There is no dotted line around a moment.
There is no white border separating one sound from the next when somebody speaks. A football match does not arrive divided into possession, transition, press, counterattack and recovery. A child looking at a kitchen does not receive a labelled inventory saying cup, spoon, steam, danger, handle, edge, heat and distance. The world simply happens.
Yet the mind cannot work with everything at once.
So one of the quietest and most important things cognition does is this: it makes pieces.
Those pieces are the starting point for this article. We will call them tokens, while being careful about what that word means in different fields. The aim is not to force every form of cognition into one technical definition. It is to build a useful reader-facing idea: before a mind can compare, remember, count, name, explain, model or plan, it usually needs something sufficiently bounded to work on.
Quick Read
A token is a particular usable unit selected from a larger flow or field. It might be an object, sound, word, symbol, event, step, quantity, feature, move, case or moment, depending on the task.
The crucial idea is that the boundary of the token is not necessarily a boundary supplied by reality itself. Minds, languages, disciplines and tools often decide where one workable unit ends and another begins.
That decision changes what becomes easy to notice.
If the token is too large, important internal differences disappear. If it is too small, structure fragments into noise. If it is badly chosen, the problem can become almost impossible. If it is well chosen, something confusing can suddenly become tractable.
One-sentence answer: A token is a bounded instance that a mind or system can treat as one workable thing for a particular purpose.
That sounds simple.
It is not.
The First Cognitive Art Is Cutting Without a Knife
Imagine standing beside a busy road in Singapore during the evening rush.
What is there?
A physicist could describe photons, pressure waves, trajectories, friction, mass and energy. A traffic engineer could see lane occupancy, signal phases, queues and flow. A driving instructor could see gaps, hazards, blind spots and right-of-way. A child might see buses, cars, lights and people. A tired commuter may notice only whether the pedestrian signal has turned green.
It is the same road.
It is not the same set of cognitive pieces.
This is our first important move. Tokenisation, in the broad cognitive sense used here, is not merely chopping the world into arbitrary fragments. It is task-sensitive segmentation. The unit that matters depends on what you are trying to notice, predict or do.
A driver does not need to consciously represent the molecular composition of the asphalt before crossing a junction. That information is real, but it is not the useful unit at that moment. The driver needs vehicles, lanes, speeds, gaps, lights and intentions.
Cognition survives by being selective.
Selection begins earlier than most of us realise.
The World Arrives More Continuously Than Thought
Consider speech.
On a printed page, words are separated by spaces. Spoken language is not delivered with visible gaps floating between every word. Acoustic information unfolds continuously, and listeners use multiple cues, expectations and learned structures to recover meaningful units.
Now consider an ordinary activity: making breakfast.
Where does one event end?
When you open the cupboard? When you take out the bowl? When cereal enters the bowl? When milk is poured? When you sit down? When the first spoonful reaches your mouth?
Researchers studying event perception have shown that people spontaneously segment continuous activity into meaningful events and subevents. Those boundaries are related to changes in movement, goals, causes and other features, and segmentation is connected with later memory and learning. The important point for us is not that the mind uses one universal set of event boundaries. It is that parsing continuous experience into units is part of ordinary cognition.
This gives us a more careful way to speak about tokens.
A token is not always a bead that already exists on a string.
Sometimes cognition helps decide where the bead begins.
A Token Is an Instance, Not Merely a Category
The word token has a long philosophical use in the type–token distinction. A type is a general repeatable kind; a token is a particular instance of that kind.
Suppose this page contains the word:
tree
and later contains the word:
tree
There are two written occurrences—two tokens—of one word type.
The distinction becomes powerful because thought constantly moves between particulars and generalities.
This dog is a particular animal. Dog is a category. This triangle on the worksheet is one instance. Triangle is a kind. This rainfall event is particular. Monsoon rainfall is a larger pattern. This argument is one occurrence. Argument is a general class of activity.
Thinking often becomes more intelligent when we know which level we are operating at.
Many reasoning errors begin when an observation about a token is quietly converted into a claim about a type.
One student cheats; therefore students are dishonest. One company fails; therefore the business model is impossible. One medicine produces an adverse effect in one person; therefore the medicine always produces that effect. One successful case is treated as proof of a universal rule.
The token may be real.
The generalisation may still be wrong.
The Same Reality Can Be Tokenised at Different Resolutions
Think about a book.
- At one resolution, the book is one object.
- At another, it contains chapters.
- Each chapter contains sections.
- Sections contain paragraphs.
- Paragraphs contain sentences.
- Sentences contain words.
- Words can be analysed into morphemes, letters, sounds or other units depending on the task.
None of these resolutions is automatically the true one.
If you are shelving books, the book is a useful token. If you are editing an argument, a paragraph may be better. If you are checking punctuation, the sentence becomes useful. If you are studying spelling, the word or letter sequence matters.
This gives Cognitive Art one of its central questions:
At what resolution should I cut this problem?
It is a deceptively sophisticated question.
Too coarse, and important differences vanish.
Too fine, and the larger structure disappears.
Granularity: When One Thing Becomes Many
Granularity is the size of the units we choose.
Suppose a student says, “I am weak at Mathematics.”
That token—Mathematics—is enormous.
It may hide algebra, fractions, geometry, problem representation, arithmetic accuracy, method selection, working memory load, interpretation of questions, checking, time management and many other distinct processes.
If the teacher accepts the large token unchanged, the intervention may become equally coarse: “Do more Mathematics.”
Now retokenise.
The student understands algebraic manipulation but repeatedly converts verbal relationships into the wrong equations.
The problem has become smaller.
And because it is smaller, it has become more repairable.
This principle travels far beyond school.
- “The economy” can be retokenised into sectors, markets, households, firms, prices, labour, credit, trade and policy mechanisms.
- “Health” can be retokenised into symptoms, functions, systems, risks, behaviours and clinical questions.
- “A bad day” can be retokenised into several separate events, one of which went badly.
- “Climate” differs from a particular weather event.
- “A war” contains campaigns, battles, logistics, information flows, political decisions and human experiences.
- “English” contains many distinct receiving and sending capabilities.
Better resolution does not always mean smaller tokens.
Sometimes the error is over-fragmentation.
A learner memorises fifty disconnected facts but cannot see the system they belong to. A footballer watches the ball but loses the shape of the team. A reader understands each sentence but misses the argument. A manager tracks dozens of metrics without seeing the organisational pattern.
Then the repair is to make a larger token.
Intelligence is not simply zooming in.
It is learning when to zoom in and when to zoom out.
Tokens Give Attention Somewhere to Land
Attention cannot give equal priority to every available feature of a scene.
When you search for your keys, the room changes cognitively. The sofa is still there. The table is still there. The colours of the curtains remain real. But your attention is organised around a question: where is the key-shaped object that matters to my next action?
A useful token therefore acts like a handle.
It gives attention something sufficiently stable to select.
This is one reason naming can be so powerful. A name does not create the phenomenon, but it can make a distinction easier to revisit, teach, compare and discuss.
A child may repeatedly experience the feeling of wanting something another child possesses. Learning the word jealousy does not manufacture the emotion. It gives the child a linguistic handle for a recurring pattern. Later, finer distinctions can appear: envy, jealousy, admiration, resentment, insecurity.
Resolution increases.
But naming also carries danger. Once a label is available, the mind may use it too quickly. A complicated situation gets tokenised as “laziness,” “talent,” “failure,” “toxic,” “genius,” “bad attitude” or “weak student,” and the label begins replacing investigation.
A token is useful when it opens thought.
It is dangerous when it closes thought too early.
Children Learn to Cut the World More Precisely
Development can be understood partly as increasing control over useful distinctions.
A very young child may use one word broadly. Every four-legged animal may briefly become “dog.” Later, distinctions stabilise: dog, cat, horse, goat. Later still come breeds, biological classifications, behaviours, anatomical structures and ecological roles.
School accelerates this process.
Primary Science teaches children that “plants” contain parts with different functions and that apparently similar materials possess different properties. Mathematics teaches that “number” includes different kinds and relationships. English teaches that “word” is not enough: nouns, verbs, modifiers, meanings, connotations, tones, clauses, claims and evidence behave differently.
Secondary school raises resolution again.
Force is not energy. Speed is not acceleration. Mass is not weight. Correlation is not causation. An example is not proof. A claim is not evidence. A narrator is not necessarily the author. A molecule is not an atom. Revenue is not profit.
Education, at its best, is partly the disciplined replacement of crude tokens with more useful ones.
The Expert Often Sees Fewer Pieces, Not More
This sounds paradoxical.
Experts know more detail. Yet in many tasks they do not consciously carry every detail as an independent unit. They organise information into larger meaningful chunks.
A beginner looking at a chess position may see many separate pieces. A skilled player is more likely to recognise configurations, threats, structures and familiar relationships. A beginner musician sees individual notes. A trained musician sees phrases, harmonic movement and form. A beginning reader decodes words; a fluent reader processes larger structures. A novice programmer may inspect every line separately; an experienced developer sees functions, interfaces, data flows and architectural patterns.
The expert has not become blind to detail.
The expert has learned which details can be bundled together without losing what matters for the task.
This is chunking as cognitive economy.
But expertise also produces a second ability: the expert can often unpack the chunk when something goes wrong.
That flexibility is crucial.
Novice: many pieces with weak structure.
Developing learner: meaningful chunks begin to form.
Expert: large chunks for speed, with access to smaller pieces when diagnosis requires them.
Bad Tokenisation Can Create Fake Problems
Suppose two people argue about whether technology is “good” or “bad.”
The argument may be difficult partly because technology is an absurdly large token.
Antibiotics, social media, water purification, facial recognition, bicycles, nuclear weapons, calculators, sewage systems and hearing aids are all technologies. Compressing them into one moral object may destroy the distinctions needed for a serious discussion.
The same happens with words such as:
- education,
- government,
- tradition,
- capitalism,
- science,
- media,
- culture,
- success,
- intelligence,
- discipline.
These can be useful umbrella terms. They become dangerous when the umbrella is mistaken for a single uniform thing.
Before solving a disagreement, ask whether the participants are using the same token boundaries.
Sometimes they are not disagreeing about reality.
They are disagreeing about what counts as the thing being discussed.
Retokenisation: Change the Pieces, Change the Problem
Retokenisation is the deliberate act of changing the units of thought.
This is one of the most transferable techniques in Cognitive Art.
Take the sentence:
I cannot write good essays.
The sentence feels final because “good essay” is being treated as one object.
Retokenise it:
- Can I understand the question?
- Can I generate a relevant claim?
- Can I find evidence?
- Can I explain why the evidence supports the claim?
- Can I order ideas?
- Can I write a clear topic sentence?
- Can I vary sentence structure?
- Can I edit for precision?
- Can I conclude without merely repeating?
The identity claim—“I am bad at essays”—has become a set of testable capabilities.
That is not motivational language.
It is better measurement.
Retokenising Time
We tokenise time constantly.
A day. A lesson. A term. A childhood. A career. An era. A century.
These units are useful, but they can distort experience.
A student says, “Today was terrible.”
Was every minute terrible?
Perhaps a test went badly at 10:30, an argument happened during lunch, Mathematics was ordinary, training was enjoyable and dinner was fine. The emotional label has grouped several hours into one token.
Retokenising does not deny the painful events.
It prevents two events from swallowing an entire day.
The same principle matters in history. “The Industrial Revolution,” “the Cold War,” “the Renaissance” and “the digital age” are enormously useful temporal tokens. But the labels can create an illusion that millions of people suddenly entered a cleanly bounded period on the same day. Historical reality is messier, uneven across regions and full of overlap.
A time token is a tool.
It is not a wall built by the universe.
Retokenising Space
Maps teach the same lesson.
A country appears as one coloured shape. A city becomes one label. A neighbourhood becomes a boundary. These are powerful tokens for navigation and administration.
But zoom in.
A city is districts, roads, buildings, institutions, homes, pipes, wires, businesses, parks, communities and individual lives. Zoom out and the city becomes part of a metropolitan region, watershed, trade network, climate system and nation.
The correct spatial token depends on the question.
“Why is this street flooding?” may require drains and elevation.
“Why is this country vulnerable to sea-level rise?” requires a different scale.
Retokenising Cause
Perhaps the most dangerous cognitive token is the single cause.
Humans like sentences with one subject and one verb:
X caused Y.
Sometimes that is appropriate.
But many real outcomes emerge from interacting conditions.
A student’s examination result may involve knowledge, sleep, preparation, question interpretation, time allocation, stress, arithmetic accuracy and marking requirements. A business failure may involve demand, cash flow, execution, financing, competition and timing. A flood may involve rainfall, drainage, terrain, land use and maintenance. A historical revolution may involve institutions, material conditions, ideas, leadership, shocks and prior tensions.
If we insist on one causal token when the phenomenon is a network, our explanation becomes neat and wrong.
Tokens in Mathematics: The Power of Treating Something as One Thing
Mathematics repeatedly turns complex structure into manipulable units.
A variable can stand for a quantity. A bracketed expression can be treated as one unit. A vector can represent magnitude and direction together. A matrix can organise many values into a single mathematical object. A function can be treated as an object that maps inputs to outputs. A geometric shape can become a token for a family of relationships.
Good algebra often depends on seeing the right chunk.
Consider:
(x + 3)² − 5(x + 3) + 6
A learner can stare at all the symbols separately. Or the learner can temporarily treat (x + 3) as one token, say u:
u² − 5u + 6.
Now familiar factorisation becomes visible.
The Mathematics did not change.
The tokenisation did.
This is why a strong Mathematics student often appears to “see” the solution. What they may actually see is a better unit.
Tokens in Science: What Counts as the System?
Science depends on deciding what is inside the system being studied and what is outside it.
A cell can be treated as one unit. Then it can be opened conceptually into organelles, membranes, molecules and reactions. An ecosystem can be treated as one system, then retokenised into organisms, populations, trophic relationships, nutrient cycles and physical conditions.
The scientific skill is not merely knowing the pieces.
It is knowing which pieces matter for the explanation.
In mechanics, whether Earth belongs inside or outside the system changes how forces and energy transfers are described. In chemistry, defining the reaction system matters. In biology, treating “the organism” as the only token can hide microbial partners or environmental dependence. In ecology, treating species independently can hide network effects.
Scientific progress often comes with new ways of dividing reality.
Tokens in Language: Words Are Not the Only Pieces
When people hear “token,” they may think of words, especially because modern computing and language technologies use tokenisation extensively.
But language itself can be divided at many levels:
- sound features,
- phonemes,
- syllables,
- morphemes,
- words,
- phrases,
- clauses,
- sentences,
- discourse moves,
- arguments,
- narrative events.
The useful level depends on the task.
A spelling error may require attention to letters and sounds. A grammar problem may require clauses. A weak paragraph may require sentence relationships. A failed essay may require argument structure. A misunderstood story may require event models rather than individual words.
This is why “improve your English” is often too large an instruction to help anybody.
Computer Tokens and Cognitive Tokens Are Not the Same Thing
We need an important boundary.
In computing, the word token has technical meanings. Programming-language tokenisers divide source text into meaningful lexical units. Natural-language processing systems may divide text into words, subwords, characters or other units according to a specific vocabulary and algorithm.
Those are engineered procedures.
Human cognition is not simply doing the same operation inside the skull.
Our use of token in Cognitive Art is broader and deliberately reader-facing. It describes the practical fact that thinking often needs bounded units. Human perception, attention, memory, language and action can involve many different forms of segmentation and representation, and cognitive science does not reduce all of them to one universal tokeniser.
Keeping this distinction matters.
A useful analogy should illuminate.
It should not quietly become a false identity.
Why Boundaries Feel More Real Than They Are
Once a token becomes familiar, we tend to stop noticing that a boundary decision was made.
Consider a school subject.
Mathematics, English, Science, History and Geography appear on timetables as separate boxes. The boxes are useful administratively and educationally. But reality is not organised according to a school timetable. Climate science requires Mathematics. Historical reasoning depends on language and evidence. Economics crosses Mathematics, psychology, politics and history. Engineering combines physics, design, materials, computation and human constraints.
The timetable token can become a cognitive wall.
Likewise, company departments, government ministries, academic disciplines and professional titles are useful organisational tokens. Yet real problems frequently cross them.
Flooding does not care which agency owns drainage and which agency owns roads.
A child does not divide neatly into academic ability, emotional regulation, sleep, family context and motivation.
The world leaks across our boxes.
The Boundary Test
Whenever a token seems obvious, try five questions.
- What did I include?
- What did I exclude?
- Why is this boundary useful?
- What changes if I move the boundary?
- Who would cut this differently?
These questions are astonishingly productive.
Ask them of “family.” Ask them of “market.” Ask them of “species.” Ask them of “lesson.” Ask them of “problem.” Ask them of “war.” Ask them of “success.” Ask them of “Singapore.” Ask them of “the internet.”
You begin to see that cognition is full of borders.
Some are excellent.
Some are provisional.
Some are inherited.
Some are wrong.
When Tokens Become Identities
A particularly powerful token is the identity label.
“I am bad at languages.”
“She is gifted.”
“He is lazy.”
“I am not a Mathematics person.”
These labels can compress many observations into one portable token. That makes them cognitively efficient and socially contagious.
It also makes them hazardous.
An identity token can hide variability across tasks, times and conditions. A student may be slow at arithmetic under time pressure but excellent at geometric reasoning. Another may read fiction fluently but struggle with dense expository prose. Another may appear inattentive during long verbal explanations but work intensely on practical tasks.
Retokenise the person into capabilities, contexts and behaviours before turning an observation into an identity.
People are larger than the labels that help us talk about them.
Tokens Can Be Socially Negotiated
Not every token is private.
Societies coordinate by agreeing on units.
A metre. A kilogram. A dollar. A legal person. A school year. A constituency. A company. A property boundary. A diagnosis. A job title. A statistical category.
Some of these correspond to physical regularities; some are conventional institutions; many combine physical and social facts.
Once standardised, they permit coordination at scale.
Imagine construction without shared units of length. Commerce without agreed currency units. Science without standard measurement. Law without defined entities and events. Education without some way of grouping stages, subjects and qualifications.
Civilisation depends on millions of shared tokens.
But standardisation creates a responsibility: never forget which units are conventions, which are measurements, which are models and which are physical entities.
Counting Requires Token Decisions
Before you count something, you must decide what counts as one.
This sounds trivial until it is not.
How many languages are spoken in a multilingual community? Do dialects count separately? How many species occupy a habitat when classifications are disputed? How many businesses exist when one corporation owns many outlets? How many “incidents” occurred when one chain of events can be grouped in different ways?
A number can look precise while resting on a contested token definition.
This is why good quantitative reasoning asks an earlier qualitative question:
What exactly is being counted?
The counting algorithm may be flawless.
The tokenisation can still be poor.
Measurement Also Requires Tokens
Measurement takes some property and makes it comparable through defined units and procedures.
But educational and social measurement often begins by constructing the thing to be measured.
What is “reading ability”? What is “engagement”? What is “poverty”? What is “well-being”? What counts as “employment”? What is “academic progress”?
Researchers and institutions need operational definitions. Those definitions are not merely clerical details. They determine which observations become tokens in the dataset and which disappear outside the frame.
This is one reason responsible research begins before statistics.
It begins with the quality of the units.
Segmentation Changes Memory
Event-segmentation research offers a useful window into how boundaries and memory interact.
When people watch ongoing activity, they tend to perceive meaningful event boundaries. Research reviewed by Jeffrey Zacks and colleagues links such segmentation with how information is encoded, updated and later remembered. Event models are maintained while an activity is coherent and updated when significant change occurs.
Think about why this matters for learning.
A lesson that feels like ninety uninterrupted minutes can be difficult to remember. A well-structured lesson contains meaningful events:
- question,
- example,
- principle,
- worked demonstration,
- practice,
- error,
- correction,
- transfer.
These are not merely timetable segments. They give the learner a structure for reconstructing what happened.
A story works similarly. Scenes, goals, changes and consequences help organise a continuous stream of sentences into meaningful events.
But More Boundaries Are Not Always Better
There is an obvious temptation after discovering tokens: divide everything.
That fails.
A musical melody chopped into isolated notes loses phrasing. A story reduced to individual sentences loses narrative motion. A body reduced to organs loses interactions. A city reduced to buildings loses flows. A student reduced to test scores loses context. A friendship reduced to individual messages loses history.
Tokenisation creates power by reducing complexity.
Every reduction also risks removing relationships.
So the mature question is not:
Can I divide this?
Almost anything can be divided.
The better question is:
What structure survives my division, and what structure did I destroy?
The Token–Relationship Trade-Off
Every token creates an inside and an outside.
That makes the inside manageable.
It can also hide what crosses the boundary.
A school can be analysed as one organisation, but students move between school and home. A company can be analysed as one firm, but suppliers, customers, regulators and infrastructure cross its boundary. A cell can be treated as a unit, but materials and signals cross its membrane. A country can be treated as one political unit, but energy, finance, information, pathogens, goods and people cross borders.
The token is therefore never the whole analysis.
After asking what is the unit?, ask what crosses its boundary?
A Token Can Be Physical, Symbolic, Temporal or Relational
It helps to widen the imagination.
- Physical token: one apple, one cell, one building.
- Symbolic token: one numeral, one written word, one traffic sign.
- Temporal token: one event, one lesson, one era.
- Action token: one move, one step, one transaction.
- Social token: one role, organisation or legal entity.
- Data token: one recorded observation or case.
- Relational token: one connection treated as the unit of analysis, such as a friendship, trade link or cause–effect relation.
- Conceptual token: one idea temporarily handled as a unit inside a larger argument.
This variety is why we should not search for a single physical thing called “the cognitive token.”
The deeper commonality is functional.
A workable boundary lets cognition operate.
Tokens and Representation Are Different
This distinction prepares the next article.
A token answers:
What am I treating as one workable unit?
A representation answers:
How is that thing being carried, modelled or presented?
The same token can have multiple representations.
Singapore can be represented by a name, a map outline, geographic coordinates, a satellite image, a statistical profile, a legal description, a story or a memory.
One country token.
Many representations.
Tokens and Abstractions Are Different
A particular Labrador called Milo can be treated as one token.
Dog is a broader category.
Mammal is broader still.
Organism is broader again.
As we move upward, many particulars are grouped according to preserved features and relationships. That is closer to abstraction and categorisation.
Tokens give us instances.
Abstraction helps us find what can survive across instances.
Tokens and Projections Are Different
A token can also become material for imagining what does not yet exist.
A chess player recognises a position and projects candidate moves. A doctor observes findings and considers possible explanations and trajectories. An engineer represents current loads and projects performance under changed conditions. A student reads the first half of a story and anticipates what a character might do.
The token is the handle.
Projection extends beyond the handle.
The Hidden Skill Behind Good Questions
A good question often contains a good tokenisation.
Compare:
Why am I bad at school?
with:
Why do I lose marks on inference questions even when I understand the passage?
The second question is not automatically solved.
But it has a better object.
Compare:
Why is society getting worse?
with:
Which measurable outcomes changed, for which groups, over what period, and compared with what baseline?
The second question creates smaller, testable tokens.
Good inquiry is frequently an act of retokenisation before it is an act of answering.
The Student Progression: From Naming Pieces to Choosing Pieces
For education, we can describe a useful progression.
Primary years: identify and distinguish
The child learns stable names and basic categories. The educational task is to help the learner notice meaningful differences without overwhelming them with premature complexity.
Lower secondary: decompose and reconnect
Subjects become more specialised. Students must learn that broad labels hide mechanisms. They break problems into parts and then reconstruct relationships between the parts.
Upper secondary and pre-university: choose the unit strategically
The learner increasingly needs to decide which representation and granularity fit the problem. A question may be difficult precisely because the obvious unit is not the useful one.
Adulthood and expertise: retokenise when the world refuses the model
Professional problems are rarely pre-divided into textbook chapters. The capable adult must create workable units, coordinate across boundaries and revise those units when evidence shows that the original cut was poor.
Five Token Failures
1. The Giant Token
Everything is compressed into one vague object: “school,” “government,” “health,” “technology.” The result is explanation without resolution.
2. The Shattered Token
The thinker divides so finely that relationships disappear. There are facts everywhere and understanding nowhere.
3. The Wrong Boundary
The thing causing the outcome sits outside the chosen frame. A company studies internal productivity while ignoring supplier failure. A student studies content while the actual problem is question interpretation.
4. The Frozen Token
A category that was once useful becomes permanent. The world changes but the unit does not. Institutions often suffer from this: old departmental boundaries remain after problems have become cross-functional.
5. The Identity Token
A temporary pattern becomes a permanent description of a person or group. The label begins predicting behaviour because nobody looks beneath it anymore.
A Practical Tokenisation Exercise
Take one problem that currently feels too large.
Write it in one sentence.
Circle the biggest nouns.
Those nouns are probably carrying your largest tokens.
For each one, ask:
- What does this contain?
- Can it be divided into distinct mechanisms?
- Which part is actually changing?
- Which part can I observe?
- Which boundary am I assuming?
- What sits just outside the boundary?
- Would an expert in another field divide this differently?
- Could two small tokens be combined into one useful pattern?
Then rewrite the question.
You may discover that the original problem has disappeared and several smaller problems have taken its place.
That is progress.
From Token to Representation
We can now see the first movement in the World of Cognitive Art:
World → boundary → workable unit.
But a token alone is not enough.
Once we have selected a unit, we must carry information about it somehow.
We can draw it.
Name it.
Remember it.
Measure it.
Describe it.
Encode it as a number.
Place it on a map.
Turn it into a diagram.
That brings us to representation.
Frequently Asked Questions
Is a cognitive token the same as a word?
No. A word can function as a token in some tasks, but cognition can also treat objects, events, actions, quantities, features, relationships and larger chunks as workable units. This article uses token as a broad reader-facing idea, not as a claim that all cognition is literally word-token processing.
Is a token always something real?
A token can refer to something physical, but it can also be symbolic, conventional or imagined. A chess move, legal contract, fictional character, mathematical variable and planned meeting can all be treated as units of thought even though they have very different relationships to physical reality.
Who decides the boundary of a token?
Sometimes physical structure strongly constrains the boundary; sometimes perception identifies meaningful changes; sometimes language, culture, science or institutions define it; and sometimes an individual selects a temporary boundary for a task. Good reasoning asks which kind of boundary is operating.
Why does granularity matter?
Because the size of the unit determines what variation is visible. A large token is efficient but can hide internal mechanisms. A tiny token provides detail but can hide the larger pattern. Useful thought moves between scales.
Can two people see different tokens in the same situation?
Yes. Expertise, goals, language, culture, attention and prior knowledge can influence what units become salient. A mechanic, artist, driver and child can attend to different workable structures in the same car.
Does learning change tokenisation?
Very often. Learners acquire new distinctions and new chunks. They learn when several details can be treated together and when one familiar-looking object must be divided into finer parts. Expertise involves both richer differentiation and more efficient chunking.
Can a token be too useful?
Yes. A familiar label can become cognitively sticky. Once a situation has been named, people may stop checking whether the label still fits. Useful tokens should remain revisable.
What is the difference between tokenisation and categorisation?
Tokenisation identifies or constructs particular workable units. Categorisation groups instances according to shared or useful structure. They interact: you need instances to categorise, and categories influence how future instances are noticed.
What is the difference between a token and a representation?
The token is the unit being handled; the representation is the form in which information about that unit is carried or modelled. The same object can have many representations.
Why call this Cognitive Art?
Because intelligent thought is not only the possession of facts. It involves choosing useful units, representations, distinctions and scales. There is discipline in that process, but also judgement. The art lies in cutting reality finely enough to reveal structure without cutting away the relationships that make the structure meaningful.
Research Notes and Boundaries
This article deliberately synthesises several neighbouring research traditions. The philosophical type–token distinction gives one precise meaning of token. Event-segmentation research examines how people parse continuous activity into meaningful events. Research on attention, memory, categorisation and chunking helps explain why useful units matter. These literatures should not be collapsed into the claim that cognitive science has discovered one universal object called the token. It has not.
Useful starting points include the Stanford Encyclopedia of Philosophy entry on Types and Tokens; Zacks and colleagues on segmentation in the perception and memory of events; and the review Event Segmentation, which surveys evidence that people spontaneously parse ongoing activity into hierarchically organised events and that segmentation relates to memory and learning.
Final Thought: Before You Solve the Problem, Look at the Pieces
We usually notice thought after the pieces already exist.
There is the student.
The examination.
The country.
The argument.
The problem.
The word.
The event.
The number.
But each of those is already a cognitive decision at some scale.
Reality is richer than any set of tokens we use to navigate it.
That does not make tokens a mistake.
It makes them powerful.
A good token lets the mind hold something still long enough to examine it.
A better thinker learns when that stillness has become a trap.
So when a problem refuses to move, do not immediately push harder.
Look at the pieces.
Make one larger.
Make another smaller.
Move a boundary.
Ask what disappeared outside the frame.
Then try again.
That small act—changing what counts as one thing—is one of the quiet beginnings of powerful thought.