Before a Computer Could Use an Ontology

Someone asks for a book about bridges. A colleague hands over a book that happens to have a bridge on its cover. The request was ordinary. The mistake is ordinary too. A picture of a thing, a book about that thing, and the thing itself have been treated as though they were interchangeable.

People who build ontologies spend a surprising amount of time untangling mistakes like this. An ontology sets out the kinds of things a system deals with, what their names mean, and how they relate. Before any software can use those distinctions, someone has to make them carefully enough that other people can understand them.

I want to begin this series well before computers, with the older work of distinguishing and describing. Library science did not invent philosophical ontology, and a library catalogue is not automatically a formal computer ontology. But librarians had to make knowledge findable under conditions that expose many of the same problems. Their solutions deserve more than a historical footnote.

A thing and the kind of thing it is

In the fourth century BCE, Aristotle examined distinctions among particular things, the kinds to which they belong, their qualities, and their relations. His Categories distinguishes an individual person or horse from the broader kinds under which each can be described. It also distinguishes a thing from something said about it, such as a quality. These are philosophical distinctions, not instructions for a library catalogue. They nevertheless give a precise early statement of a problem that remains familiar. Aristotle, Categories, translated by E. M. Edghill

Imagine an equipment register. “Pump” names a kind of equipment. “Pump P-17” identifies a particular pump. “Red” describes a quality. “Installed in Building C” records a relationship to another particular thing.

Put all four into one list of interchangeable tags and a person may still work out what happened. Software has less room for that kind of kindness. A search for equipment types might return a building, a colour, and a serial number unless the system has some way to distinguish their roles.

This is a present-day example, not evidence that Aristotle designed asset management. The continuity lies in the question. What kind of statement are you making when you attach a name to something? An ontology practitioner needs an answer because different statements support different operations.

A pump can move to another building without becoming another kind of equipment. Its paint can change without changing its identity. If a model cannot express those possibilities separately, a routine maintenance update may force someone to rewrite far more than they intended.

A catalogue reaches beyond the shelf

By the sixteenth century, printed scholarship presented another problem. A reader could not discover a work merely by inspecting the books in one room.

Conrad Gessner’s Bibliotheca universalis, published in Zürich in 1545, recorded writings in Latin, Greek, and Hebrew. Its title alone tells us something important about its boundaries. “Universal” did not mean every language, every community, or everything people knew. The 1548 Pandectae offered a subject-organized approach to this bibliographic enterprise. Readers could approach writings through their subject as well as their authors. The 1545 edition in the Universal Short Title Catalogue; Gessner’s digitized 1548 Pandectae

A bibliography describes writings whether or not one particular library owns a copy. That distinction is useful today. A research catalogue may list a report, while an access system says where a copy can be obtained. The report’s identity and the availability of a particular file answer different questions.

Suppose an engineering firm has three copies of the same report: one in an archive, one in a project folder, and one attached to an email. Treating every file as a different intellectual work exaggerates what the firm knows. Treating them as one physical object makes it impossible to track which copy carries an annotation or which file was corrupted.

The problem has changed its materials. It has not become simpler.

The limits belong in the description

Gessner’s language boundary also provides a useful way to examine a modern ontology. What does its apparently general name leave out?

An organization may have a model called “People.” In practice, it may contain only current employees in one country’s payroll system. Contractors, former staff, external authors, and people whose names use another script may be absent. Nothing is necessarily wrong with a payroll model. Trouble starts when someone assumes its scope is the same as its title.

A useful description would make the coverage explicit. It would also say what inclusion means. Is a person listed because they are employed, have an account, contributed to a project, or appear in a source document? Those conditions overlap, but they do not identify the same population.

Modern ontology development makes scope a practical starting point. Natalya Noy and Deborah McGuinness ask builders to establish the domain, intended use, and questions the ontology should answer before committing to a class structure. The aim is a useful model for a purpose, not a complete inventory of existence. Noy and McGuinness, Ontology Development 101

The early history therefore leaves a very usable set of concerns. Keep the object separate from its description. Distinguish the particular thing from its kind and qualities. Explain the limits of the collection. Give readers more than one sensible route into it.

None of those concerns requires a graph database. All of them become expensive to recover after a database has confused them.

The next difficulty appears as soon as the catalogue grows. One person may publish under several names, and several people may share a name. Before a catalogue can bring their work together, its makers have to decide what they mean by “the same author.”