Introduction
The sentence you want to fix may be the sentence you need
Imagine opening yesterday's lesson transcript and seeing “Did you went last week?” The impulse to repair it is understandable. A document full of visible mistakes can feel unfinished, especially when it will be shared with a learner or colleague. Yet that particular sentence may be the reason the teacher discussed auxiliary verbs. Replace it silently with “Did you go last week?” and the explanation that follows appears unnecessary. Worse, the learner seems to have produced the target question before receiving help. The relevant distinction is between readable presentation and altered evidence. Punctuation, spacing and consistent turn labels can make the exchange easier to follow. Supplying a different verb changes what the exchange records. Before editing the sentence, decide whether the document should preserve an attempted form, provide a corrected model or do both in clearly separated places. That decision determines which changes belong in the transcript and which belong in accompanying teaching notes.
Language teaching also makes ordinary editorial shortcuts unreliable. A repeated phrase may be a requested retry. A teacher's apparently redundant question may invite self-correction. A short answer can matter because it follows a specific prompt. Lyster and Ranta's original classroom study distinguishes corrective feedback from the learner responses that follow it; the distinction alone explains why collapsing the two into one polished statement loses information. This guide uses that narrow conceptual point, rather than treating the study as a test of transcript cleanup. For an editor, the immediate task is to preserve who supplied the language and what happened next. A clear record can later support discussion about teaching choices. It cannot establish, by wording alone, why a learner answered as they did, whether the response was independent or whether the same form will be available in another conversation. Those questions need evidence beyond a cleaned page.
References: Corrective feedback and learner uptake, Example transcription instructions
Choose what the notes will be used to do
A learner reviewing vocabulary, a tutor planning the next session and a researcher examining interaction need different documents. Agree on the intended reader, editing boundary and review question before spending time on cosmetic changes. For ordinary lesson review, the reader might need the original attempt, the teacher's response and a usable model nearby. For detailed interaction analysis, pauses, restarts and overlaps may deserve closer treatment. For a practice handout, the reader may need only selected target examples, but that handout should carry its own label. British Council practitioner guidance discusses correction in relation to communication, accuracy and the activity in progress. The editorial implication proposed here is to ask what the lesson was trying to make visible. A transcript prepared for reviewing question formation should not normalize away auxiliary errors merely because a general writing tool considers them awkward. Purpose guides selection; it does not authorize inventing better source speech.
A modest editing agreement can fit into a few sentences at the beginning of a working copy. State that spoken grammatical forms are retained, recognition corrections require a sound check, teacher additions are marked and uncertain wording stays unresolved until reviewed. If routine fillers will be removed, say which ones and under what conditions. Give the untouched draft a different filename from the reading version. That small separation creates an audit trail without requiring elaborate software. It also makes later disagreement easier to handle: a colleague can compare the revised passage with the retained source instead of guessing which version is original. The examples below offer one workable agreement, not a mandatory research notation or an institutional assessment rule. A school or project may need its own conventions. Follow the requirements that actually apply, while keeping the fundamental distinction between reported speech and editorial help visible to the reader.
References: Error correction in language teaching, CHAT transcription manual
1. Preserve the lesson sequence before polishing it
1.1 Build a small case that exposes the editing risk
Use a turn identifier and a speaker label for every contribution. Then distinguish the initial attempt, teacher response, retry, new prompt and instruction where those roles are known from the exchange. These are working annotations for review, not words to insert into the speaker's mouth. They help the editor notice when two similar sentences perform different jobs. A teacher saying a target sentence and a learner repeating it afterward are separate events even when the wording matches. A transcript that preserves only the best version of each sentence cannot show that relationship. Nor should role labels be assigned from grammar alone: teachers can deliberately produce errors for discussion, and learners can produce accurate sentences on their first attempt. In a real recording, identify the speaker and interaction from available evidence. When that evidence is inadequate, leave the role uncertain rather than making the page look more complete through inference.
The following original text case was written for this guide; it is not a classroom excerpt. L1, Learner: “Yesterday I go to the library.” T2, Teacher: “Yesterday I went to the library. Try the sentence again.” L3, Learner: “Yesterday I went to the library.” T4, Teacher: “Now ask a question about last week.” L5, Learner: “Did you went last week?” T6, Teacher: “After did, use the base form: did you go?” L7, Learner: “Did you go last week?” T8, Teacher: “Keep both first attempts in the practice notes.” There are eight turns, four assigned to each speaker, with two initial attempts and two retries. Their roles are known because the example was authored that way. No timing or pronunciation can be recovered from this text. Its limited purpose is to make editing choices observable without borrowing a real learner's mistakes or presenting invented dialogue as a measured teaching result.
References: Feedback and learner response as separate events, Documenting transcript conventions
1.2 Notice what a perfect-looking rewrite removes
Now make just two replacements. Change L1 to match L3, and change L5 to match L7. The counterexample still has eight turns, four learner contributions, four teacher contributions, the same order and plausible sentences. A quick count would find nothing missing. Nevertheless, the two initial forms have disappeared. The teacher's responses remain on the page, but the reader cannot see what prompted them. This is why completeness checks need more than paragraph totals. A transcript can preserve every speaker slot while changing the instructional story inside those slots. For a lesson about past forms, those two small replacements are material edits. They change the apparent starting point of both exchanges. A tidy export, a successful spelling check or a matching turn count cannot establish that the cleaned text still represents the original attempts. Comparison must reach the wording that carries the lesson's relevant contrast.
The opposite mistake is to treat every visible error as untouchable. A recognizer may print a verb that the learner did not say, merge a teacher's interruption into a learner's line or omit an audible auxiliary. Fixing those mistakes can restore the record. The decisive question is what supports the replacement. If the recording clearly contains “went,” correcting an erroneous “go” is a transcription correction. If the recording contains “go” and the editor prefers “went,” the change is a teaching correction. The two actions can produce the same final string while having different justifications. Record the reason where it matters. Do not label the authored counterexample above as a recognition failure: it contains no recording and no recognizer. Its value is narrower and concrete. It shows that fluent rewriting can erase an attempted form while leaving the document's outer structure intact and superficially reassuring.
References: Preserving speech rather than tidying grammar, CHAT distinctions for correction and reformulation
1.3 Keep a practice sheet separate from the record
A practice sheet may legitimately present target forms, teacher explanations, new exercises, and selected examples. It serves a different purpose from the transcript. In the authored case, a short practice sheet can extract “Yesterday I went to the library.” from T2 and “did you go?” from T6, labeling both as teacher models. The second extraction deliberately stops at the words the teacher actually supplied. Adding “last week” would create an adapted practice sentence, which is also useful if labeled as an adaptation. That distinction keeps attribution accurate without preventing teaching. A learner should be able to tell whether a line is something they said, something their teacher modeled or something prepared afterward for practice. The practice sheet may omit most of the exchange. Its selectivity becomes a problem only when it is presented as a complete transcript or used to imply that the learner independently produced its polished examples.
The original case was checked as text in three separate forms: a reading copy that retained all eight turns, a counterexample that changed exactly L1 and L5, and a two-item teacher-model worksheet. Eight consistency checks passed, including preserved source identifiers, the two changed turns and the teacher origin of both extracted models. This is text bookkeeping, not a language assessment. It does not test whether a student learns more from one layout, whether software identifies speakers correctly or whether any transcription service recognizes these sentences. A teacher can adopt the same separation without running a script: keep the record, mark teaching additions and compare the selected models with their source turns. The important result is inspectable provenance. Readers can follow each model back to a specific teacher contribution instead of trusting that a fluent sentence somehow appeared in the lesson in the form shown.
References: Purpose-specific transcription decisions, Feedback and uptake research abstract
2. Make the first editing pass about the source
2.1 Start with recoverable speech and a protected draft
Before editing, establish which recording produced the text, which lesson segment it covers, which draft is untouched, which speakers are known and which conventions you will apply. A downloaded lesson video, a clipped exercise and a complete session recording are different source scopes. If the transcript begins halfway through a correction exchange, the missing prompt may explain an otherwise puzzling response. Note the boundary instead of inventing a lead-in. Keep the source recording recoverable while you work, and distinguish local time within a clip from time within the original session. A transcript generated from a shortened recording should not acquire the full session's title and duration merely because the names look similar. These details make the later sound check possible. Without them, an editor can spend considerable effort polishing a sentence that belongs to another version of the lesson or cannot be located reliably in the retained recording.
WhisperAITranscribe, an audio and video transcription tool, can help produce the draft to review once you have a usable source. The editorial responsibility begins with that draft rather than ending there. Keep the initial output before making changes, then review the passages that matter for the lesson's target language. This guide does not establish a native lesson-analysis feature, automatic teacher-model extraction or classroom speaker assessment in the product. Those are separate capabilities and should not be assumed from the existence of a transcript. If you work in another editor afterward, carry the source identity and turn references with the text. The resulting document should make it possible to return from a disputed sentence to the relevant recording, not merely to another increasingly polished copy. That is especially valuable when several lesson clips contain the same exercise prompt and very similar answers.
References: Documenting source and transcription decisions, Timed text cue model
2.2 Correct recognition errors without correcting the learner
For each doubtful passage, distinguish heard wording, draft wording, proposed replacement, supporting evidence and remaining uncertainty. Listen to enough surrounding speech to identify the turn, rather than deciding from the suspicious word alone. A teacher may be quoting the learner's earlier attempt; a learner may be asking whether a form is wrong. Neither belongs to the same editorial category as an accidental omission by the recognizer. If the sound is clear, make the transcription correction and note it when the target form is consequential. If the sound is unclear, retain an uncertainty marker and a source location. A grammar rule can tell you which sentence is conventional; it cannot tell you which sentence was spoken. This matters most when the plausible alternatives are precisely the contrast under discussion, such as a base form after an auxiliary or a singular ending that may have been clipped by the recording.
Avoid turning confidence into a binary choice between leaving everything untouched and silently fixing everything. A working note might say that the auxiliary is audible, the following verb is uncertain and the teacher supplies a clear model in the next turn. That is more useful than attributing the model backward to the learner. It allows a later reviewer to focus on the unresolved sound while keeping the surrounding teaching sequence readable. The UK Data Service's example instructions explicitly distinguish punctuation from changing grammar and require unclear portions to be marked rather than guessed. Those are project-specific instructions, not a universal lesson protocol; they provide a clear illustration of the boundary. The workflow proposed here applies that boundary to lesson review. If the audio remains indecisive, an honest gap preserves more information than an attractive sentence whose certainty comes entirely from the editor's knowledge of the expected answer.
References: Unclear wording and grammar in an example protocol, Transcribing revisions distinctly
2.3 Use punctuation to clarify the exchange conservatively
A first reading pass can improve turn boundaries, sentence punctuation, capitalization and spacing without substituting target-language forms. Even these changes deserve care. A question mark can turn a repeated model into an apparent question. Quotation marks can distinguish a teacher mentioning a word from using it in an ordinary sentence. A colon may make an explanation easier to follow, but it should not imply an exact pause pattern. Where punctuation would settle an uncertain interpretation, retain a brief editorial note instead of forcing the speech into a neat written category. In the authored case, T6 is an explanation followed by a model question. Presenting the whole line as an unmarked learner question would change both its speaker and function. Good punctuation helps the reader see the teacher's explanation while leaving the actual words intact. It should not manufacture a different interaction simply because the written result scans more smoothly.
Choose a manageable convention for incomplete speech. If a learner abandons a sentence and starts again, a dash or a marked restart can preserve the relationship without reproducing every acoustic detail. Explain the convention once so another editor does not later interpret it as missing text. Do not use punctuation as a substitute for information you have not checked. A long row of dots does not measure a pause, and a comma cannot prove that two speakers' fragments belong together. For a simple review copy, the aim is enough structure to follow who said what and what came next. A more detailed analysis may need another representation. Keep that decision explicit. The reading version should remain a readable rendering of the source exchange, while any separate analytic notation records additional decisions under its own conventions and review standards.
References: Example punctuation and transcription conventions, CHAT transcription categories
3. Retain details that carry the teaching point
3.1 Distinguish retries from disposable duplication
Repeated wording can mark a requested retry, an independent contribution or a restart within one turn. Those functions should not be collapsed through a blanket instruction to remove repetitions. TalkBank's CHAT manual distinguishes retracing with correction from fuller reformulation, and discusses repeated words across interactions as potentially independent communicative acts. You do not need to adopt its notation to use that conceptual distinction. In the authored case, L3 and L7 are later learner turns responding to instruction. Removing either would erase the learner's immediate response; merging it with the teacher's model would obscure ownership. Conversely, a duplicated paragraph introduced by copying a transcript twice is a document error rather than a spoken retry. The editor must locate the repetition in the source process before deciding what to do. Similar text alone cannot establish whether an item is accidental duplication, a fresh attempt or a deliberate teaching move.
If a drill contains several nearly identical responses, preserve enough context to make the selected scope clear. A reading extract may include the first exchange and a note that later practice was omitted, provided the note is accurate and the complete record remains available where appropriate. It should not imply that every learner response has been reproduced. If exact turn counts matter, do not replace repeated turns with a vague summary. If counts do not matter, avoid adding a page of mechanically repeated sentences merely to imitate completeness. The useful decision is whether a reader needs each event to answer the stated review question. In all cases, distinguish an omission made for presentation from an omission already present in the recording. An editor's selection is a new action; it should not disappear into the transcript as though the selected text were the whole lesson.
References: Repetition and retracing distinctions, Correction in relation to lesson activities
3.2 Keep pronunciation claims within what you checked
An orthographic transcript records words in writing; a pronunciation observation makes a claim about sound. Do not treat one as an automatic substitute for the other. A correctly spelled target word can appear in a draft even when its pronunciation is uncertain. A misspelled word can reflect a recognition error rather than a learner's spoken production. If pronunciation is the lesson focus, return to the recording and specify what you can actually hear. A teacher's feedback about a vowel can be transcribed as teacher speech without the editor independently endorsing the diagnosis. Likewise, a learner repeating the printed word after feedback does not establish that the sound changed. The text example in this guide contains no audio, so it supports no pronunciation judgment at all. Maintaining that limit is part of accurate transcription, not a reason to avoid making useful notes about a real recording that has actually been reviewed.
When a pronunciation note is necessary, keep it outside the reported words and identify its basis. A note may record that the teacher modeled a word, asked for another attempt or referred to stress. A stronger observation about the learner's sound needs the appropriate listening and review. Do not add technical phonetic symbols merely because they make the page look scholarly; use a convention that the intended reader and reviewer understand. Nor should a speech recognition correction be repackaged as a diagnostic score. The recognizer's preferred spelling is evidence about its output, not an assessment of the speaker. If a learner asks why the transcript displays a target form they do not remember saying, compare the relevant audio rather than defending the polished text. The same approach prevents an edited transcript from becoming an unintended and unsupported pronunciation report.
References: Choosing the detail a transcript preserves, Speech detail in transcription conventions
3.3 Separate translations from the words being translated
A multilingual lesson may contain source-language speech, teacher translation and editorial translation in the same document. Label their origins before making the page uniform. If a teacher explains an English auxiliary in another language, that explanation is part of the exchange. Replacing it with fluent English changes what the teacher said. A later translation can sit nearby as reading assistance, with a label that prevents it from being mistaken for the original turn. Keep target-language examples intact where their form is the teaching point. A translated equivalent may convey the topic while losing the very auxiliary, ending or word order being discussed. The editor should therefore decide what needs to remain visible at the level of form as well as meaning. Where an unfamiliar phrase cannot be checked, carry the uncertainty forward rather than normalizing it into the language the editor knows best.
A draft from WhisperAITranscribe can enter this workflow as a starting text, while the editor determines which passages require bilingual review. Keep the original script and any translation in separate, identified passages instead of silently swapping one for the other. If you prepare a practice translation of the authored example, label it as an adaptation and retain L1 through T8 as source references. That lets a teacher explain the idea in another language without pretending the translated dialogue occurred in the lesson. Avoid assigning one speaker's translation to the other speaker's original turn. When a learner translates a teacher's model aloud, that response is a separate event with its own wording. The document becomes easier to trust when each language choice has a visible owner, even if the final reading copy contains fewer acoustic details than a specialist transcript would retain.
References: Explicit transcription conventions, Maintaining separate interactional contributions
4. Make annotations useful without overstating them
4.1 Separate what happened from what you infer
Keep observed wording, teacher feedback, editorial explanation and assessment distinct. The authored case shows L1 followed by T2 and L3. It does not establish why the learner chose the first verb, whether the teacher had addressed it earlier or whether the retry would transfer to a new sentence. A note can accurately describe the visible sequence without filling in those missing facts. For a real lesson, a teacher may know additional context; identify that context as a teacher note rather than inserting it into the transcript. Avoid converting every deviation into a diagnosis. A form may be quoted, playful, task-driven or simply uncertain in the recording. When the purpose is lesson review, an actionable description such as “compare the first attempt with the teacher's model” is often more useful than a confident claim about the learner's underlying knowledge. It also leaves room for the learner to explain what they intended.
The research abstract cited here treats feedback moves and learner responses as distinct parts of classroom interaction. It does not validate the eight-turn example or establish a general learning outcome for this editing method. That limit matters when preparing a report for somebody who was not in the lesson. A well-organized transcript can show that a response followed a model, but the sequence alone cannot prove durable improvement. If a teacher later evaluates a comparable task, keep that evidence separate and date it. Do not upgrade an immediate retry into a claim of mastery through a heading or summary sentence. The proposed annotations are meant to preserve the basis for discussion. They should make interpretation more accountable, not disguise interpretation as something the recording itself directly demonstrates about every learner or every future use of the language form.
References: Original feedback and uptake study abstract, Context-sensitive correction guidance
4.2 Use role labels that remain open to review
After producing a draft with WhisperAITranscribe, an editor can add stable turn IDs, known speaker labels, source locations, working role labels, review notes and revision dates in the review document. These are proposed editorial practices, not a claim about automatic product features. Keep a role label modest when the source is ambiguous. “Teacher model” may be clear in the authored T2, while a real speaker's rising repetition could be a confirmation request rather than a model. A label such as “possible clarification request; review audio” communicates the uncertainty without stopping the whole workflow. If the speaker is unknown, use a stable neutral identifier instead of guessing from vocabulary, accent or grammatical accuracy. The record should not become more certain simply because later editing requires a clean label. A visible unresolved label is easier to correct than a confident but invented attribution repeated throughout a polished document.
One useful convention is to keep working annotations in a separate copy or margin, leaving the reported wording easy to identify. The reader can then distinguish an editorial classification from an utterance. If a teacher changes a classification during review, retain the turn ID so notes and practice sheets still point to the same passage. Do not renumber everything merely because a heading changes or a paragraph is split. If two fragments are found to belong to one contribution, explain the merge in the working notes. If one line contains two voices, split it with a source check rather than silently deciding which voice matters more. Small, traceable corrections keep the review process proportionate. They also prevent a later document export from turning provisional labels into apparently verified statements about the speaker or the function of the entire exchange.
4.3 Do not invent timing from the layout
A review copy prepared after WhisperAITranscribe may include clip references, cue intervals, turn locations and untimed notes, but those categories should remain distinct. A subtitle cue tells a player when to display text. It is not automatically a word-level boundary or a measurement of how long a learner considered an answer. The W3C WebVTT model specifies timed cues; a lesson editor still needs the underlying sound and appropriate boundaries for any claim about response timing. If the teacher's prompt and learner's answer share one cue, subtracting its endpoints gives the cue's interval, not the learner's pause. A document's paragraph break has even less temporal authority. Do not convert line positions into speaking times because the page needs neat timestamps. Where no timing was measured, leave the turn untimed or attach only a source reference that genuinely helps the reviewer find it.
For a clipped lesson, record whether locations refer to the clip or the original session. A turn at one minute in an excerpt may occur much later in the full recording. If you change the excerpt, old locations may no longer point to the same sound. Keep enough source identity to detect that mismatch before a teacher follows a reference and hears the wrong exercise. The authored eight-turn case deliberately includes no timestamps. Adding evenly spaced times would create an appearance of measurement without evidence. That restraint also applies to summaries such as “the learner answered immediately” or “needed a long pause.” Such statements require more than the order of text turns. The record can preserve the sequence accurately while staying silent about timing that was never observed. Precision in a transcript should come from checked information, not from decorative numbers.
References: WebVTT cue timings and structure, Detail and conventions in transcription
5. Review the copy against its intended use
5.1 Compare edits where they can change the lesson
Prioritize target forms, speaker ownership and instructional sequence when reviewing changes. A replaced verb, a deleted retry or a teacher sentence moved into a learner paragraph deserves closer attention than a corrected double space. Compare the working draft with the retained original, and listen where a material correction depends on sound. In the authored case, the important comparison is not whether both versions contain eight turns. It is whether L1 and L5 retain their initial wording and whether the responses remain attached to the correct speakers. A reviewer can then understand why T2 and T6 exist. For a real lesson, apply the same logic to the language being practiced. If articles, pronouns or endings are the focus, small words become high-priority checks. Review effort should follow the consequence of an edit rather than the visual size of the changed passage on the page.
Ask another reviewer a concrete question when their help is needed. “Check whether this auxiliary is audible” is more useful than “make this natural.” “Confirm which speaker supplies this example” is more focused than “improve the transcript.” A general request for better prose can undo the care taken in earlier passes. Give the reviewer the editing agreement and identify which versions are source, reading copy and practice sheet. If their correction introduces new wording, ask whether it restores the recording or supplies teaching material. Both contributions may be welcome, but they belong in different places. The outcome should be a document whose key changes have reasons a reader can understand. It need not contain an exhaustive log of every punctuation mark. It does need enough traceability that a disputed teaching point can be resolved from evidence rather than settled by whichever version sounds most fluent.
References: Source-faithful editing instructions, Keeping feedback and response distinct
5.2 Hand the learner a usable set of documents
A useful handoff distinguishes the review transcript from the practice resource in plain language. The transcript can show what was said, with uncertainty and editorial conventions explained. The practice resource can present selected models, explanations and new tasks. A learner should not need specialist notation to discover which document records their speech. Put the distinction near the beginning, where it affects interpretation. If only an extract is shared, state the selected scope. If the teacher has added a translation or explanation afterward, identify it as a teaching note. Avoid a filename that simply says “corrected transcript” when it contains both source corrections and rewritten learner answers. That label leaves the most important ambiguity unresolved. A slightly more specific description helps the learner use the material confidently without mistaking editorial help for a record of independent performance in the lesson.
If WhisperAITranscribe supplied the starting draft, keep that workflow detail subordinate to the learner's actual task. They need to know what to review and how to find the relevant passage, not the internal history of every export. A compact note can explain that the reading copy preserves attempted forms while the accompanying practice page contains teacher models. Carry the source turn IDs into the practice page when they help the learner connect the two. In the authored case, T2 and T6 are enough to identify the extracted models without reprinting the entire conversation twice. If you adapt a model into a new exercise, say that it is adapted. That allows useful teaching work to continue while keeping the record honest. The handoff succeeds when the recipient can tell what happened, what remains uncertain and what has been prepared afterward to help them practice.
References: Making transcription conventions explicit, Correction choices in language activities
5.3 Stop when the record is clear enough for the job
Define completion criteria, unresolved passages, review ownership and separate teaching additions before making endless cosmetic passes. In a workflow using WhisperAITranscribe, completion can mean that the relevant turns have been checked, material edits are explained and the reading copy is easy to follow. It need not mean that every sentence sounds like finished written English. A draft from WhisperAITranscribe can remain visibly conversational after careful review because the lesson itself was conversational. Replacing that texture with formal prose may create a more elegant page while making it harder to understand how the exchange unfolded. Decide whether another edit helps the stated reader answer the stated question. If it only makes a learner's original attempt less recognizable, put the improved form in the practice resource instead of continuing to polish the transcript.
Before closing work that began with WhisperAITranscribe, read the final passage as someone who did not attend the lesson. Can that reader identify the learner's first attempt, the teacher's contribution and the later response? Can they distinguish an uncertain word from an editor's explanation? Can they see where an extract ends and whether the practice model was copied or adapted? Those are practical acceptance questions, not a learning score. Keep any unresolved sound check visible and assigned to somebody who can review it. Save the retained draft and the final reading copy with distinct names. Then stop. Further editing should have a concrete purpose, such as resolving a speaker, checking a target form or preparing another exercise. A transcript that remains faithful and usable has reached a better stopping point than one made immaculate through changes nobody can justify from the source.
6. Conclusion
6.1 A readable transcript should preserve what made it a lesson
The useful result of working with WhisperAITranscribe is a draft that can become a source-faithful record, with attempted forms, teacher ownership, learner retries, visible uncertainty and separate practice materials. The central editing decision is whether a change restores the speech or rewrites it for teaching. Keep that distinction visible and many smaller choices become easier. A missing audible auxiliary can be restored after a sound check. An unconventional verb that was actually spoken can remain in the transcript while a target form appears in the practice notes. A teacher's model can be extracted with its source identified. None of those actions requires pretending that a polished page proves a learner's ability. They simply make the documents more useful for reviewing the exchange and planning what to do next. Accuracy includes preserving the parts that an ordinary prose editor might most want to smooth away.
The case in this guide provides a useful final comparison because its limits are unusually clear. All eight turns were invented as text, so their speaker assignments and intended functions are known by construction. The reading version preserves them. The polished counterexample changes exactly two turns, making each initial attempt identical to the later retry. The practice worksheet extracts two teacher models and identifies their sources. Those results demonstrate differences between these text versions; they do not measure a real lesson, a student's progress, speech recognition quality or the effectiveness of a teaching approach. Keep that distinction when applying the example elsewhere. With an actual recording, source checks replace the certainty provided here by authorship. You may discover an incorrect speaker label, an inaudible ending or a missing prompt. Preserve those discoveries in the working copy, then decide what the reader needs to understand the selected exchange. The aim is an inspectable record, not a claim that one editing style answers every pedagogical question. A teacher can still prepare encouraging feedback, clearer explanations and fresh practice tasks. They simply belong beside the record as identified teaching work rather than inside it as speech that never occurred.
6.2 Use a short routine that protects attribution
Step 1: retain the recoverable source and untouched draft, then state what the reading copy is meant to preserve. Step 2: use WhisperAITranscribe for the transcription stage where appropriate, keeping the resulting text separate from later teaching additions. Step 3: check material wording and speakers against the sound, preserve relevant attempts and retries, and leave unresolved passages marked. Step 4: prepare a labeled practice sheet, compare it with its cited source turns and review the final transcript for changed ownership or missing teaching contrasts. The principle linking these actions is traceability. A reader should be able to move from a useful model back to the teacher's contribution, or from a disputed learner form back to the recording. The authored example demonstrates that text counts alone cannot supply that traceability. The proposed routine supplies a way to inspect it without turning ordinary lesson review into an elaborate research project.
For a practical final review, open the transcript and practice sheet side by side. Select one target exchange and follow its references in both directions. Start from the learner's initial attempt, locate the teacher's model and find the later response. Then start from a model on the practice sheet and locate the turn that supplied it. If the model includes new material, check that the adaptation is labeled. If a sound remains uncertain, confirm that the uncertainty has survived the export instead of disappearing into polished punctuation. Finally, inspect the document title and opening explanation: they should describe the actual scope, including any omitted part of the lesson. This small review focuses on relationships that a spell checker cannot verify. It also gives the next editor a clear starting point. They can resolve an identified question rather than repeat the entire cleanup or make another unrestricted prose pass. Where the source cannot settle a point, preserve that limit and continue with the portions that can be used responsibly.
6.3 Keep the original attempt visible beside the help
A clear lesson record protects the starting point, the response, the source of help and the limits of the evidence. In the eight-turn example, changing just two learner turns makes the conversation look more accurate while removing both initial forms. That is a small edit with a large interpretive effect. Use WhisperAITranscribe to begin a transcription workflow, then make the review copy readable without crediting a learner with a teacher's language. Keep corrected examples and new exercises available in a separate teaching resource. The final check after working with WhisperAITranscribe is whether another reader can still tell what was attempted, what was supplied and what happened next. When those relationships remain clear, the transcript can support a useful discussion of the lesson while leaving pronunciation, independent ability and lasting learning to evidence that actually addresses those questions.
References: CHAT manual on distinct speech events, Corrective feedback and learner response