wide net is cast at this stage so that important concepts are not overlooked.

Selecting terms and choosing descriptors 🔗

Once a pool of candidate terms exists, they are grouped by meaning. For each group, a single preferred term is chosen as the descriptor, often in consultation with subject specialists and after reviewing which terms appear most commonly in the literature. The remaining synonyms become non-preferred entry terms. Spelling variants matter here too; a popular variant should be retained as a lead-in term so that searchers using it are still guided to the right descriptor.

Establishing relationships and adding notes 🔗

With descriptors selected, the equivalence, hierarchical, and associative relationships are mapped between them, using the BT, NT, RT, USE, and UF notation. This is the intellectual heart of the work, because the quality of these links determines how well the finished tool guides retrieval. Facet analysis, the technique of breaking a subject into its fundamental categories, often underlies this structuring and has been incorporated into ISO 25964. Scope notes, history notes, and definitions are added to each descriptor record so that meaning is never ambiguous.

The role of computers 🔗

Modern thesaurus construction relies heavily on dedicated software. Manual thesaurus management is laborious because every relationship is reciprocal, so a change to one term must be reflected in its partner. Software enforces this automatically. The AGROVOC thesaurus maintained by the Food and Agriculture Organization of the United Nations offers a clear example. Its web-based maintenance interface allows editors to search and browse terms, add new ones, edit relationships, modify scope notes and definitions, and delete obsolete terms, all while keeping the underlying database consistent.

Computers have also transformed what a thesaurus can become. Many traditional thesauri are now expressed using the W3C standard SKOS (Simple Knowledge Organization System), which represents hierarchical links as skos:broader and skos:narrower and associative links as skos:related. This lets a thesaurus be published as linked open data, mapped to other vocabularies, and reused across the web. Real-world examples such as the Educational Resources Information Center (ERIC) Thesaurus, Medical Subject Headings (MeSH) used by the National Library of Medicine, and the Art and Architecture Thesaurus with its tens of thousands of concepts show how large and sophisticated these tools have grown.

Maintenance and updating 🔗

A thesaurus is never finished. Knowledge advances, new concepts appear, and old terms fall out of use, so a thesaurus must be continually revised to stay useful. New descriptors are added, outdated ones are flagged with history notes, and relationships are refined as understanding of the field deepens. Like all ISO standards, ISO 25964 itself is reviewed periodically, which reflects the same principle of ongoing revision at the level of the standards that govern these tools. A well-maintained thesaurus remains a living map of its subject, while a neglected one slowly loses its grip on the literature it was built to organize.

What do you think? If you were building a thesaurus for a fast-changing field like artificial intelligence, how would you decide which emerging terms deserve to become preferred descriptors? And do you think automated, machine-generated tagging will eventually replace carefully constructed thesauri, or will human-defined concept relationships always remain essential for precise retrieval?

References

1. https://en.wikipedia.org/wiki/Controlled_vocabulary
2. https://en.wikipedia.org/wiki/ISO_25964
3. https://marciazeng.metadataetc.org/Z3919/43hierarchy.htm
4. https://www.niso.org/standards-committees/iso-25964
5. https://www.researchgate.net/publication/267207310_Thesaurus_construction_guidelines_an_introduction_to_thesauri_and_guidelines_on_their_construction
6. https://aims.fao.org/publications/web-based-tool-manage-multilingual-thesauri-example-agrovoc
7. https://en.wikipedia.org/wiki/Art_%26_Architecture_Thesaurus