Thesaurus Standards for Taxonomies

Presentation Description

This talk, developed by Heather Hedden and Joseph Busch, discusses the similarities and differences of thesauri and taxonomies and the relevance of thesaurus standards for taxonomies. Then we provide an overview of the Simple Knowledge Organization System known as SKOS and explain how the revised ISO 25964 Part 1—the Thesaurus standard—aligns and compares with SKOS.

Date: April 15, 2026

Venue: Taxonomy Bootcamp London 2026

Location: London

Broad Topic: Taxonomy

Presenter: Joseph Busch, Principal Analyst, Taxonomy Strategies

Watch this Presentation

Download Presentation: MP4 | PDF

Download Slide Script: PDF

[Slide 1]

This is a brief talk about thesaurus standards for taxonomies developed by Heather Hedden and Joseph Busch.

[Slide 2]

We have been colleagues for quite a few years, and recently we have both participated as members of the ISO Technical Committee 46-Subcommittee 9-Working Group 8 for revision of ISO 25964 the thesaurus standard.

[Slide 3]

In this talk we will 1) discuss thesauri vs. taxonomies, 2) provide an introduction to standards used for thesauri and taxonomies, 3) talk about the relevance of thesaurus standards for taxonomies, and then we’ll tell you about 4) ISO 25964 Part 1 of the thesaurus standards and its recent revisions, 5) the Simple Knowledge Organization System known as SKOS, and 6) How ISO 25964 Part 1 aligns and compares with SKOS.

[Slide 4]

This diagram shows types of knowledge organization schemes or KOS on a continuum of simple to complex. Taxonomies and thesauri are located in the middle of this continuum with KOS like term lists (sometimes referred to as vocabularies) and authority files that list named entities such as people, organizations, places, events, topics and other types of named entities as more simple or less expressive and ontologies as more complex or more expressive. Note that the main distinction between taxonomies and thesauri strictly speaking is that thesauri have associative relationships as well as equivalent and broader and narrower relationships.

[Slide 5]

Some additional differences between taxonomies and thesauri include:

Thesauri support research as well as retrieval and are optimized for indexers to use when indexing content. Thus, thesauri are appropriately focused on concept scoping, disambiguation, and relationships to similar concepts (think Roget’s). The hierarchies in thesauri may not be particularly large and enumerated. They are better for multiple overlapping subject areas and very large vocabularies. The approach is term-centered and what terms are linked to or from it. Thesauri strictly follow ISO or ANSI/NISO thesaurus standards.
Taxonomies on the other hand support indexing content for search and retrieval and other uses and are optimized for end users browsing. Thus, all terms in a taxonomy belong to a limited number of major hierarchies (or facets). They support classification, categorization, concept organization (think the Linnaean taxonomy). They are better for subject areas of restricted scope and a restricted number of concepts. The approach is a top-down browse navigation from broader to narrower. Taxonomies do not as strictly follow ISO or ANSI/NISO thesaurus standards.

[Slide 6]

There are two basic types of standards – 1) There are specifications (measurements, protocols, coding, etc.) that support exchange and interoperability, and 2) there are design standards that support an expected user experience by different users’ types without training and support a high-quality outcome. For knowledge organization systems (KOS) specifications may also be called “data models.” Some examples of this type of specification are MARC, Dublin Core, RDF, RDFS, and SKOS. Some specifications may be used in combination for different types of data in the same resource, for example, Dublin Core and SKOS may be encoded using RDF (Resource Description Framework). KOS design standards for thesauri include ISO 25964 Thesauri and Interoperability with Other Vocabularies, and ANSI/NISO Z39.19 Guidelines for the Construction, Format, and Management of Monolingual Controlled Vocabularies.

[Slide 7]

Let’s take a little deeper look at these thesaurus standards. ISO 25964 Thesauri and Interoperability with Other Vocabularies has two parts. Part 1: ISO 25964-1: 2011 Thesauri for Information Retrieval was reviewed and confirmed in 2022. It was adopted by the British Standards Institute as BS ISO 25964-1. Revision work has been completed by Technical Committee 46, and the revision should be published in 2026. Part 2: ISO-25964-2: 2013 Interoperability with Other Vocabularies was reviewed and confirmed in 2023. It was adopted by the British Standards Institute as BS ISO 25964-2. Revision work has just recently been started.

[Slide 8]

ISO 25964 Part1 includes definitions, vocabulary control and thesaurus objectives, defines concepts and their scope notes, identifies thesaurus terms forms of entry, provides guidelines on when to include complex/compound concepts as terms or when to decompose them into simpler terms, defines equivalence/synonyms, defines hierarchical and associative relationships, discusses facet analysis, provides best practices for thesaurus construction and management, identifies thesaurus management software functional requirements, provides a standard data model in the unified modeling language (UML), discusses integration of thesauri with information applications, and identifies standard thesaurus data exchange formats and protocols.

Examples of ISO 25964 Part1 and ANSI/NISO Z39.19 design standards are:

Use nouns or noun phrases for concept/term labels.
Use plural for countable nouns
No duplicate concepts are allowed—Concept labels must be unique.
No relationship clashes are allowed. For example, a pair of concepts can be either hierarchically or associatively related to each other, but not both.
No circular relationships are allowed. For example, if concept A is narrower to concept B, and concept B is narrower to concept C, then concept C cannot be narrower to concept A.

[Slide 9]

Why use thesaurus standards for taxonomies? Data model standards exist for taxonomies (such as SKOS) but standards for design/quality do not. The distinctions between thesauri and taxonomies are often blurred. Both are increasingly built on the SKOS data model. Both tend to be built with the same software tools. Thesauri have more features than taxonomies do; taxonomies do not have features that thesauri lack (although the structural design tends to be different). The most important feature of taxonomies, hierarchical relationships, functions the same in both taxonomies and thesauri. Understanding how to implement hierarchical relationships correctly is very important.

ISO 25964 Part 1 and ANSI/NISO Z319.19 have extensive explanations and examples of types of acceptable hierarchical relationships. Inconsistencies may be tolerated by human users, but not by AI uses.

[Slide 10]

As mentioned, Heather and I have been part of the working group that has worked on revising ISO 25964 Part 1. These are the main changes in this revision. Overall, examples and references have been updated for currency and accuracy, and a significant number of multi-language examples have been added. To align more closely with SKOS the definitions and terms (clause 3) have been updated to add: concepts, candidate concept, lexical labels, sequence numbers, classification codes, GUIDs and UUIDs, information retrieval, information use, information management and mapping, etc. The equivalence relationships have been updated for both monolingual (clause 9) and multilingual (clause 10). Relationships between concepts (clause 11) has been updated.

Guidance on complex concepts (clause 8) has been updated. A section on persistent identifiers – PIDs (clause 13.5) has been added. The Data Model and identifiers guidance (clause 15) has been revised. Guidelines for thesaurus management software (Annex C) has been added.

[Slide 11]

This was my first experience participating in an ISO standard working group. Overall, I think that ISO 25964 Part 1 is an excellent thesaurus primer! However, I had hoped to make more changes to the standard, but it is difficult to change the scope of an ISO standard. An ISO standard needs to be consistent with other ISO standards in same domain and there are more than 900 ISO terminology-related standards! ISO standards are not responsive to fast moving technical areas like AI. The World Wide Web Consortium—W3C (and other communities of practice) are a more responsive. It is also challenging to identify and include examples in multiple languages (which are needed for a language-based standard).

[Slide 12]

We’ve talked a lot about SKOS the Simple Knowledge Organization System specification. But what is it exactly? It’s a W3C recommendation first published in 2009. SKOS is “a common data model for sharing and linking knowledge organization systems [or KOS] via the Web.” It’s a data model to represent various types of knowledge organization systems such as thesauri, taxonomies, classification schemes, subject heading schemes, terminologies, etc. A KOS specified according to SKOS is machine-readable and interchangeable. The SKOS specification is based on RDF (Resource Description Framework) and encoded in XML or JSON-LD.

[Slide 13]

This table is an overview of the SKOS specification elements grouped by type. I’ve highlighted the elements that correspond directly to the ISO 25964 Part 1 thesaurus standard. The labels in the ISO standard are intended to be human readable, while the elements in the SKOS specification are intended to be machine-readable. In the same way that the ISO thesaurus standard is being revised to align more readily with SKOS, a SKOS extension called SKOS-XL or SKOS eXtension for Labels makes it possible to readily extend beyond the initial SKOS elements.

[Slide 14]

How does ISO 25964 Part 1 align and compare with SKOS? The ISO 25964 Part 1 introduction states:

“In the revision of this document, the working group struggled with the use of preferred and non-preferred terms as the term of art for the main terms applied and their equivalence terms. “Concept label” was added to insure a clear cross walk with the W3C standards definitions and vocabularies. We have added the definitions for each area within the definition’s sections providing a cross walk to other major terminology standards including the ISO 704, ISO 30042 -TBX, and ISO 1087 terminology standards as well as the definitions from the W3C SKOS and OWL communities. Many other appellations are used for both labels. Preferred term is often called default term or default concept as well as concept label, and descriptor. Non-preferred term may be referred to as alternative term, alternative label, AltLabel,synonym, and others. This is not an exhaustive list for either but indicative. We encourage the community to use the term label best suited to their own community.”

In fact, the two standards are now more closely aligned in many critical ways. This include aligned definitions, use of concept identifiers and resource locators (URIs), similar definition of concepts and terms, similar relationships – BT / NT is the same as broader concept / narrower concept and RT is the same as related concept, a similar equivalence model with exact, close, broad, narrow, and related matching the same as exactMatch, closeMatch, broadMatch, narrowMatch, and relatedMatch. ISO 25964 has several features that SKOS lacks including: sub-types of hierarchical relationships for generic-specific, whole-part, and generic-instance; and conceptual mapping of thesauri to other types of vocabularies (this is largely in ISO 25964 Part 2). SKOS adds several features that ISO 25964 lacks including: collections and members (collections are containers to organize members), hidden labels as another kind of alternative label, documentation of example and change notes, a transitive super-property for inferencing transitiveBroader and transitiveNarrower, and the use of SKOS-XL for further alignment.

[Slide 15]

One further point of comparison between SKOS and the ISO thesaurus standard is concepts vs. terms. In the thesaurus standard model, there are two types of terms: 1) preferred terms and 2) nonpreferred terms. The preferred term is in an equivalence relationship with any number of nonpreferred terms. Content is indexed/tagged to preferred terms only. In the SKOS vocabulary model concepts have preferred labels and nonpreferred labels. This is not a relationship; just multiple labels for a concept. Content is indexed/tagged to concepts, as named by their preferred labels.

[Slide 16]

These are links to some key sources for the SKOS and thesaurus standards.