When the Catalogue Became Machine-Readable

The computer could read the record. That was progress. Whether it could tell an author from a publisher depended on what the people preparing the record had put there.

By the 1960s, libraries already had elaborate ways of describing their collections. Names, titles, editions, subjects, and shelf locations were not an undifferentiated pile of words. A catalogue card carried distinctions developed through generations of practice. Moving it into a computer meant deciding which of those distinctions the machine would need.

Henriette Avram worked on that problem at the Library of Congress. The resulting MARC format, short for Machine-Readable Cataloging, made bibliographic records exchangeable between computer systems. The MARC pilot ran from 1966 to 1968 and eventually involved twenty libraries. Avram’s final report recorded a practical experiment in distributing usable descriptions, with institutions testing what they received. Library of Congress, Avram MARC Development Collection

It is a good place to pause in this history. Much of what gets called an ontology problem today first arrives on somebody’s desk as a migration problem.

What survives the move?

Consider a catalogue card containing a person’s name twice: once because they wrote the book, and once because the book discusses them. A human reader may recognize the difference from the arrangement and punctuation. A computer needs an explicit indication of what each occurrence does.

MARC uses numbered fields, indicators, and subfields to distinguish kinds of content and details within them. In MARC 21, a title occupies a different field from a topical subject heading. A name used as a subject is distinguished from a name used as an author access point. These markers let software treat the entries differently. Library of Congress, Understanding MARC Bibliographic

The format did not make cataloguing judgment disappear. It gave the result of that judgment a structure that software could carry and use.

That distinction is easy to miss during a modern platform replacement. A supplier says it can import every record. The import completes. The record count matches. Everyone is relieved until someone searches for reports written by a particular engineer and gets reports that merely mention her.

The words survived. Their roles did not.

For an ontology practitioner, a useful migration test follows the meaning through the move. Take one record with several contributors, a revised title, a publication date, and a subject that resembles a person’s name. Can the receiving system still answer the questions the old system answered? Can it distinguish who did the work from whom the work concerns?

Counting rows helps catch omissions. It does not settle those questions.

The benefit was shared work

Frederick Kilgour’s work with OCLC brought another part of the opportunity into view. Libraries could reuse one another’s cataloguing instead of describing the same publication repeatedly.

On August 26, 1971, Ohio University’s Alden Library entered the first record in OCLC’s online shared cataloguing system. Milton Hodnette entered a children’s book, The Rand-McNally Book of Favorite Pastimes. The early system still produced physical catalogue cards. Online cooperation and paper drawers overlapped rather than changing places overnight. Ohio University Libraries’ account of the 1971 launch

I like the ordinariness of that first book. The lasting change did not require an extraordinary object. It required a description that another library could use.

There is a familiar professional equivalent. Several project teams buy the same piece of equipment. Each team creates a supplier name, a product description, and a set of classifications. A shared product record could save repeated work, but only if everyone agrees about what belongs in it.

The manufacturer’s model number may be shared. The location of a particular installed unit cannot be shared in the same way. Neither can the date that one team inspected its unit.

A reusable record needs a boundary. Put local facts into the common description and one team’s update can become another team’s false information. Keep every fact local and the organization pays to describe the same product again and again.

Library cooperation makes that tension unusually easy to see: many libraries can describe the same edition while owning different copies.

A format is an agreement with limits

MARC is an encoding format, not a complete formal ontology. Cataloguing rules determine much of the description; controlled vocabularies govern many names and subjects; the format carries the resulting values and their roles. Treating all three as one standard conceals where a disagreement actually lives.

Suppose two departments exchange an “owner” field. The software reads both perfectly. One department means the legal owner of an asset; the other means the employee responsible for its maintenance. They have achieved technical exchange without achieving agreement about meaning.

The repair belongs in the model and the mapping. Give legal ownership and maintenance responsibility separate relationships. Explain which source supplies each one. Then make the conversion preserve that distinction.

Sometimes the old record simply does not say which meaning was intended. A conscientious migration retains that uncertainty or sends the record for review. Silently choosing the more convenient interpretation makes the new system look tidier at the expense of its usefulness.

Avram’s achievement belongs in this story because knowledge has to travel between actual systems, with their formats, conventions, and awkward exceptions. An elegant model that cannot survive that journey is of limited help to a working librarian or an engineer trying to find a document.

By this point, computers could receive descriptions and support shared cataloguing. AI researchers were asking a further question: could a machine use explicit relationships to work out something that had not been entered as a complete answer? That question would demand more from the meaning of every connection.