Turn Speaking Mistakes Into Targeted Drills: An Error-to-Practice Router

Do not repeat an entire conversation every time one sentence goes wrong. Save the exact attempt, identify the smallest likely bottleneck, choose a drill that matches it, then retest the same task. A missing word, unstable verb form, unclear sound, long pause, and missed follow-up are different problems. They should not all produce the same generic instruction to “practise more.”

The router below is a LinguaLive editorial tool, not a diagnostic test. One recording cannot prove why an error happened, and an automated label may be wrong. Use the first classification as a working hypothesis and change it when the retest produces different evidence.

Start with evidence you can replay

Keep three things before correcting anything:

the prompt or situation;
the learner's exact words, including the gap or repair; and
what the listener needed to understand or do.

This prevents a polished correction from replacing the original evidence. It also separates a harmless variation from a mistake that blocked the task. The CEFR Companion Volume organises communicative performance around activities and strategies, including spoken production, interaction, and repair. It does not prescribe this router, but it supports reviewing language in relation to the communicative task rather than counting every difference from a model sentence as equally important.

If the note came from an AI system, first use the AI correction review checklist. Do not build a week of drills around a rule that has not been verified.

Use the five-branch error router

What happened in the attempt?	Working bottleneck	Matched micro-drill	Retest evidence
You knew the idea but could not retrieve a word or phrase	Retrieval	Prompt-to-phrase recall in several short contexts	The phrase appears without looking and still fits the message
The form changed unpredictably across similar sentences	Form control	Contrast two or three meaningful sentence frames	The target form survives a new subject, time, or purpose
The listener could not identify a key word although the form was known	Sound or intelligibility	Hear, mark, produce, record, and compare the critical contrast	The word is understood in a fresh sentence without a visual clue
The message stopped for long planning gaps	Timing or formulation	Prepare key points, deliver once, shorten planning, then repeat	The task finishes with fewer disruptive gaps, not merely faster speech
The sentence was correct but the exchange stalled	Interaction	Practise clarification, repair, and response-dependent follow-ups	The next turn responds to what was actually said

The branches can overlap. If a word disappears because its sound form is weak, retrieval and pronunciation may both matter. Choose the first branch that can be tested cheaply; do not turn uncertainty into a large remedial syllabus.

Route one mistake from capture to retest

1. Write the task outcome

Use an observable sentence such as “confirm the delivery date,” “explain why I changed my mind,” or “ask one relevant follow-up.” Avoid goals such as “speak better” because they cannot tell you whether the retest worked.

2. Mark the smallest consequential moment

Select one timestamp or one short excerpt. Ask whether the moment changed the meaning, made the listener ask for repair, or repeatedly interrupted the task. If not, log it as low priority. A noticeable accent feature is not automatically an intelligibility problem.

3. Choose a provisional branch

Use the table, then write why you chose that branch. “Retrieval: I recognised the word after seeing it” is more useful than “vocabulary mistake.” If two branches are plausible, design the first drill to distinguish them.

4. Run a small matched drill

Keep the drill narrow enough to complete in a few minutes. For a retrieval gap, hide the answer and retrieve the phrase from a meaning or situation cue. A controlled study indexed by Ghent University compared word writing with retrieval practice for L2 vocabulary; it is evidence about vocabulary learning under that study's conditions, not proof that this router or every retrieval drill is superior. Here, retrieval is used as a testable practice choice when the evidence points to a retrieval gap.

For a form problem, contrast the exact meanings that require different forms. For a sound problem, keep the key word inside a complete sentence. For a timing problem, preserve the message while reducing preparation gradually. For an interaction problem, make the next prompt depend on the previous answer.

5. Retest the original task, then transfer once

First repeat a comparable version of the original task. Then change one feature: the person, time, reason, or follow-up. A memorised line that works only on the training prompt is not yet evidence that the bottleneck is resolved.

6. Retire, revise, or escalate

Retire the drill when the target survives two reasonably comparable attempts and no longer blocks the task. Revise the branch if the drill did not change the evidence. Escalate consequential or ambiguous language to an authoritative reference or qualified person instead of increasing repetition.

Worked example: the missing deadline phrase

Original task: Tell a colleague that the draft can arrive on Friday, but the final file is needed by Monday morning.

Evidence: The learner pauses, says “until Monday,” then restarts. When shown “by Monday morning,” they recognise that it expresses the intended deadline.

Working branch: Retrieval, with a possible meaning contrast between “by” and “until.”

Drill: Use five situation cards: arrival deadline, continuing availability, meeting start, payment deadline, and travel duration. Retrieve the relevant time phrase, then say one complete message. Check the contrast with a reliable reference rather than accepting an unverified generated explanation.

Retest: Repeat the project update with different days, then answer an unexpected follow-up about what happens if the deadline moves. If the phrase is accurate in isolated cards but disappears in the exchange, the next branch may be timing or interaction rather than vocabulary knowledge.

Worked example: a conversation that stops after every answer

Original task: Discuss weekend plans for three minutes.

Evidence: The learner answers each question clearly but never reacts to the partner's answer. Grammar correction would not solve the task.

Working branch: Interaction.

Drill: Listen to one answer and choose a branch based on its content: person, reason, time, contrast, or consequence. The guide to practising follow-up questions shows how to build response-dependent turns instead of memorising a fixed list.

Retest: Use a new topic and count developed exchanges, not total questions. If the learner asks a relevant follow-up but cannot formulate the wording, route that smaller moment again as retrieval or form.

A one-week retest card

Field	Entry
Task and intended outcome	What needed to happen?
Exact evidence	Quote or timestamp; do not rewrite it
Communicative impact	Blocked, repairable, or low priority
Working branch	Retrieval, form, sound, timing, or interaction
Matched drill	What will be repeated, contrasted, or varied?
Immediate retest	What changed on a comparable task?
Transfer retest	Did it survive one changed prompt?
Decision	Retire, revise the branch, or escalate

The card can fit inside the 10-minute speaking session template or a longer routine. If recordings show that imitation is the right next move, compare shadowing, repeating, and retelling before choosing the drill. When the evidence points to planning or timing rather than missing language, use the speaking-under-time-pressure ladder to add the clock one controlled step at a time.

Limits and disclosure

This router has not been validated as an assessment or instructional intervention. It cannot infer a cognitive cause from one sample, assign a CEFR level, distinguish every language variety, or determine whether an accent feature needs changing. Stress, task familiarity, disability, recording quality, first-language influence, and the interaction partner can all affect the evidence.

LinguaLive develops commercial AI language-learning software. No proprietary learner outcomes, product-accuracy benchmark, customer example, or conversion data was used here. The neutral next action is to route one verified mistake, run one small drill, and keep the before-and-after sample.

Frequently asked questions

Should I correct every mistake in a recording?

No. Start with moments that changed the message, caused a repair, or repeatedly blocked the task. Logging every difference can hide the few practice decisions that matter for the next attempt.

How many times should I repeat a drill?

This guide does not set a universal number. Stop when a comparable retest gives enough evidence to retire or revise the branch. Repetition without a retest can make a training line familiar without showing transfer.

Can AI identify the right drill automatically?

It can suggest a hypothesis, but its transcript, correction, explanation, and cause label can each fail. Preserve the original attempt, verify consequential language, and treat the retest as the decision point.

Sources and editorial review

This guide was checked against its primary official or academic reference on 5 August 2026. Language usage can vary by region, relationship, and situation. Review the primary source.