Wednesday, January 20, 2021

Hierarchies in Taxonomies, Thesauri, Ontologies, and Beyond

Hierarchies are a defining feature of taxonomies, and they are also characteristic of other controlled vocabularies or knowledge organization systems, such as classification schemes, thesauri, and ontologies. The problem is that the definitions and rules for hierarchies vary depending on the kind of knowledge organization system, so you cannot assume that a hierarchy in one system converts to a hierarchy in another system.

“Hierarchy” can have various types and uses. Not all kinds of hierarchies are reflected in even in taxonomies, which tend to be quite flexible. The rules are stricter when it comes to thesauri. Finally, in ontologies, there is only one kind of hierarchy.

The hierarchies permitted in thesauri are specified in the ANSI/NISO Z39.19 and ISO 25964-1 standards, as a reciprocal inverse relationship pair of Broader term (BT) / Narrower term (NT). There are three kinds specified in these standards:

  • Generic-specific   which refers to “is a” or “are a kind of”
        Basketball is a kind of sport.
        Basketball BT Sports;  Sports NT Basketball
        Baketball has broader concept Sports; Sports has narrower concept Basketball
  • Generic-instance – which refers to “is a named entity instance of”
      Michael Jordan is a named basketball player.
      Jordan, Michael BT Basketball players; Basketball players NT Jordan, Michael
      Jordan, Michael has broader concept Basketball players; Basketball players has narrower concept Jordan, Michael
  • Whole-part – which refers to “is in” or “is an integral part or component of”
       (not to be confused with “part” as a participant taking part in, or member of)
       Locker rooms are in athletic facilities.
       Locker rooms BT Athletic facilities; Athletic facilities NT Locker rooms
       Locker rooms has broader concept Athletic facilities; Athletic facilities has narrower concept Locker rooms

The types of hierarchies permitted in taxonomies include all of those designated for thesauri, plus a little more flexibility due to the absence of the associative relationships. In thesauri, if the relationship between a pair of concepts is better described as associative (“Related term” - RT) than hierarchical, then they cannot be hierarchically related. In a taxonomy which lacks associative relationships, in some cases a relationship that is not accepted as hierarchical in a thesaurus may be accepted as hierarchical in a taxonomy. An example is the pair of concepts Stress and Stress management. Technically, the relationship between these two concepts is associative and not hierarchical, because Stress management is not a kind of or a part of Stress. But in a taxonomy (not a thesaurus), designating Stress management as a narrower concept of Stress may be acceptable.

As for classification schemes, despite their name, they do not always conform to class-subclass (as "is a kind of") conventions. For example, in the Dewey Decimal Classification system, 910 Geography & travel comes under 900 History. But geography and travel are not kinds of/sub-categories of history. Classification schemes may have a tendency to force a hierarchy when it’s not really an accepted taxonomic hierarchy.

Despite the looser rules for hierarchies of taxonomies and classification schemes, there are also kinds of hierarchies that are not taxonomic hierarchical. These include organizational chart hierarchies, hierarchies of (military) rank, family tree hierarchies, the ordering of social sciences concepts of as Maslow’s hierarchy of needs, or Bloom’s Taxonomy of learning objectives. The hierarchies in these cases are not broader/narrower, but rather reflect importance, influence, sequence, or some other aspect of the notion of hierarchical order. In taxonomies and thesauri, concepts in such organizational hierarchies need to be treated instead as siblings at the same level all sharing the same broader concept, such as Learning objectives as the single broader concept for all six of Bloom's learning objectives, Needs as the single broader concept for all five of Maslow's needs, Military ranks as the single broader concept for all ranks, or Job titles as the single broader concept for all job titles.

Finally, in ontologies, hierarchies may be of less significance, but they are still a feature. While relations between concepts/entities are “semantic,” with specific descriptive labels, and thus are not necessarily hierarchical, there are may be hierarchical relations between classes, when designating subclasses of classes. However, the kind of hierarchical relationship that is created between ontology classes and subclasses is limited strictly to the generic-specific type, for “is a kind of.”


These distinctions in hierarchies have ramifications if you want to combine, import, or convert one knowledge organization system to another. When converting a thesaurus to a taxonomy, it is possible that some of the associative relationships could be accepted as hierarchical. When converting a taxonomy to a thesaurus, existing hierarchical relationships should be reviewed to see if any should be converted to associative.

Converting a taxonomy or thesaurus to an ontology would require identifying and remove whole-part hierarchical relationships (and adding new broader concept relations to the orphaned concepts) and converting generic-instance hierarchical relationships to class-individual relationships rather than class-subclass. In fact, this may involve so much effort, which cannot be automated, that the better approach to converting a taxonomy to an ontology is probably to apply a more generic ontology as a layer to the existing taxonomy/thesaurus, which some software tools, such as PoolParty, support. Extending a taxonomy into an ontology is the subject of my next conference presentation “Ontology Design by Enriching Taxonomies” at the Data-Centric Architecture Forum on February 3.

No comments:

Post a Comment