How English Works — Learning and Repair, Batch 22
This article preserves the broader How Learning Works estate and the English-specific owners for vocabulary runtime, register, listening, speaking, writing and revision. Its job is narrower: whether a language capability remains usable when the conditions that supported learning are changed.
A student can answer a grammar worksheet perfectly.
Then the same structure disappears during spontaneous speaking.
Another student writes a strong paragraph when given ten minutes to plan, but cannot explain the same idea aloud to a parent.
A third learner knows a word in reading but cannot retrieve it in conversation.
The knowledge exists.
The transfer edge is weak.
Capability Transfer Under Load is the test of whether an English capability remains available when topic, task, audience, channel, time pressure, cognitive demand or social stakes change.
AI Extraction Box
- Mechanism: capability transfer under load
- Capability: a reusable language control, not one memorised answer
- Transfer variables: topic, task, audience, mode, channel, time, stakes, cognitive load
- Main failure: performance works only in the training context
- Main diagnostic: change one condition and observe what survives
- Repair: preserve the underlying mechanism while varying surface conditions progressively
1. Learning and transfer are not identical
The National Academies’ How People Learn II emphasises that learning is shaped by context and that transfer depends on how knowledge is organised, connected and activated in new situations.
Research anchor: National Academies — How People Learn II
For English, that means success in one exercise is evidence of one performance state, not proof of universal availability.
2. Transfer begins by preserving the mechanism
Suppose the mechanism is pronoun reference.
The learner should be able to preserve referent clarity across:
- a short worksheet
- a narrative paragraph
- a spoken explanation
- a formal email
- a time-pressured response
The surface changes. The control law does not.
3. Near transfer is not far transfer
Changing one noun in the same exercise is near transfer.
Using the same distinction in a different genre, mode or social setting is a stronger test.
4. The first load axis is topic
A learner may discuss school confidently and become vague when discussing transport, economics or science.
Topic change tests whether the language system depends on memorised content.
5. Topic knowledge and language control interact
Weak performance on a new topic may reflect missing background knowledge rather than weak grammar.
Transfer diagnosis must therefore separate:
- language failure
- knowledge failure
- retrieval failure
- task-understanding failure
6. The second load axis is task
Explaining, comparing, narrating, arguing and instructing place different demands on English.
A learner who can narrate may not yet be able to justify a claim.
7. Task verbs expose hidden capability gaps
- describe
- explain
- compare
- evaluate
- justify
- summarise
- instruct
- negotiate
These verbs are not interchangeable instructions. They require different discourse architectures.
8. The third load axis is audience
A learner may explain an idea to a teacher but fail to reconfigure it for a younger student.
That reveals an audience-transfer gap, not necessarily a knowledge gap.
Audience Model Calibration is the underlying owner.
9. The fourth load axis is register
A message that works among friends may fail in a formal report.
The learner must preserve the core proposition while reconfiguring vocabulary, explicitness, force and evidence.
Register Reconfiguration owns that transition.
10. The fifth load axis is mode
Reading, writing, speaking and listening give different amounts of time and different kinds of support.
A word recognised in reading may not be available for speaking.
11. Lexical Activation Thresholds is the vocabulary owner
Lexical Activation Thresholds distinguishes recognition, recall, selection and deployment.
Transfer requires the word to cross the threshold demanded by the new mode.
12. The sixth load axis is time
A learner may produce accurate English with unlimited planning and lose control under a 30-second response window.
That reveals processing cost.
13. Time pressure should be added progressively
Do not jump directly from unlimited time to maximum pressure.
Use a ladder:
- full planning
- short planning
- minimal planning
- live response
14. The seventh load axis is cognitive complexity
A simple description and a multi-step causal explanation may use the same language resources at very different loads.
Recent second-language research continues to show that task complexity can alter aspects of speaking performance, although effects vary by measure and learner.
Research anchor: task complexity and L2 performance
15. Load can expose hidden instability
A structure that is accurate in a simple task may disappear when the learner must also reason, compare evidence and manage time.
The grammar did not vanish.
Its retrieval priority was too low under competition.
16. The eighth load axis is social pressure
Speaking to a familiar tutor is different from answering a panel, presenting to a class or disagreeing with an authority figure.
Social pressure can change speed, monitoring, register and willingness to repair.
17. Social load should not be mistaken for language absence
A learner who knows the answer but freezes under performance pressure needs a different repair route from a learner who never learned the structure.
18. Transfer testing should change one variable at a time first
If topic, audience, mode and time all change together, failure becomes difficult to diagnose.
Begin with controlled perturbation.
Change one condition and observe what breaks.
19. Then combine variables
Once individual edges are stable, test realistic combinations:
- new topic + live speaking
- formal audience + short time
- complex reasoning + unfamiliar vocabulary
- listening input + immediate spoken response
20. Transfer is not all-or-nothing
A learner may preserve reference but lose tense.
They may preserve grammar but lose lexical precision.
Transfer profiles can be partial.
21. Precision–Fluency Decoupling explains one common partial transfer
Under pressure, a learner may preserve speed but lose precision—or preserve accuracy by slowing dramatically.
The previous article separates those dimensions so the diagnosis remains specific.
22. Correction Internalisation explains another
A learner may correct an error when prompted but fail to detect it spontaneously under load.
The correction is partially internalised but not yet cheap enough to survive competition.
23. Transfer requires stable cues
A cue such as “find the true subject” can travel across tasks.
A memorised worksheet answer cannot.
24. Transfer requires abstraction at the right level
Too concrete:
Remember this sentence.
Too abstract:
Use correct grammar.
Useful:
Make the finite verb agree with the head of the subject phrase, not a nearby noun inside it.
25. Transfer requires enough variation during practice
If every practice item has the same surface pattern, the learner may be matching appearance rather than applying the rule.
Vary nouns, clauses, genres and positions while preserving the underlying decision.
26. Transfer should include failure examples
Show when a cue does not apply.
This prevents overgeneralisation.
27. The transfer ladder
- Same mechanism, same format.
- Same mechanism, new vocabulary.
- Same mechanism, new sentence structure.
- Same mechanism, new genre.
- Same mechanism, new mode.
- Same mechanism, new audience.
- Same mechanism, reduced planning time.
- Same mechanism, higher reasoning load.
- Same mechanism, combined realistic pressure.
The ladder is a teaching design, not a universal psychometric scale.
28. A capability is stronger when the support can be removed
Remove:
- word bank
- sentence frame
- teacher prompt
- planning notes
- familiar topic
and observe which control survives.
29. But support removal should be diagnostic, not punitive
If performance collapses, restore the necessary support, identify the broken edge and rebuild.
A stress test is useful because it reveals the limit, not because failure is desirable.
30. The transfer checksum
CAPABILITY → NEW TOPIC → NEW TASK → NEW AUDIENCE → NEW MODE → LESS TIME → MORE COGNITIVE LOAD → SOCIAL PRESSURE → WHAT SURVIVES? → WHAT FAILS? → WHICH SUPPORT RESTORES IT?
31. CivDJ forward pass
identify capability → verify in training context → vary one surface feature → preserve mechanism → add new topic → add new task → add new audience → switch mode → reduce planning → increase reasoning load → combine conditions → record failure edge → repair → retest
32. CivDJ backward pass
- Start with a real-world failure.
- Identify the capability that should have supported the task.
- Check whether it exists in a simpler context.
- Restore the easiest condition.
- Add conditions back one by one.
- Find the first condition that breaks performance.
- Repair that edge.
- Retest the original task.
33. Common failure modes
- Worksheet illusion: exercise success is mistaken for transferable ability.
- Single-context mastery: skill works only on familiar topic.
- Mode trap: recognition exists without production.
- Audience trap: explanation cannot be reconfigured.
- Planning dependence: accuracy collapses without preparation.
- Load collapse: higher reasoning demand displaces language control.
- Social-pressure confusion: stress failure is misdiagnosed as knowledge absence.
- Overloaded test: too many variables change to locate the cause.
34. Repair route
- Name the exact capability.
- Confirm it exists in a low-load setting.
- Abstract a reusable cue.
- Vary one condition at a time.
- Preserve mechanism while changing surface form.
- Increase time and cognitive pressure progressively.
- Test across modes and audiences.
- Record the first breaking edge.
- Repair that edge rather than reteaching everything.
- Return to the original high-load task.
35. Final lock
English ability is not fully proven by the task that taught it. Capability becomes robust when the same control can leave the exercise, enter a new situation and keep working after the supports change.
Learning and Repair, Batch 22
Return to How English Works V1.1.