Who Decides Where a Book Belongs?

A book about the history of mathematics could be placed with history or with mathematics. Either location has a reason behind it. Put the only copy in one place and the reader browsing the other place may never encounter it.

The shelf forces a decision. A catalogue can offer other routes, but someone still has to describe the subject and explain its relationship to neighbouring subjects. That work is an early, very practical version of a problem ontology teams face whenever a record fits more than one category.

Melvil Dewey’s original text is more candid about those decisions than the apparent neatness of his numbering system might suggest.

Dewey’s useful compromise

The first Dewey Decimal Classification appeared in 1876. Its numbers grouped books by subject and supported their arrangement and discovery. Dewey also supplied a subject index, so readers did not have to know the number before looking for a subject. OCLC’s publication timeline

In the original preface, Dewey explicitly describes sacrificing theoretical exactness to practical usefulness. He says that a history of mathematics belongs with mathematics because the subject, rather than the form of treatment, should determine its place. He also acknowledges that some topics have been fitted under headings to which they do not strictly belong. The number system was an arrangement for a working library, not a claim to have discovered the final structure of all knowledge. Dewey, A Classification and Subject Index, 1876

That admission matters. A tidy hierarchy often looks more inevitable than the decisions that produced it.

In a modern engineering catalogue, a report on corrosion in railway bridges might sit under materials, transport, or structures. If the system permits several subject relationships, it need not pretend that one of those concerns cancels the others. If an application requires one primary category, the team should explain what “primary” is for.

Perhaps the category routes the report to a reviewer. Perhaps it controls where the report appears in a menu. Those are different purposes, and they may justify different choices.

When a tree says too much

A hierarchy usually places narrower categories beneath broader ones. That can be useful when the relationship is genuinely “a kind of.” A suspension bridge is a kind of bridge.

A bridge deck, however, is part of a bridge. A bridge inspection is an activity concerning a bridge. The organization responsible for the bridge is neither a part nor a kind of it. A diagram can place all those labels beneath “Bridge,” but the visual arrangement alone does not tell a computer which relationship applies.

This is where a modern ontology must be more explicit than a navigation menu. If the model treats a deck as a kind of bridge, a query counting bridges could count components as whole structures. If it treats inspection as a component, a maintenance application may look for a physical location where none exists.

These are illustrative errors, but they show why a professional asks awkward questions about a perfectly attractive diagram. Does the line mean “is a kind of,” “is part of,” “is about,” or merely “appears underneath on this screen”?

A catalogue can be useful without being a formal ontology. Trouble begins when an application imports its arrangement and silently gives every line a stronger meaning.

The words also place people

Classification becomes more consequential when its categories describe people. A heading can help readers find material while also preserving assumptions that the people concerned reject.

In Prejudices and Antipathies, published in 1971, librarian Sanford Berman challenged Library of Congress subject headings concerning people. His criticism made the wording and social consequences of catalogue entries a subject of professional scrutiny. This was not an objection to organizing collections at all. It was an objection to treating inherited descriptions as beyond correction. Berman, Prejudices and Antipathies

For an ontology team, the corresponding work is specific. Who chose the label? Whose language does it use? Is a category describing how people identify themselves, how an institution classifies them, or what a historical source says about them? Those assertions should not be collapsed into one unqualified property.

An archive may need to preserve an offensive term because it appears in a source. The search interface does not therefore have to present that term as its preferred description of living people. A record of historical wording and a current display label can serve different purposes.

Changing a label is only part of the job. Existing searches, reports, links, and exported records may still depend on the earlier wording. The team needs a migration plan that preserves access and explains what changed. Where the concept itself changes, an old label cannot simply be attached to the new concept as though nothing happened.

Consistency leaves room for correction

Imagine a catalogue that originally placed “maintenance contractors” under “employees.” The mistake has since spread into a dashboard and an access-management tool. Correcting the hierarchy will affect counts and permissions as well as the catalogue.

A responsible change would identify the affected records, distinguish employment from contractual relationships, test the applications that use those relationships, and preserve the earlier version so that an old report remains interpretable. Agreement about the correction does not make its consequences disappear.

The practical principle from the library is therefore more demanding than “choose one term and stick to it.” People need consistent descriptions they can use, and they need a way to improve those descriptions when the existing arrangement misleads them.

The next part expands the scale. Once records describe material scattered across many institutions, local consistency is no longer enough. Paul Otlet and his collaborators tried to make a shared bibliographic system work across those boundaries, one card at a time.