How to Improve Learning Transfer | Deep Structure, Variation & New Problems

A student can solve ten percentage questions in a percentage chapter and still miss the eleventh question when the same relationship appears inside a shopping problem, a Science graph or an unfamiliar examination context. The learner may know the procedure and still fail to recognise when the knowledge is useful.

That gap is transfer.

Learning transfer improves when a learner becomes better at recognising the underlying structure of a problem, separating that structure from its surface features, selecting the relevant knowledge without being told what topic is being tested, and adapting the knowledge when the new situation is not identical to the old one.

This article continues the eduKateSengkang How to Improve series after How to Improve Learning From Mistakes. It is the practical companion to How Learning Transfer Works | When Knowledge Survives a New Problem. The canonical page explains the mechanism. This guide focuses on building the capability deliberately.

The examples below are instructional examples rather than records of actual students or official examination items. Their purpose is to expose the decisions a learner must make when knowledge has to travel.

The central problem: school often tells the learner what to retrieve

A chapter heading is a cue. A worksheet titled “Simultaneous Equations” is a cue. A teacher saying “Today we are practising inference questions” is a cue. These supports can be useful while a method is being learned, but they also remove one important decision: which knowledge applies here?

Transfer becomes visible when that cue is weakened or removed.

A learner who can solve a ratio question only when the worksheet is labelled “Ratio” may have procedural skill without reliable selection. A learner who can identify cause and effect in a Science chapter but not in a newspaper paragraph may know the language without recognising the relationship in another setting. A student who can write a strong claim-evidence paragraph on a rehearsed topic may still struggle when the evidence arrives as a chart instead of prose.

Improving transfer therefore requires practice in recognition, selection, adaptation and verification, not only repetition.

What actually has to travel?

Knowledge is not one uniform object. Different things can transfer:

a fact,
a concept,
a causal relationship,
a mathematical structure,
a problem-solving strategy,
a checking routine,
a writing move,
a method-selection rule,
a way of evaluating evidence.

The transfer test should match what was learned. If the target is a vocabulary word, a new sentence may be enough. If the target is a method-selection skill, changing only the numbers proves little. If the target is a causal model, the learner should be able to reason through a changed system where the same mechanism still matters.

A useful question before designing transfer practice is:

Which part of the original learning should remain valid when the surface changes?

Separate deep structure from surface features

Surface features are what a problem looks like. Deep structure is the relationship that determines what kind of reasoning or method fits.

Consider these two situations:

A shop reduces a $50 item by 20%.
A tank loses 20% of its water from an original 50 litres.

The nouns differ. The underlying multiplicative relationship is similar.

Now consider:

A shop reduces a $50 item by $20.
A shop reduces a $50 item by 20%.

The surface is almost identical, but the underlying relationship differs. A learner who follows keywords rather than structure can easily confuse them.

Transfer improves when learners are asked not merely to solve, but to state what relationship makes this problem belong with another problem.

Do not vary everything at once

A completely unfamiliar problem can be a poor diagnostic because several things change simultaneously: context, vocabulary, representation, number of steps, method selection and time pressure.

Build transfer through controlled variation. Change one dimension while keeping the target structure stable.

Change the numbers but keep the context.
Change the context but keep the representation.
Change the representation but keep the relationship.
Change which quantity is unknown.
Change the order in which information is presented.
Mix the target method with plausible alternatives.
Reduce scaffolding.

When the learner fails, controlled variation helps identify what changed too far or which recognition cue is still missing.

Build a transfer ladder

A practical progression can move through increasingly distant cases:

Repeat: same structure and nearly identical surface.
Near transfer: same structure, changed numbers or wording.
Context transfer: same relationship in a different setting.
Representation transfer: move between words, diagrams, tables, graphs or equations.
Selection transfer: mix several possible methods and remove the topic label.
Adaptive transfer: the old method must be modified because the new problem is similar but not identical.
Cross-domain transfer: a reasoning pattern learned in one subject becomes useful elsewhere.

This ladder is not a universal scientific scale. It is a practical way to increase distance gradually instead of jumping from worked example to extreme novelty.

Worked case 1: transfer in Mathematics

Suppose a learner has learned to solve:

3x + 5 = 20.

The student can execute the algebra accurately. That proves a useful skill, but it says little about transfer.

Step 1: change the numbers

Use 4x − 7 = 21. This is a near-transfer check. The structure remains an unknown transformed by multiplication and addition or subtraction.

Step 2: change the representation

“Four identical notebooks and a $7 voucher give a final cost of $21. What is the price of one notebook?”

The learner must now construct the equation rather than receive it.

Step 3: change which quantity is unknown

“A notebook costs $7. Four notebooks and a delivery charge cost $34. What is the delivery charge?”

The story still contains multiplicative and additive relationships, but the unknown occupies a different role.

Step 4: add a plausible competitor

Mix the problem with ratio, percentage and simultaneous-equation questions. Do not announce which chapter each problem belongs to. The learner must decide whether a single linear equation is the correct model.

Transfer is not proved by faster algebra alone. It is increasingly supported when the learner can build and select the model independently.

Ask “why this method?” before “can you do this method?”

Execution practice answers one question: can the learner carry out a method?

Transfer requires another: can the learner recognise when the method belongs?

Before calculating, ask the learner to state:

what the problem is asking,
which relationship matters,
which method fits,
which nearby method does not fit,
what evidence in the problem supports the choice.

This separates method recognition from method execution and makes selection visible enough to teach.

Worked case 2: transfer in English comprehension

A student learns to support an inference with textual evidence in a story. The learner can answer questions about a character’s emotions when the passage contains obvious body-language clues.

Near transfer might use another story with different characters and another emotion.

A more demanding transfer changes the text type. Use a short speech:

“We have delayed the launch twice, and I will not ask the team to promise another date until the safety checks are complete,” the manager said.

Ask what the statement suggests about the manager’s attitude toward the next launch date.

The surface has changed from narrative emotion to stance in reported speech. The deeper operation remains similar: identify a claim about attitude, locate evidence, and explain how the wording supports the inference without exceeding it.

A learner who has transferred the skill may say that the manager is cautious about committing to another date because they explicitly refuse to promise one before safety checks are complete.

The transfer target is not the vocabulary word “cautious.” It is evidence-governed inference.

Change genre to reveal whether the learner knows the operation

A reading skill that survives only one genre may still be cue-bound.

Move the same reasoning operation through:

story,
dialogue,
news report,
advertisement,
speech,
chart plus caption,
short argument.

Do not change every literacy demand at once. If the new genre introduces too much unfamiliar vocabulary, a failure may reflect language load rather than the inference skill being tested.

Worked case 3: transfer in Science reasoning

Suppose a learner has understood the idea of a fair comparison in an investigation where different amounts of light are tested while other relevant conditions are kept the same.

A weak transfer test changes only the values of light intensity.

A stronger transfer test changes the scientific context completely. Imagine two paper towels being compared for absorbency. If one towel is larger, the comparison is confounded because size changes alongside towel type.

The surface differs from the light investigation. The underlying reasoning remains: when the aim is to test the effect of one factor, other relevant conditions should not vary systematically with it.

Then change the representation. Give a table of trials rather than a written method and ask which column reveals a condition that was not controlled.

The learner is no longer recalling the words “fair test.” They are applying the relationship to new evidence.

Do not confuse vocabulary transfer with conceptual transfer

A student may use the phrase “controlled variable” in several topics while still failing to recognise which condition threatens a comparison.

Conversely, a learner may reason correctly in ordinary language before remembering the technical term.

Test both when both matter:

Can the learner identify the relationship?
Can the learner explain why it matters?
Can the learner name it accurately?

Terminology supports precise communication. It should not become a substitute for the underlying concept.

Use comparison to make transfer cues visible

Transfer is difficult when learners do not know what feature should trigger the old knowledge.

Place two problems side by side and ask:

What looks different?
What relationship is the same?
Which information is decorative?
Which feature determines the method?
What would have to change before another method became better?

The learner is not simply noticing similarity. They are learning which similarity matters.

Use contrast to prevent false transfer

Transfer can fail by overuse as well as underuse. A student sees one familiar feature and applies an old method when it should not be used.

Contrast a target case with a near neighbour:

percentage decrease versus subtraction of a fixed amount,
correlation versus causal evidence,
inference versus explicit statement,
area versus perimeter,
direct proportion versus merely increasing together.

Ask for the decisive difference. A learner who knows when not to transfer is developing stronger boundary knowledge.

Interleave when the learner can execute but cannot choose

Blocked practice is useful when a procedure is new. The method is held stable so the learner can concentrate on execution.

Once execution is reasonably secure, mix related problem families. Interleaving removes the guarantee that the previous method remains correct.

For example, mix:

ratio, percentage and fraction comparison,
linear and simultaneous equations,
direct statement and inference questions,
cause, correlation and coincidence,
mean, median and mode.

Require the learner to name the decision cue before solving selected items. The aim is not random difficulty. It is discrimination among plausible alternatives.

See How Interleaving Works in Learning.

Vary practice without destroying the target

Variation is useful only when the learner can still see or reconstruct the stable rule.

If every example changes in too many ways, the learner may see a collection of unrelated cases instead of one transferable structure.

A strong sequence alternates stability and variation:

make the rule clear,
change one surface feature,
compare the examples,
state what stayed invariant,
change another feature,
later remove the explicit comparison.

See How Practice Variability Works in Learning.

Use self-explanation to extract the transferable rule

After solving, ask the learner to explain:

why the method worked,
what feature triggered it,
what feature did not matter,
what nearby case would require another method,
where else the same relationship might appear.

This shifts attention from answer production toward an abstract rule that can be retrieved later.

See How Self-Explanation Works in Learning.

Use analogy carefully

Analogy can support transfer by showing that two different situations share a relationship.

But an analogy becomes dangerous when surface similarities are imported as if every feature matches.

After using an analogy, always ask:

Which part maps across?
Which part does not?
Where would the analogy break?

Transfer requires selective mapping, not wholesale copying.

See How Analogy Works in Learning.

Remove the chapter label

One of the simplest transfer interventions is to stop telling the learner what topic each question belongs to.

Use a mixed set and ask the learner to classify the problem before solving:

What kind of relationship is present?
Which method family is relevant?
What evidence supports that classification?

Then occasionally include a problem that looks like one topic but belongs to another. This tests whether the learner relies on deep cues rather than superficial keywords.

Remove the worked example gradually

A worked example can reveal expert structure. Transfer requires that the learner eventually generate the structure without it.

Fade support:

complete worked example,
worked example with explanations missing,
completion problem with later steps missing,
similar independent problem,
mixed problem without method cue,
changed-context problem.

If performance collapses only when the example disappears, the learner may have been following rather than generating.

Retrieve before transfer

A learner cannot transfer knowledge that is unavailable when needed.

Before blaming transfer, check retrieval. Can the learner explain the relevant concept or method without the original notes?

If the knowledge itself cannot be retrieved, repair availability first. Then test selection and application.

This prevents a memory failure from being misdiagnosed as failure to recognise structure.

Understand before transfer

Shallow knowledge tends to stay attached to the form in which it was learned.

A learner who memorises “multiply by 0.8 for a 20% discount” can succeed on familiar discount questions. A learner who understands that 80% of the original remains can adapt more easily when the unknown changes or when the problem asks for the original value.

Understanding creates more routes into the knowledge. See How to Improve Understanding.

Transfer and schema formation

Transfer becomes easier when separate examples are organised into a schema: an internal structure that captures what matters across cases.

Instead of storing twenty isolated examples, the learner begins to store relationships such as:

part-whole proportion,
conservation under transformation,
claim supported by evidence,
change one variable while controlling relevant alternatives,
unknown quantity represented by a linear relationship.

The examples become instances of a more general structure.

See How Schema Formation Works in Learning.

Transfer and mistakes

A correction can appear successful and still fail to transfer.

After repairing an error, ask:

Can the learner perform the same operation on a fresh item?
Can they recognise the operation after the context changes?
Can they distinguish a near neighbour where the repair should not be applied?
Can they still do it after a delay?

This is why How to Improve Learning From Mistakes ends with fresh, changed and delayed cases rather than the corrected original.

Transfer and feedback

Feedback can accidentally reduce transfer when it becomes too specific to the original page.

“Change this sentence to this” may repair one sentence. “Your evidence is present, but the paragraph does not explain how it supports the claim” gives the learner a relationship that can travel to another paragraph.

After feedback, always ask where the same lesson will be tested again. See How to Improve Learning From Feedback.

Transfer and metacognition

The learner must notice that old knowledge might be relevant.

Useful self-prompts include:

What does this problem remind me of structurally?
Which relationship have I seen before?
What is different this time?
Does the old method still fit every condition?
What would prove that I transferred the wrong rule?

Transfer is partly a retrieval problem and partly a monitoring problem. The learner must both bring knowledge to mind and evaluate whether it belongs.

See How to Improve Metacognition.

Transfer under examination conditions

Examinations often remove many of the cues available during teaching. Topics are mixed, wording changes, time is limited and several methods can appear plausible.

Prepare transfer progressively:

secure the knowledge,
practise the method,
mix nearby methods,
change context and representation,
remove topic labels,
add time pressure only after selection is reasonably stable.

Adding the clock too early can hide whether the failure comes from transfer or time control.

See How to Improve Exam Performance.

Transfer in AI-assisted learning

AI can make one solution easy while leaving transfer completely untested.

After AI support:

close the response,
state the underlying rule in your own words,
solve a fresh problem,
change the context,
include one near neighbour where the rule should not apply,
verify factual claims separately where necessary.

If the learner can reproduce only the original answer, the tool may have solved the task without strengthening transfer.

See How AI-Assisted Study Works.

A ten-minute transfer drill

Minute 1: State the rule or relationship from memory.
Minutes 2–3: Solve one familiar example.
Minutes 4–5: Solve a changed-context example.
Minutes 6–7: Compare a near neighbour where another method applies.
Minutes 8–9: Explain the decisive cue.
Minute 10: Predict where else the same structure could appear.

The routine is a teaching device, not a fixed evidence-based dosage. Complex transfer may require substantially more time.

A thirty-minute transfer workshop

Choose one well-understood concept or method.
Retrieve it without the notes.
Solve one standard problem.
Identify its deep structure.
Change one surface feature.
Change the representation.
Mix in a plausible competitor.
Ask the learner to classify before solving.
Use one case requiring adaptation rather than direct copying.
Finish by explaining what stayed invariant across the set.

If the learner fails early, return to understanding or retrieval. If execution remains strong but classification fails, concentrate on discrimination and selection rather than more blocked repetition.

A weekly transfer audit

Select a small number of important skills and ask:

Can the learner use the skill without the chapter label?
Can they recognise it in another representation?
Can they distinguish a near neighbour?
Can they explain why the method fits?
Can they adapt it when one condition changes?
Does the transfer survive a delay?

Transfer audits are most useful for high-value knowledge: foundational relationships, recurring examination operations and ideas that connect several topics.

For parents: ask for another example, not another explanation

If a child says they understand, ask them to create or solve another case.

Useful questions include:

Can you show me a different example?
Can you show me a case where this rule would not work?
What would this look like in another subject or context?
How would you recognise this if the worksheet did not tell you the topic?

Parents do not need to know the full subject solution. They can ask the child to make the connection visible and then verify uncertain subject details with the appropriate teacher or source.

For teachers: plan the future problem while teaching the current one

When teaching an important method, prepare at least one later problem that changes the surface and one near neighbour that should trigger a different response.

This changes lesson design from:

Can students do today’s example?

to:

What future variation would show whether students learned the relationship rather than the page?

The transfer task does not have to appear immediately. In many cases it is more informative after the original example is no longer fresh.

For tutors: stop carrying the classification decision forever

A tutor can unintentionally destroy transfer evidence by announcing the method before every problem.

Instead of saying, “This is a percentage question,” ask:

What relationship do you see?
What method family might fit?
What alternative are you ruling out?
Which part of the question made you decide?

As the learner improves, reduce those prompts. Transfer strengthens when classification increasingly moves inside the learner.

Common transfer traps

Chapter dependence: the heading supplies the method.
Keyword matching: one familiar word triggers the wrong rule.
Single-example learning: the example becomes the concept.
Blocked-practice comfort: execution looks strong because selection is removed.
Too much variation too soon: the stable structure disappears inside novelty.
No contrast cases: the learner never learns when not to apply the rule.
Representation dependence: knowledge works only as words, equations or diagrams, but not across forms.
Immediate-only testing: the learner transfers while the lesson is fresh but not later.
Transfer by analogy without limits: irrelevant features are carried across.
Far-transfer fantasy: expecting an abstract skill to appear automatically in a very distant domain without explicit bridges.

How to know transfer has improved

The learner recognises the relevant relationship without a topic label.
Surface changes produce less performance collapse.
The learner can explain why one old method applies and another does not.
Knowledge moves between words, diagrams, tables, graphs or equations more reliably.
Changed contexts no longer hide familiar structure as easily.
The learner detects near neighbours where transfer would be wrong.
Less external prompting is needed to select a method.
Transfer survives after a delay.
The learner can adapt rather than merely copy when the new problem differs in an important way.

Do not claim more transfer than you tested

Success on one changed question supports a narrow claim. It does not prove that the learner can transfer the idea anywhere.

Use precise language:

“The learner solved a new numerical form independently.”
“The learner selected the method correctly in this mixed set.”
“The learner transferred the relationship from a word problem to a graph.”
“Cross-topic transfer has not yet been tested.”
“The skill remains dependent on a prompt.”

This protects the distinction between observed capability and optimistic extrapolation.

The deepest transfer test

After learning something important, ask the learner to face a new problem where the old knowledge is useful but not announced.

Then ask two questions:

What from the old learning still applies here?
What is different enough that I must adapt rather than copy?

Those two questions capture the balance at the heart of transfer. Too little connection and the learner treats every problem as new. Too much connection and the learner forces old methods onto situations where they no longer fit.

Continue the How to Improve route

Final principle

Learning is not finished when the student can reproduce what was practised.

The stronger test arrives when the wording changes, the representation changes, the context changes, the topic label disappears and several methods become plausible.

If the learner can still recognise the deep structure, choose the relevant knowledge, adapt it to the new conditions and verify that the result makes sense, the knowledge has begun to travel.

Small Group Tutorials

How to Improve Learning Transfer | Recognise the Deep Structure, Change the Surface and Use Knowledge Somewhere New