Kind, Particular, Relationship, or Value: Four Questions Before You Type the Axiom

Open almost any first ontology and look at the class hierarchy. It is usually impressive. Dozens of classes, nested neatly, every one with a label. Then look closer. One of the classes is a specific instrument with a serial number. Another is called InstalledAt. Another is SerialNumber. Somewhere near the bottom is a class named after a person, with exactly one conceivable member: that person.

The file opens. Nothing is syntactically wrong. The hierarchy looks busy, which is easy to mistake for thorough. But when a reasoner pass or a query finally asks a real question — which instruments are installed at which site, and who operates them? — there is nothing to join. The relationships were never properties, the particulars were never individuals, and the values were never data.

The Class, Individual, and Property Decision Kit (for Protégé) is built for the moment before that happens. It is a decision pass: four questions you answer for every new name, a small right-modelled ontology that already answers them, and a deliberately wrong teaching file that makes the four usual mistakes so you can recognize them in the editor.

Get it here: Class, Individual, and Property Decision Kit (for Protégé)

What are the four mistakes?

They are worth naming exactly, because each has a different repair.

1. A particular thing modelled as a class. Anemometer-A1 is one physical instrument. Making it a class implies there could be many Anemometer-A1s, each an instance of it. There cannot. It is an individual of the class Instrument.

2. A relationship modelled as a class. “Installed at” connects two things: an instrument and a site. As a class, InstalledAt has nowhere to put either end of the connection. As an object property, installedAt runs from Instrument to Site, and every assertion of it says exactly which instrument is where.

3. A value modelled as a class. A serial number is not a kind of thing and not a separate entity; it is a value of a thing. As a class, SerialNumber cannot hold the actual string. As a data property, serialNumber runs from Instrument to xsd:string, and the serial NFS-AN-014 sits right on the instrument it identifies.

4. A person modelled as a class of one. Maya Chen as a class is the same error as the anemometer, with a name attached. She is an individual of the class Person.

Every one of these looks harmless in the hierarchy. Every one of them quietly removes something a later query needed.

Why do these distinctions matter so much?

Because the line between kinds, particulars, relationships, and values is one of the oldest and most productive ideas in organized thought — and every time someone draws it clearly, the system downstream gets more useful.

Aristotle’s Categories separated “this man” from “man”: the particular from the kind it belongs to. He also set relation and quantity apart as their own categories, distinct from the things that stand in relations or have quantities. More than two millennia later, the four questions in this kit are recognizably the same cut.

Carl Linnaeus organized Systema Naturae (first edition, 1735) around ranks of kinds, and natural history came to anchor each named species to physical specimens in collections. The species is the class; the specimen in the drawer is an individual. Confusing the two would make a collection impossible to reason about — and taxonomists have been careful not to for a very long time.

In 1976, Peter Chen published “The Entity-Relationship Model — Toward a Unified View of Data,” which gave database designers three separate things to identify: entities, relationships, and attributes. Generations of data modelers learned to ask, of every noun in a requirements document, “is this a thing, a link between things, or a property of a thing?” OWL’s classes and individuals, object properties, and data properties are a richer version of the same discipline.

A first ontology that makes the four mistakes is not careless. It is skipping a question humans have been asking, productively, since antiquity.

What are the four questions?

The kit puts them in a table, and they are short enough to memorize:

  • Is this a kind of thing, with many possible particulars? Model it as a class. In the right file: Instrument, Site, Person.
  • Is this one particular thing? Model it as an individual of the right class. In the right file: Anemometer-A1, Dock-North, Maya Chen.
  • Is this a relationship between two things? Model it as an object property. In the right file: installedAt (Instrument → Site) and operatedBy (Instrument → Person).
  • Is this a value of a thing, not another entity? Model it as a data property. In the right file: serialNumber (Instrument → xsd:string).

The guide adds a quick tiebreaker that works surprisingly often. If you would give it a proper name or a serial, it is probably an individual. If you would say it in a sentence as a connection — “the instrument is installed at the site” — it is probably an object property. If you would write it on a datasheet as a string or a number, it is probably a data property.

What is in the kit?

The zip contains a 4-page guide, two ontologies in two syntaxes each, and a README.

**Northshore-Field-Station-RIGHT.owl**, plus a Turtle copy of the same graph:

  • Classes: Instrument, Site, and Person, with Instrument declared disjoint with Person.
  • Object properties: installedAt (Instrument → Site) and operatedBy (Instrument → Person).
  • A data property: serialNumber (Instrument → xsd:string).
  • Individuals: Anemometer-A1 (an Instrument with serial NFS-AN-014, installed at Dock-North, operated by Maya Chen), Dock-North (a Site), and Maya Chen (a Person).

**Northshore-Field-Station-WRONG-teaching.owl**, plus a Turtle copy: the same four things modelled as classes, on purpose, labelled for practice only. It is not a starter for a real project, and the files say so.

The guide covers the four questions, the contents of the right file, a first-hour walk through each file, a table of the four errors and their repairs, and how to start your own ontology afterward.

Both OWL files and both Turtle copies were parsed as RDF with rdflib 7.6.0 when the kit was built, and a structural check confirmed that the right file models the anemometer, the dock, and Maya Chen as individuals with the correct properties, while the teaching file models all four as classes. The Northshore Field Station is fictional — a made-up site that gives the two files a shared story. Its ontology IRI, https://organa.agency/examples/northshore-field-station-modeling, is an example IRI for the kit, not a live lookup service.

What does the first hour look like?

The kit is written for Protégé Desktop 5.6.9, the current stable line. (Protégé 6.0 is still in alpha.)

The right file first. Open Northshore-Field-Station-RIGHT.owl with File → Open. In the Entities tab, open the class hierarchy: three classes, and a disjointness axiom between Instrument and Person. Open Object properties and look at the domains and ranges of installedAt and operatedBy. Open Data properties and find serialNumber. Then open Individuals, select Anemometer-A1, and read everything about it in one place: its type, its serial, its site, and its operator. Select Dock-North and Maya Chen and confirm they are individuals, not classes.

If you like, start a reasoner from the Reasoner menu — HermiT is the usual choice when it is installed with Protégé — but that is optional. The point of this hour is the modelling decisions, not a clash. Stop the reasoner before you switch files.

Then the teaching file. Open Northshore-Field-Station-WRONG-teaching.owl in a new window. Walk the class hierarchy and find all four errors. Then, for each one, write down which of the four questions would have caught it. That small act of writing is what turns recognition into reflex.

The guide’s summary of the repair is worth quoting directly: “The repair is not to add more classes. The repair is to put the particular, the relationship, and the value in the right kind of axiom.”

What happens after the hour?

Start your own file with File → New, set an IRI you actually control, and save. Then, for every name you are about to add, answer the four questions first — on paper, or in a comment — and only then type the axiom. Add classes only for kinds. Add individuals for particulars. Add object properties for relationships. Add data properties for values.

Add disjointness where two kinds truly cannot share a member — instrument and person, in the teaching story — and not as decoration. A disjointness axiom is a claim the reasoner will hold you to.

How does this sit next to the other Protégé and ontology tools?

The First Consistent Ontology Hour is a reasoner hour: a consistent file, then one deliberate clash to read and clear. This kit is the decision that should happen before a clash is even possible. If you own both, run the decision pass first and the reasoner pass after it — not instead of it.

The Ontology & Knowledge-System Starter Kit is a vault-and-Turtle knowledge-graph starter, not a Protégé decision kit. If you own more than one of these, they complement each other; none substitutes for another.

The kit does not include the Protégé application or a reasoner binary; you install those from official project sources. Protégé is a trademark of its trademark owner (the Protégé project / Stanford University, as used for the official editor). This is an independent tool for people who use Protégé. It is not an official Protégé product.

A hierarchy with something to join

You are not paying for a longer class list. You are paying for four questions you will reuse on every new name for as long as you build ontologies, a small right file that already answers them, and a teaching file that puts the four usual mistakes in front of you so you recognize them the next time they try to sneak into your own work. That is the difference between a hierarchy that looks thorough and one that can actually answer a question.

Get the Class, Individual, and Property Decision Kit (for Protégé): Class, Individual, and Property Decision Kit (for Protégé)

It will be listed on our Tools for Protégé page alongside the rest of the tools in that line. You can browse every Organa Obscura tool on the Products page.

Leave a Reply