How English Works | The Ontology

A taxonomy can tell us that a Labrador is a kind of dog, and that a dog is a kind of mammal.

Useful.

But suppose we need to say more.

A dog can have an owner.

An owner can be a person or organisation.

A veterinary appointment concerns an animal.

A medicine has an active ingredient.

A student enrols in a course.

A course is taught by a teacher.

A teacher may also be a parent.

Suddenly the problem is no longer merely:

Which category does this thing belong to?

It becomes:

What kinds of things exist in this domain, what properties can they have, how may they relate, and what else follows when those statements are combined?

That is ontology territory.

An ontology is English made explicit enough that relationships themselves become part of the model.

The word sounds philosophical because its ancestry is philosophical. In information science and knowledge representation, however, an ontology has a practical job: it provides a formal vocabulary for describing a domain, including classes, individuals, properties and constraints among them.

Quick Read

One-sentence answer: an ontology works when it makes the important kinds of things in a domain, the relationships among them and the logical constraints on those relationships explicit enough that different people or systems can share not only labels but consequences.

W3C’s OWL 2 Web Ontology Language models knowledge using classes, properties, individuals and data values. Its expressive power goes beyond a simple hierarchy: an ontology can say that two classes are equivalent, that two classes are disjoint, that a property links one kind of thing to another, or that membership in a class depends on a relationship.

This gives us the ontology’s central architecture:

classes: kinds of things;
individuals: particular things;
properties: relationships or attributes;
domains and ranges: the kinds of things a property connects;
restrictions: conditions that describe or constrain a class;
equivalence and difference: statements about when concepts mean the same thing or cannot overlap;
inference: consequences that follow from the model even when nobody typed them explicitly.

The Ontology Is Not the Taxonomy

A taxonomy is principally an organisation of concepts.

An ontology can express a richer network of meanings.

Taxonomy:

Animal → Mammal → Dog

Ontology:

Dog is a Mammal.
A Dog may haveOwner a Person.
A GuideDog is a Dog that performs an AssistanceRole.
A VeterinaryVisit concerns an Animal.

The taxonomy gives us the ladder.

The ontology gives us the web of claims around the ladder.

taxonomy = organised concept hierarchy
ontology = explicit model of entities, relations and constraints

This distinction protects the new article from colliding with How English Works | The Taxonomy. Taxonomy asks where concepts sit. Ontology asks what they are allowed to mean together.

Classes Are Types of Thing

A class groups individuals that share a defined conceptual type.

Examples:

Student
Teacher
Book
School
Assessment

The important move is that the class is not merely a folder name.

It can participate in logical statements.

Every SecondaryStudent is a Student.

Now if Maya is asserted to be a SecondaryStudent, a reasoner can infer that Maya is also a Student.

No additional sentence about Maya was required.

The class relation carried the consequence.

Individuals Are Particular Things

Ontology separates the idea of a class from particular members of that class.

Class:

School

Individual:

a particular school

Class:

Book

Individual:

one particular physical or digital book

This seems simple until ordinary language begins blurring type and instance.

We say:

The tiger is endangered.

Do we mean one tiger?

The species?

The class of tigers generally?

Formal ontology forces us to decide which level we are modelling.

Ontology begins by refusing to let “a kind of thing” and “this particular thing” remain casually interchangeable.

Properties Turn Nouns into a World

A list of classes is still static.

Properties create relations.

Examples:

student enrolledIn course;
teacher teaches course;
book writtenBy author;
city locatedIn country;
assessment hasScore number.

The verbs are doing ontological work.

They tell the model how one thing can connect to another.

W3C OWL distinguishes object properties, which connect individuals to individuals, from data properties, which connect individuals to data values.

So:

Maya enrolledIn Course-7

is structurally different from:

Maya hasScore 82

The first connects two entities.

The second connects an entity to a value.

Domain and Range Give a Property Semantic Shape

Suppose a property is:

teaches

A domain may indicate that things using this property as subject are Teachers.

A range may indicate that the object is a Course.

This can support inference.

If a model contains:

Alex teaches English101

and the ontology says the domain of teaches is Teacher, a reasoner may infer:

Alex is a Teacher.

This is powerful.

It is also dangerous when the model author thinks “domain” merely means “the property is usually used here.” In formal ontology, statements can have logical consequences.

Ontology vocabulary is not decorative metadata. Definitions can make new claims appear.

Restrictions Describe Classes Through Relationships

We can define a class not only by its name but by what relations its members have.

For example:

EnrolledStudent = Student who is enrolledIn at least one Course.

Or:

Parent = Person who hasChild at least one Person.

Now class membership can arise from relationships.

This is one of the points where ontology becomes much more expressive than taxonomy.

The system can reason from relational structure, not only from preassigned category labels.

Equivalent Does Not Mean Merely Similar

In ordinary English, people casually write:

These are basically the same.

Formal ontology requires more care.

If two classes are declared equivalent, they share membership conditions in the model.

That is much stronger than saying they are related or often overlap.

For example:

AdolescentLearner ≡ Learner aged 13–18

is a definitional claim.

But:

Teenager is associated with SecondaryStudent

is not equivalent. A teenager may not be in secondary school, and a secondary-school system may include students outside that age band.

Ontology design punishes fuzzy equivalence because one loose equals-sign can propagate incorrect classifications through the whole graph.

Disjointness Says Two Classes Cannot Share a Member

Suppose an ontology says:

DraftDocument and PublishedDocument are disjoint.

If one document is later asserted to belong to both, the model exposes an inconsistency.

Disjointness is therefore a powerful constraint.

It should be used only when reality genuinely supports the exclusion.

Bad model:

Student and Employee are disjoint.

Many real people are both.

A rigid ontology can make perfectly ordinary reality look impossible.

Before declaring two categories disjoint, find the person, object or event most likely to belong to both.

Inference Is the Ontology’s Hidden Output

Ontology is interesting because the explicitly written statements are not necessarily the complete knowledge state.

Suppose:

Maya is a SecondaryStudent.
Every SecondaryStudent is a Student.
Every Student is a Learner.

The model can infer:

Maya is a Student.
Maya is a Learner.

Those conclusions were not separately asserted.

They follow from the model.

asserted facts + ontology rules → inferred facts

This is where tiny wording choices become consequential. Change the class hierarchy or property definition and the inferred world can change everywhere.

The Open-World Assumption Changes What Missing Means

OWL-style reasoning commonly operates with an open-world assumption.

If the ontology does not say that Maya has a sibling, we cannot automatically infer:

Maya has no siblings.

We know only that the model currently contains no such assertion.

This differs sharply from many databases and forms, where a missing value may be treated as absent, false or incomplete according to local rules.

not known to be true ≠ known to be false

This connects directly to The Schema and The Empty State. Different systems assign different meanings to silence.

Ontology Does Not Automatically Make the Knowledge True

An ontology may be logically elegant and factually wrong.

Suppose it declares:

Every swan is white.

A reasoner can use that rule perfectly.

The world can still contain black swans.

Formal consistency and empirical truth are different questions.

reasoning can faithfully amplify a bad premise

This is the ontology version of the Schema lesson: structurally valid does not mean true.

Ontology Alignment Is Hard Because Different Systems Carve Reality Differently

System A uses:

Teacher

System B uses:

Educator

Are they equivalent?

Maybe.

Perhaps System B includes tutors, lecturers and trainers while System A means only school-employed teachers.

Vocabulary matching alone is insufficient.

Ontology alignment must compare definitions, scope, relations and consequences.

same label can hide different concepts
different labels can hide the same concept

This is why controlled labels, taxonomies and ontologies are connected but distinct layers.

Primary School: Build a Tiny World of Things and Relationships

Use familiar objects:

Student
Teacher
Book
Library

Then add relations:

Student borrows Book;
Teacher recommends Book;
Library contains Book.

Ask:

Is “Book” a thing, a category of things, or one particular book?

Children begin noticing the difference between types, instances and relations without needing formal notation.

Lower Secondary: Find the Bad Inference

Give students:

Every prefect is a student.
Maya is a student.
Therefore Maya is a prefect.

The conclusion does not follow.

Then compare:

Every prefect is a student.
Maya is a prefect.
Therefore Maya is a student.

Now the inference is valid.

This teaches that hierarchy has direction.

Upper Secondary: Audit a Formal World Model

What classes exist?
Which things are individuals rather than classes?
What properties connect them?
What do property domains and ranges imply?
Which classes are equivalent?
Which are disjoint?
Which restrictions create class membership?
What facts are inferred rather than directly asserted?
What does missing information mean?
Which real cases would break the model?

This turns ontology into a lesson in vocabulary, logic, modelling and the limits of representation.

Ten Failure Modes of Ontology English

Taxonomy collapse. The ontology is only a category tree with no meaningful relations.
Type-instance confusion. A class and one particular member are modelled as though they were the same level.
Property vagueness. Relations such as “relatedTo” carry too little meaning to support reliable inference.
Domain-range surprise. Property declarations create classifications the model author did not intend.
False equivalence. Similar concepts are declared identical.
Over-disjointness. The ontology forbids legitimate overlap found in real life.
Inference blindness. Authors inspect only asserted facts and miss consequences generated by the model.
Closed-world instinct. Missing assertions are incorrectly treated as false statements.
Logical-truth confusion. Consistent reasoning is mistaken for empirical correctness.
Alignment by label. Different systems are merged because words match rather than because meanings match.

How to Build a Better Ontology

Define the domain and the questions the ontology must answer.
Separate classes from individuals.
Name properties narrowly enough to carry useful meaning.
Check what domains and ranges will imply.
Use equivalence only when membership conditions genuinely align.
Use disjointness only after testing overlap cases.
Make restrictions explicit.
Run inference and inspect unexpected consequences.
Keep empirical evidence separate from logical consistency.
Version the ontology when conceptual commitments change.

Then perform the impossible-world test:

What perfectly real person, object or event would this ontology be unable to represent without contradiction?

The Deeper Idea: Ontology Is a Machine for Making Meanings Collide

A glossary can define terms one at a time.

A taxonomy can arrange concepts.

An ontology forces definitions to meet each other.

If Teacher is a Person, Tutor is equivalent to Teacher, and Volunteers are disjoint from Employees, what happens when one tutor volunteers and is paid for another role?

The model must survive contact between its claims.

concepts → relations → constraints → inference → consequences

That is why ontology belongs to How English Works.

It is the point where vocabulary stops being a list of words and becomes a formal world whose sentences can generate new sentences.

Reader Checklist

What domain is being modelled?
What are the classes?
What are the individuals?
Which properties connect them?
What do those properties imply?
Which concepts are equivalent or disjoint?
What facts can be inferred?
What does missing information mean?
Could the model be logically valid but factually wrong?
Which real edge case most threatens the ontology?

Research and Further Reading

W3C — OWL 2 Web Ontology Language Primer

Final idea: an ontology is good when it makes a domain precise enough to reason about while remaining humble enough to admit that the model is not the world.