Which Book Do You Mean?

“We already have that book.”

Sometimes that settles the matter. Sometimes the speaker means a different translation, an earlier edition, or a copy that nobody can find.

Libraries have spent a long time making those differences manageable. The problem reaches back to the catalogues we encountered earlier in this series: bring related writings together without pretending that every version and every copy is interchangeable. Modern bibliographic models make the levels of identity explicit.

They offer useful company for anyone trying to untangle the latest approved policy from six files called “final.”

Four meanings inside one familiar word

The 1998 Functional Requirements for Bibliographic Records model, usually called FRBR, distinguished work, expression, manifestation, and item. The 2017 IFLA Library Reference Model, prepared by Pat Riva, Patrick Le Bœuf, and Maja Žumer, consolidated that bibliographic modelling tradition while retaining those four entities. It organized the model around what users need to do with bibliographic information. IFLA Library Reference Model

A work is the intellectual or artistic creation. An expression is a particular realization of it, such as its text in a particular language. A manifestation is an embodiment with shared production characteristics, such as a published edition. An item is an individual exemplar.

The terminology takes a moment. An ordinary library request makes its purpose clearer.

A reader wants Pride and Prejudice. Another wants it in Spanish. A researcher needs a particular publisher’s edition. A conservator is examining the copy with a damaged binding. These requests can concern related resources while requiring different answers.

The physical copy does not become a different literary work because somebody spills coffee on it. A translation does not become the same text as the English original merely because both realize the same work.

Those distinctions stop facts from landing in the wrong place. The stain belongs to the item. A translator’s contribution concerns a particular expression. A publisher and publication date concern the manifestation being described.

Why an office should care

Suppose a company maintains a safety procedure. The procedure has a revised text, an approved PDF, a printed copy beside a machine, and a translation used by another team.

If every object is simply “the procedure,” a search result cannot reliably tell a worker which one applies. If every file is treated as unrelated, the organization loses the history connecting them.

A model can distinguish the continuing procedure, its revisions, its language versions, its released files, and individual controlled copies. That does not mean copying the library model wholesale into an office. It means using the same discipline: decide which level of identity each fact describes.

The approval applies to a particular revision, possibly in a particular language. A file checksum identifies particular bytes. The location beside a machine describes a copy. The procedure’s subject may remain stable across all of those changes.

Once the distinctions are explicit, an update becomes more precise. Replace the released PDF without claiming that every printed copy has already been replaced. Mark an older revision as superseded without erasing the evidence that it governed work last year.

Anyone who has tried to investigate a decision made under an old procedure will recognize the value. “Current” is not an adequate description of the past.

Another model, another set of boundaries

The Library of Congress’s BIBFRAME initiative brings bibliographic description into a linked-data approach. BIBFRAME 2.0 uses three core levels: Work, Instance, and Item. Its Work includes conceptual information such as authors, languages, and subjects; Instance describes a particular published form; Item describes a particular copy. Library of Congress, BIBFRAME Manual

The names resemble those in the IFLA model, but the models do not divide the material identically. In particular, three BIBFRAME levels do not map one-to-one onto the four familiar IFLA entities.

That is an important lesson for ontology integration. Matching labels is not enough to match concepts.

Imagine joining two project systems that both have an “asset” class. One means a manufactured product design. The other means an installed unit with a serial number. A direct class equivalence would confuse a reusable design with each object made from it.

The mapping has to explain the difference and preserve the useful relationships. An installed unit can be an instance of a product model without being identical to that model. The apparently small word “of” is doing real work.

Start with the questions that fail

A useful ontology workshop can begin with the team’s recurring disagreements about sameness.

Are these two records duplicate descriptions of one report, or descriptions of two revisions? Does a changed filename create a new document? Does an altered technical conclusion? Can two people approve different files while believing they approved the same revision?

Work through actual examples before drawing the hierarchy. The point is to establish the identity rules the work requires, including the changes that preserve identity and those that create a new version or a different object.

Then try the model against retrieval. Can a reader find every translation of a work without receiving unrelated books on the same subject? Can a project engineer find the revision valid on a specified date without accidentally opening the latest one?

The library model earns its complexity through those questions. Each distinction should help someone find, identify, select, obtain, or explore what they need.

As we move into scientific datasets, the same discipline becomes harder to avoid. Researchers may use different words for the same biological function, or the same word for different measurements. Joining the records requires an account of what the records mean, not merely a column name that happens to match.