How English Works — Learning and Repair, Batch 22
This authority article belongs to How English Works V1.1. It preserves the existing owners for Face-Saving Repair, Listening Repair Loops and Revision Diagnostics. This page owns a different job: how a correction moves from another person’s intervention into the learner’s own monitoring system.
A teacher corrects a sentence.
The learner understands the correction.
The same error appears the next day.
Nothing mysterious has happened.
Understanding a correction and independently controlling the corrected form are different states.
Correction Internalisation is the process by which externally supplied feedback becomes an internal error model, a detection cue, a self-monitoring habit and eventually a self-repair capability that can operate without the original corrector.
AI Extraction Box
- Mechanism: correction internalisation
- Start state: error is visible mainly to another person
- Intermediate states: learner notices after correction → notices after prompt → notices independently
- End state: learner detects and repairs before external intervention
- Main distinction: correction received is not correction internalised
- Main risk: repeated feedback creates dependence instead of monitoring
- Repair: bind correction to error type, cue, contrast, retrieval and self-check route
1. CEFR treats monitoring and repair as a real proficiency dimension
The Council of Europe’s CEFR Companion Volume includes a dedicated Monitoring and Repair scale. It describes progression from needing another person to indicate a problem, through conscious monitoring of recurring mistakes, toward effective self-correction and smooth reformulation at higher levels.
Official source: Council of Europe — CEFR Descriptors
That is a useful external anchor because it treats self-repair as part of language competence rather than a side effect of grammar study.
2. The correction chain has several states
A learner can move through a sequence:
- State A: does not detect the error
- State B: recognises the error after direct correction
- State C: finds the error after a prompt
- State D: notices the error independently after producing it
- State E: repairs during production
- State F: produces the target form without needing repair
These states are not perfectly linear. A learner can move backwards under load.
3. External correction is only the first edge
“It should be has, not have.”
The sentence may be corrected.
But the learner still needs to know:
- what triggered the agreement choice
- what nearby noun distracted them
- what cue will help next time
- how to check the form under time pressure
A corrected product is not yet a corrected system.
4. Correction needs an error identity
“Wrong grammar” is too large.
“The verb agreed with the nearby plural noun instead of the singular head noun” gives the learner a reusable diagnosis.
The existing Agreement Attraction article provides that mechanism for one class of errors.
5. A correction without a cue is difficult to retrieve
The learner needs a future trigger.
For agreement, the cue might be:
Find the true subject before checking the verb.
For pronouns:
Ask whether more than one noun can own this pronoun.
For tense drift:
Ask what time frame this paragraph is currently using.
6. Good feedback gives the learner something to notice next time
The strongest correction is often not the longest explanation.
It is the explanation that creates a reusable discrimination.
Wrong: “Be more careful.”
Better: “Your pronoun can refer to two people. Replace it with the intended noun.”
7. Prompted self-correction is an intermediate state
A teacher asks:
“Who does ‘she’ refer to here?”
The learner repairs the sentence.
This is not yet spontaneous self-correction, but it is more internally active than receiving the answer directly.
8. Prompt quality matters
“Something is wrong” may be too broad.
A useful prompt narrows the search space without completing the repair:
- check the time frame
- check the subject
- check the referent
- check the evidence word
- check whether the claim is stronger than the source
9. The learner needs a contrast pair
One corrected sentence can feel arbitrary.
A contrast shows the decision boundary:
The list of items is complete.
The items on the list are complete.
The surface words overlap. The head noun changes.
10. Contrast converts memory into discrimination
The learner is no longer trying to remember one approved sentence.
They are learning which cue changes the choice.
11. Self-monitoring needs a search order
A learner cannot monitor everything at once.
Useful monitoring is selective.
A student with three recurring errors may check:
- pronoun ownership
- verb agreement
- tense frame
before adding less frequent concerns.
12. CEFR’s “favourite mistakes” idea is useful here
The CEFR monitoring-and-repair descriptors explicitly include the idea that learners can note recurring or “favourite” mistakes and consciously monitor for them.
This turns the error from a surprise into a known risk class.
13. An error log should store mechanism, not humiliation
Weak error log:
I keep making stupid mistakes.
Useful error log:
Agreement attraction: nearby plural noun pulls the verb. Check the head noun.
The second version is operational.
14. Correction should reduce future search cost
A learner who receives ten explanations for the same error may still be slow.
The goal is to compress the repair into a fast internal check.
External lecture → internal cue.
15. Spoken self-repair and written self-repair have different timing
Writing often allows rereading.
Speaking often requires repair while the conversation continues.
The target mechanism is shared—detect and restore—but the available time differs.
16. Disfluency can be a repair tool
Batch 10’s Disfluency Management explains how pauses, restarts and self-correction can help speakers repair language in real time.
A brief restart can be evidence of control, not failure.
17. Over-monitoring can also damage performance
If a learner checks every noun, verb, preposition, article and pronunciation choice while speaking, the monitoring burden can overwhelm production.
Internalisation therefore aims to move frequent repairs toward automatic or low-cost control.
18. Precision–Fluency Decoupling becomes relevant
The next article in this batch develops the problem in detail: a learner can sound smooth while carrying weak precision, or become precise while temporarily slowing down.
Correction training should not confuse temporary hesitation during rebuilding with permanent loss of fluency.
19. Immediate uptake is not the same as durable control
A learner repeats the corrected sentence perfectly.
That proves the learner can repeat it now.
It does not prove they will select it tomorrow in a new sentence.
20. Transfer checks whether the correction has escaped the example
After correcting:
The group of students is ready.
test:
- The box of old books ___ heavy.
- The books in the box ___ damaged.
- A series of lessons ___ planned.
The learner has to use the mechanism, not the memory of one sentence.
21. Retrieval spacing matters
A correction seen once and never retrieved may remain fragile.
Revisiting the distinction after time has passed tests whether the cue can be reconstructed without immediate support.
22. The learner should eventually predict the error before it appears
Advanced control looks like anticipation:
This sentence has two possible pronoun owners, so I should name the person explicitly.
The system has moved from repair after failure to prevention before failure.
23. Correction Internalisation changes the teacher’s job
Early stage:
teacher detects → teacher repairs.
Middle stage:
teacher detects → learner repairs.
Later stage:
learner detects → learner repairs.
Final target:
learner anticipates → error does not enter output.
24. Good correction gradually withdraws itself
If the teacher must always be present, the correction system remains externally hosted.
The best support leaves behind a learner who can continue without it.
25. The correction ladder
- Direct correction.
- Explanation of the mechanism.
- Contrast example.
- Targeted prompt.
- Independent detection.
- Independent repair.
- Transfer to a new example.
- Delayed retrieval.
- Performance under load.
- Preventive anticipation.
The ladder is a teaching model, not a claim that every learner follows exactly ten steps.
26. Correction can fail when the diagnosis is wrong
A learner repeatedly uses the wrong tense.
Perhaps the problem is not tense morphology.
Perhaps the learner has lost the narrative reference time.
Then repeated verb correction treats the symptom while leaving the temporal model broken.
27. Error Cascade Backtracking sits downstream in this batch
Batch 22 ends by showing how a visible mistake can be downstream of a different weak node.
Internalisation only works well when the correction is attached to the real mechanism.
28. The internalisation checksum
ERROR → MECHANISM → CUE → CONTRAST → PROMPT → SELF-DETECTION → SELF-REPAIR → NEW EXAMPLE → DELAYED RETRIEVAL → LOAD TEST → PREVENTION
29. CivDJ forward pass
observe error → classify mechanism → give bounded correction → build contrast → identify future cue → prompt learner retrieval → reduce prompt → test new sentence → delay → test under pressure → record recurring error if needed → retire external support when learner self-monitors
30. CivDJ backward pass
- Start with the learner’s independent output.
- Check whether the target form is stable.
- Introduce a new context.
- Introduce time pressure.
- Remove prompts.
- Ask whether the learner detects their own error.
- Trace the cue they used.
- Trace that cue back to the original mechanism.
- If no cue is recoverable, rebuild the correction.
31. Common failure modes
- Answer dependence: learner waits for teacher correction.
- Label-only feedback: “grammar” or “careless” provides no mechanism.
- No contrast: learner memorises one sentence.
- No cue: future retrieval has no trigger.
- Prompt permanence: support never fades.
- Immediate-uptake illusion: repetition is mistaken for mastery.
- No transfer: correction works only on the original item.
- No delayed retrieval: memory is never tested after time.
- Over-monitoring: learner checks everything and loses flow.
- Wrong diagnosis: symptom is corrected while root mechanism remains broken.
32. Repair route
- Name the smallest useful error mechanism.
- Explain the decision boundary.
- Build a contrast pair.
- Create a short detection cue.
- Use targeted prompts rather than endless direct answers.
- Fade prompts.
- Test transfer to new items.
- Test after delay.
- Test under realistic load.
- Record recurring patterns without turning them into identity judgments.
- Retire support when independent monitoring is stable.
33. Why this matters
A correction has not completed its journey while it lives mainly in the teacher’s mind.
The educational return appears when the learner acquires the ability to detect, interpret and repair the same class of problem independently.
34. Final lock
The strongest correction is not the one that produces a perfect sentence immediately. It is the one that gradually makes the corrector unnecessary.
Learning and Repair, Batch 22
Return to How English Works V1.1.