How Knowledge Compilation Works | How Slow Rules Become Fast Procedures

Quick Read: From Knowing That to Knowing How

Skill-acquisition theories influenced by John Anderson’s ACT and ACT-R architectures distinguish between declarative knowledge and procedural knowledge.

Declarative knowledge contains facts, examples and explicit rules: “to form this tense, use this rule”; “to solve this equation, perform these transformations.”

Procedural knowledge is organised as condition-action routines that can execute more directly when the right situation appears.

A 2025 paper in Studies in Second Language Acquisition describes a three-stage skill-acquisition account grounded in ACT-R: learners begin with declarative knowledge, develop procedural knowledge through practice, and move toward increasingly efficient performance. In this tradition, practice can transform explicit knowledge into skill-specific production rules.

Knowledge compilation is the learning system’s way of turning instructions into executable routines.

The Driving Lesson

A beginning driver thinks in sentences.

Mirror. Signal. Check. Brake gently. Turn. Straighten. Watch the lane. Check the speed.

Every action competes for conscious attention.

An experienced driver still performs these operations, but many no longer require full verbal control. Perception and action have become coupled through practice.

The skilled driver does not forget the rules. The rules have partly disappeared into the procedure.

That is why expertise can feel mysterious to novices. Experts often cannot easily narrate every micro-decision because those decisions are no longer individually occupying consciousness.

Declarative Knowledge Is Not a Weak Form of Learning

There is a temptation to dismiss explicit knowledge as “just theory.”

That is a mistake.

Declarative knowledge gives the learner a portable description before the skill exists. It allows a teacher to explain a rule, a textbook to record it, a student to inspect it and a learner to apply it deliberately.

Without declarative knowledge, many complex skills would have to be discovered through enormous amounts of unguided trial and error.

The problem is not having explicit rules.

The problem is stopping there.

Why Rules Are Slow at First

When a learner first applies an explicit rule, several operations are separate.

• recognise the situation;
• retrieve the relevant rule;
• hold the rule in working memory;
• interpret what each part means in the current case;
• execute the action;
• check whether the action worked.

Each operation costs time and attention.

Practice creates opportunities for the system to connect the condition directly to the appropriate action.

Proceduralisation

One classic idea in knowledge compilation is proceduralisation: parts of the explicit instruction are transformed into procedures that can operate without repeatedly retrieving the declarative rule in full.

Instead of thinking, “The equation is balanced, so if I subtract 4 from the left I must subtract 4 from the right,” the experienced learner simply performs the balancing move while preserving equality.

The principle has not vanished. It has been compiled into the action.

Composition

Another idea is composition: several small production steps can combine into a larger routine.

A beginning algebra learner may separately identify like terms, move terms, simplify coefficients and isolate the variable. An experienced learner can sometimes treat the whole familiar pattern as one chunked transformation.

This reduces decision overhead.

The same phenomenon appears in reading. Skilled readers do not consciously decode every familiar word letter by letter. Musicians do not consciously name every note before every movement. Typists do not spell each key location in working memory.

Why Compiled Knowledge Feels Fast

Compilation reduces interpretation.

The learner no longer needs to reopen the full manual for every familiar case. A known condition triggers a practiced response.

This creates speed, consistency and lower working-memory demand.

It also explains why experts can devote more attention to higher-level structure. A strong writer can think about argument because basic sentence construction is relatively automatic. A mathematician can think about strategy because elementary manipulations do not consume the whole workspace.

Knowledge Compilation and Automaticity Are Related but Not Identical

Automaticity describes performance that requires relatively little conscious control.

Knowledge compilation describes one theoretical route by which explicit knowledge becomes procedural and more efficient.

A routine may become highly fluent through repeated procedural use. Compilation helps explain how the representation itself changes during that development.

The distinction matters because “practise until automatic” says little about what exactly is being compiled.

The Danger: Wrong Knowledge Can Compile Too

Practice does not inspect morality or correctness before strengthening a routine.

If a learner repeatedly applies an invalid shortcut and receives no corrective feedback, the shortcut can become faster.

A student misreads “percentage increase” and repeatedly divides by the final value. A writer uses a memorised phrase in every conclusion. A programmer reaches for the same inefficient pattern. A musician rehearses the same fingering mistake.

Once compiled, a bad routine can be harder to interrupt precisely because it runs quickly.

Quality control must happen before speed becomes the main objective.

Why Feedback Matters During Compilation

Early practice is where the system decides which condition-action relationships deserve to become reliable.

Fast feedback helps keep the compiled route aligned with the target.

But feedback should become less intrusive as the learner gains control. Permanent correction after every micro-step prevents the learner from developing internal monitoring.

The sequence should move from external quality control toward self-correction.

Why Worked Examples Help Before Compilation

A novice needs something correct to compile.

Worked examples reduce the search space and expose expert sequencing. Self-explanation connects the sequence to underlying principles. Completion problems then require more learner execution. Independent practice allows the procedure to stabilise.

This creates a natural instructional chain:

Observe → explain → complete → execute → vary → retrieve → transfer.

See How Worked Examples Work for Performance and How Completion Problems Work.

Knowledge Compilation in Mathematics

Mathematics makes compilation visible because novices often verbalise every step.

Consider solving a linear equation.

At first, the learner may explicitly ask: What is the inverse operation? Which side should I change? What must happen to preserve equality?

With practice, familiar transformations become fast enough that working memory can focus on the strategic structure of the problem.

This is beneficial until the learner sees a case where the compiled routine does not fit. Then adaptive expertise must reopen the procedure.

Continue through the Mathematics Learning Hub.

Knowledge Compilation in English

Language contains many rules that begin explicitly and later become procedural.

A learner first remembers subject-verb agreement as a grammar rule. Later the correct form is produced with little conscious thought.

Writers compile larger routines too: opening a paragraph, integrating evidence, punctuating dialogue, checking pronoun reference, shifting register for audience.

The danger is compiling formulaic writing. Efficient structure should reduce mechanical load, not replace judgement.

Continue through the English Learning Hub.

Knowledge Compilation in Second-Language Learning

Skill-acquisition theory has been particularly influential in second-language research.

A learner may begin with explicit grammar knowledge: “third-person singular takes -s.” During actual conversation, consciously applying the rule to every sentence is too slow.

Practice can make the rule increasingly procedural so correct production occurs under communicative time pressure.

The 2025 Cambridge study on the three-stage model tests aspects of this declarative-to-procedural progression empirically, showing that the model remains an active research framework rather than only a historical theory.

Knowledge Compilation in Science

Science requires both conceptual reasoning and procedural routines.

Reading a graph, converting units, identifying variables, balancing equations, setting up an experiment and checking significant figures can become compiled procedures.

That frees attention for interpretation and mechanism.

But scientific reasoning must remain able to interrupt routines when evidence is surprising.

Continue through the Science Learning Hub.

Knowledge Compilation in Examination Technique

Exam advice often remains declarative.

“Read the question carefully.” “Show your working.” “Answer the command word.” “Check your units.” “Plan before writing.”

Students may agree with every sentence and still fail to execute under pressure.

The advice has not become a procedure.

Compilation requires repeated cue-action practice. When the command word appears, the answer-form routine should activate. When a numerical answer is reached, the unit-and-reasonableness check should run.

eduKateSG already has a distinct canonical owner for this performance problem in How Exam Technique Fails | Why Good Advice Breaks When It Never Becomes a Procedure. The present article owns the general cognitive mechanism behind that transition.

The Cue Is Part of the Procedure

A procedure is useless if the learner cannot recognise when to run it.

This is a major reason blocked practice can create fragile skill. Every question in the block already belongs to the same method, so the learner never practises recognising the condition.

Good compilation includes the trigger.

Not merely: “How do I complete the square?”

Also: “What features tell me that completing the square is useful here?”

Why Interleaving Helps Method Selection

Once procedures are stable, mixed practice can train cue discrimination.

The learner encounters several possible methods and must select the right production rather than repeating the last one.

This is one way compiled procedures become more useful rather than merely faster.

See How Interleaving Works.

Why Variable Practice Helps Generalisation

A procedure compiled under one narrow surface may become bound to that surface.

Variable practice changes parameters and representations so the condition-action relationship has to recognise deeper structure.

Series 0031 develops this in How Variable Practice Works.

Why Retrieval Matters After Compilation

A compiled routine that is never retrieved after delay can still decay.

Successive relearning protects access across time. The learner runs the procedure again after spacing, receives correction if it fails, and returns later.

Compilation makes a routine efficient. Spaced retrieval helps keep it available.

The Speed-Accuracy Trade-Off

Students often chase speed too early.

That is dangerous because compilation is partly a compression process. If the underlying routine is wrong, compression can make the error fast.

Accuracy and understanding should stabilise before speed becomes the dominant goal.

A useful progression is:

• accurate with support;
• accurate independently;
• accurate across variation;
• accurate after delay;
• then increasingly fast.

The Fluency Illusion

Fast performance can look like understanding.

Sometimes it is.

Sometimes the learner has compiled a narrow routine that only works on familiar surfaces.

Transfer tests reveal the difference. Change the representation. Remove the cue. Alter the constraint. Ask why the method works.

If speed disappears completely, the procedure may have been overfitted to the practice format.

Compilation and Schemas

Schemas compress knowledge into structured chunks. Compiled procedures compress sequences into executable routines.

The two interact.

A strong schema helps the learner recognise the situation. The compiled procedure helps the learner act within it.

See How Schemas Work.

Compilation and Cognitive Load

Procedural fluency reduces working-memory demand.

This is one reason foundational fluency can improve higher-order performance. If every elementary operation requires conscious control, complex reasoning becomes crowded.

But automaticity is not valuable because consciousness is bad. It is valuable because consciousness is scarce.

Compilation moves reliable work out of the scarce channel so attention can be spent on decisions that still need it.

Compilation and Adaptive Expertise

Series 0029 introduced adaptive expertise as efficiency without rigidity.

Knowledge compilation explains where much of the efficiency comes from.

The adaptive problem begins when compiled routines become so fast that the learner stops noticing whether the case still fits.

Strong expertise therefore needs both layers:

• a fast procedural layer for familiar work;
• a monitoring layer that can interrupt the procedure when evidence becomes unusual.

See How Adaptive Expertise Works.

When a Compiled Routine Should Be Reopened

Experts need triggers that return a procedure to conscious inspection.

• repeated unexpected errors;
• a new context;
• a boundary condition;
• contradictory evidence;
• a changed rule or standard;
• a result outside the normal range;
• or a learner who can execute but cannot explain or transfer.

When these signals appear, slow down.

Decompile mentally. Inspect the assumptions. Rebuild if necessary.

The First Weak Link

If a student knows the rule but cannot perform under time pressure, the first weak link may be compilation rather than understanding.

If a student performs fast but collapses on a changed representation, the first weak link may be narrow cueing.

If a student repeatedly makes the same fast mistake, the first weak link may be a wrongly compiled production.

Different states require different repairs.

Use the Diagnostics & Recovery Hub for the wider diagnostic system.

A Compilation Protocol for Students

• Understand the rule: know why the step is valid.
• Observe a correct model: study a complete worked route.
• Explain: justify important steps.
• Complete: fill missing steps in partial examples.
• Execute: solve independently.
• Correct: repair errors before they become fast.
• Repeat: build reliable condition-action links.
• Vary: change numbers, representations and contexts.
• Mix: practise choosing among nearby procedures.
• Time: add speed only after accuracy is stable.
• Return: retrieve after delay.
• Transfer: test on an unfamiliar problem.

A Teacher Protocol

When students say, “I understand it when you explain it but I cannot do it myself,” treat the statement as useful diagnostic information.

The declarative representation may exist while the procedural route is still thin.

Do not simply explain again.

Shift the learner into execution with support that fades: completion problems, prompted practice, immediate correction, then independent trials. Once performance stabilises, vary and mix.

A Parent Protocol

If a child can tell you the rule perfectly but takes too long to use it, the problem may not be memory of the rule.

Ask the child to perform several short, accurate applications with immediate checking. Keep the examples simple enough that the same core routine repeats. Then gradually vary the surface.

The aim is to turn explanation into execution without losing understanding.

Knowledge Compilation and AI

AI creates an unusual risk for skill acquisition.

If a tool performs the procedure every time, the learner may accumulate declarative familiarity without enough personal execution for compilation.

The student recognises good output, understands the explanation and still cannot produce the routine independently.

Tools should therefore sometimes step aside.

Use AI to model, compare, generate practice and provide feedback—but preserve learner attempts where the skill itself must become executable.

The Deep Principle: Expertise Is Compressed History

A skilled action looks simple because the learner is carrying a great deal of past reasoning inside it.

Rules were learned.

Examples were inspected.

Errors were corrected.

Steps were repeated.

Conditions became familiar.

Sequences compressed.

Eventually, what once required a paragraph of conscious instruction becomes one fast move.

That is not the disappearance of knowledge.

It is knowledge changing form.

Research and Further Reading

• Testing the Three-Stage Model of Second Language Skill Acquisition
• ACT-R Work on Declarative Knowledge, Procedural Knowledge and Transfer
• How Schemas Work
• How Variable Practice Works
• Study & Learning Methods Hub

---

eduKateSG Learning Node Series · 0032 of the continuing series. Previous: 0031 — How Variable Practice Works. Continue through the Study & Learning Methods Hub.