To translate interview transcripts and oral histories accurately into any language, preserve who is speaking, what each person actually claims, and how certain the speaker is. The most useful translation is not the smoothest story: it is the version in which a historian, family member, student or researcher can still identify a question, a recollection, a correction, a hesitation and an editor’s intervention without confusing them.
People searching for interview transcript translation, oral history translation, multilingual interview transcripts or AI translation of recorded interviews often need a practical workflow rather than a list of word equivalents. Begin with the approved source recording or transcript, map speakers and timestamps, keep attribution and uncertainty intact, then check the completed target-language text against the original turn by turn.
This guide explains how to translate oral-history interviews from first principles, including spoken repairs, dialect, code-switching, sensitive memories, disputed claims, redactions and publication excerpts. The central rule is simple: make the conversation understandable in another language without silently changing its evidence. Every example below is an illustrative English-language editorial scenario, not a purported translation of an actual archived recording.
What a trustworthy oral-history translation must preserve
An interview is a sequence of accountable speech acts. Its facts include not only the content of sentences but also their authors, the order of questions and answers, the strength of claims and the gaps in knowledge. A translation must retain these relationships even where the target language uses different grammar or conversational habits. A speaker’s uncertainty is information; an interviewer’s suggestion is information; a correction is information. Removing any of them changes what later readers can legitimately conclude.
A practical speaker–meaning–certainty–provenance ledger contains five columns: turn identifier, speaker, proposition, certainty or source of knowledge, and editorial status. The target transcript should reproduce that ledger accurately. Use it before line editing, because fluency checks do not reliably reveal misplaced responsibility. Good prose can be misleading prose when the attribution map is wrong.
The ten editorial principles
Start with a speaker-and-source map
Before translating, identify every interviewer, narrator, witness and editor, then record how the source distinguishes their voices. A speaker turn is evidence about who made a statement, not merely a typographic convenience. The translation should retain speaker labels, order, question boundaries and quotation boundaries. If the archive provides a recording, transcript and interview summary, do not quietly treat those as interchangeable versions. The recording may carry timing or audible hesitation that a cleaned transcript intentionally omits. Name the version used and preserve the editorial status of any supplied transcript.
Distinguish transcription, translation and editing
A verbatim transcript, an edited transcript, a translated transcript and a published narrative serve different readers. Translation changes the language, not the documentary status. Removing a repeated word for readability is an editorial act and should follow the institution’s transcription policy rather than a translator’s instinct. If the commission requires a readable rather than verbatim version, declare the intervention and retain a reliable path to the unedited record. Never represent an invented smooth sentence as a precise quotation from a speaker.
Protect uncertain memories
Oral histories frequently describe events many years later. The narrator may know some dates exactly, infer others and openly doubt a sequence. Preserve phrases such as ‘I think’, ‘perhaps’, ‘around that time’ and ‘I cannot remember’. Turning qualified recollection into confident historical fact damages the evidence. Equally, avoid weakening a firm first-hand statement merely because the translator finds it surprising. A careful translation reflects the speaker’s certainty, not the translator’s estimate of what probably happened.
Separate the speaker’s claim from the historical record
A narrator can recall an event differently from an official document or another witness. The translator is not the adjudicator. If verification is required, put factual corrections or contextual notes outside the attributed speech with a clear editorial label. Do not silently correct names, dates, places or causal explanations in a quotation. Readers should be able to tell what the interviewee actually said, what the interviewer asked and what an archivist later verified.
Listen for discourse rather than single words
Interview speech is interactive. Small words such as ‘well’, ‘right’, ‘you know’, ‘actually’ or ‘so’ can signal hesitation, agreement, a change of topic or disagreement. A word-for-word replacement may give the wrong interpersonal function. Translate the move the utterance makes in the exchange while retaining meaningful emphasis and hesitation. This is especially important when a pause precedes a sensitive claim, a laugh softens disagreement or a question invites a particular answer.
Make ambiguity visible when it cannot be solved
An unclear pronoun, noisy recording, unfamiliar nickname or ambiguous cultural reference should not become a guessed certainty. Consult surrounding turns, any available recording, approved terminology notes and the commissioning archive. Where uncertainty remains material, use a consistent marker such as ‘[unclear]’ or an editor’s note according to the transcript policy. A visibly unresolved word is more honest than an attractive but fabricated answer.
Treat accents, dialects and multilingual turns respectfully
A speaker’s accent does not mean the person is uneducated, comic or confused. Do not imitate accent through caricatured spelling unless the research purpose specifically requires documented phonetic representation. Code-switching, borrowed terms and dialect vocabulary can communicate community identity, intimacy and audience alignment. Decide which words should remain in the original language, which need a gloss and which can be translated normally. Preserve meaningful differences in register without stereotyping the speaker.
Preserve oral-history ethics and access conditions
Interview collections may include consent limits, embargo periods, pseudonyms, redactions and sensitive information about third parties. These are not stylistic suggestions. Translation must respect the permissions attached to the source and avoid re-identification through explanatory glosses. A redaction in the original should not be reconstructed from context in the translated version. Record whether the speaker approved quotation, translation or public circulation as separate matters; one permission does not automatically imply all three.
Design a two-layer quality check
First check content fidelity: speakers, events, time references, certainty, quotations, contradictions, redactions and editorial notes. Then check target-language readability: clear sentences, coherent paragraphing, natural speech patterns and consistent terminology. A translation can pass one layer and fail the other. In final review, compare the target transcript back to the source turn by turn, then read the target alone to see whether a reader could reconstruct the conversation without mistaking an editor’s voice for a narrator’s.
Keep the work auditable
Use stable turn numbers, timestamps or line identifiers so a historian can trace a translated quotation to the source. Preserve version numbers and correction logs when interviewees amend transcripts. If an AI tool proposes a translation, its output needs the same source comparison and human editorial approval as any other draft. Avoid uploading restricted interview data into unapproved systems. The record of choices should allow a future reviewer to understand why a culturally specific term was retained or a passage was marked inaudible.
Case laboratory: twenty-eight decisions that change the historical meaning
The following cases show where a fluent transcript can become unreliable. The illustrative source snippets are written in English so the evidence relationship is visible regardless of the reader’s language pair. A real translation should use the natural conventions of its target language, while preserving the documented intent, speaker and certainty. Read each example as a decision rather than a phrasebook entry.
1. The first answer after a leading question
Illustrative source moment: ‘You were frightened, weren’t you?’ — ‘I suppose I was, yes.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Turning the response into ‘I was frightened’ erases the interviewer’s framing and the narrator’s cautious assent. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep the question and answer as separate turns. Translate ‘I suppose’ as a qualification rather than dropping it for elegance. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Read the target exchange without the source: can a historian still see that the interviewer suggested the adjective? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
2. A self-correction about the year
Illustrative source moment: ‘We moved in ’68—no, wait, ’69, after my sister was born.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Choosing only 1969 conceals a visible correction, while printing both as equally certain dates misreads the repair. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Retain the correction in an approved readable-verbatim form; identify the later year as the speaker’s corrected recollection. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the reader distinguish an error repaired by the narrator from a discrepancy inserted by an editor? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
3. A deliberate pause before a difficult memory
Illustrative source moment: ‘I heard the knock … and then my father went outside.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Removing the pause may flatten its effect; treating it as proof of trauma is an inference the record does not establish. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Follow the source transcription policy for meaningful pauses. Do not invent a psychological explanation or add dramatic punctuation. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Check that pause notation matches other interviews and that the translator has not interpreted silence as a fact. Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
4. Overlapping speech during an interview
Illustrative source moment: A narrator starts ‘The train had already—’ as the interviewer begins ‘Was that before dawn?’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Combining both voices into a single polished sentence changes the order of interaction and obscures interruption. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep overlapping turns and interruption marks where the archive requires them; preserve the narrator’s incomplete phrase. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can two reviewers follow who spoke first and whether an interrupted claim was actually completed? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
5. Backchannels that are not full agreement
Illustrative source moment: The interviewer says ‘Mm-hm’ while the interviewee describes a disputed event.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Rendering ‘Mm-hm’ as ‘I agree’ falsely creates corroboration by the interviewer. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Treat the sound as an acknowledgement or listening cue unless context clearly supports assent; use the transcription convention. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Ask whether the chosen phrase would lead a reader to infer an endorsement absent from the recording. Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
6. Uncertain identification of a person
Illustrative source moment: ‘I think her name was Mrs Tan, or perhaps Chen.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Deleting one name or choosing the more familiar one creates a false identification. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Preserve the competing possibilities and mark any external verification as a separate editor’s note. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Would a future researcher be able to search both possible names without mistaking either for confirmed identity? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
7. An indirect quotation inside a story
Illustrative source moment: ‘My uncle said, “Don’t sign anything yet,” and I listened.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: A careless quotation layout can make the uncle’s words appear to be the interviewee’s advice today. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Maintain nested quotation and the attributed speaker; translate imperative force without changing who issued it. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Check who gave the warning, to whom it was addressed and whether the narrator followed it. Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
8. A memory with a relative date
Illustrative source moment: ‘That happened a couple of weeks after the festival.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Replacing the phrase with a fixed calendar date introduces a historical calculation the narrator never made. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Translate the relative interval and the festival reference; add a separately sourced note only if dates are needed. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Verify that neither approximate duration nor reference event becomes a fictitious exact date. Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
9. An unfamiliar nickname
Illustrative source moment: ‘Everyone called him Ah Boy, even when he was older.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Translating the nickname as if it were a legal given name may erase its social meaning. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep the established name form in the transcript and offer a short gloss in a separate note when useful. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the reader tell the difference between a nickname, a name and an editor’s explanation? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
10. A dialect word for a household object
Illustrative source moment: ‘She kept it in the tingkap,’ with the term documented as a regional window word.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: An overly specific modern replacement can narrow the reference to a feature the speaker did not identify. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Check the dialect and context, choose a defensible equivalent, and preserve the original term in a note if the object matters. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Could another competent translator reconstruct the same object from the supplied context? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
11. Irony delivered with laughter
Illustrative source moment: ‘Oh, the management was wonderful,’ followed by a short laugh and a story about late wages.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: A literal compliment without laughter or subsequent context can reverse the speaker’s attitude. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Preserve the utterance and recorded laughter while keeping the irony available to the target reader through context. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the translated passage read like unqualified praise when the exchange suggests criticism? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
12. Strong emotion without invented diagnosis
Illustrative source moment: ‘I couldn’t speak for a while,’ the speaker says, with a long silence.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Calling the narrator ‘traumatised’ is a diagnosis, not a translation of the words. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Render the statement and silence accurately. Any emotional interpretation belongs in analysis attributed to the analyst. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the translated text distinguish observable behaviour from explanations supplied by a researcher? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
13. An allegation about a third party
Illustrative source moment: ‘People said the supervisor took the money. I never saw that myself.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Dropping the second sentence turns reported gossip into an apparent first-hand accusation. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Preserve reportative language, attribution and the explicit limitation on personal knowledge. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: List every claim in the translated passage and mark whether it is witnessed, heard from others or inferred. Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
14. A question that changes topic
Illustrative source moment: ‘Before we talk about the strike, can you tell me about your first job?’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: A smooth narrative rewrite can erase the interviewer’s ordering and a change of topic. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Maintain the question’s topic shift and the fact that the narrator is responding to a new prompt. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Could a reader locate the moment the interviewer steered the conversation? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
15. An inaudible proper noun
Illustrative source moment: ‘We took the bus to [inaudible] and walked home.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Guessing a plausible town name can contaminate later geographic research. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Use the archive’s approved unreadable-audio marker, with a timestamp that permits another listener to review. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the final edition make uncertainty clear rather than hiding it in a confident place name? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
16. Editor-supplied words in brackets
Illustrative source moment: ‘I spoke to [the headmaster] next morning’ where the bracketed phrase was supplied during transcription.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Removing brackets makes an editorial clarification appear to have been spoken. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Retain a consistent distinction between spoken words and bracketed editorial insertions. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can a reader trace which words were audible and which were supplied for clarity? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
17. Translated segments aligned to timestamps
Illustrative source moment: A recording has one segment at 00:14:20–00:14:45 containing two speaker turns.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Moving the second speaker’s line to the next minute may break audio alignment. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep turn anchors linked to the original timing even if target-language sentence lengths differ. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Open the recording at the cited timestamp: is the translated claim still traceable? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
18. A pseudonym protected by consent
Illustrative source moment: The archive uses ‘Participant R’ because the interviewee approved anonymous quotation only.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Reinstating a real name gleaned from context violates the record’s access conditions. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Respect the archive’s pseudonym and redaction map; do not reveal identities through footnotes or extra biographical detail. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Could the target transcript identify a person more readily than the source? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
19. A mixture of two languages
Illustrative source moment: The speaker says an English sentence, switches to a family-language expression, then returns to English.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Flattening every register into one formal voice hides a deliberate audience or intimacy shift. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Record meaningful code-switching, retain a term with an explanatory gloss where it carries cultural force, and avoid needless exoticising. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the translation preserve the speaker’s relationship to the quoted expression? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
20. A sigh followed by disagreement
Illustrative source moment: The respondent sighs before saying, ‘That is not how I remember it.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Changing the sentence into a polite agreement to reduce conflict destroys its central meaning. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep explicit disagreement and the recorded vocal cue without inflating the cue into hostility. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the reader still identify two conflicting accounts? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
21. A pronoun with more than one possible referent
Illustrative source moment: ‘He gave it to him the next day’ follows references to three male relatives.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Selecting a particular name from context may create a chain of custody never stated. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Preserve ambiguity where possible or flag it with a clear note. Consult adjacent turns before expanding pronouns. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the target name a recipient that the original left unresolved? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
22. A passage removed by the archive
Illustrative source moment: ‘I then went to [redacted] before returning home.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Inferring the deleted place from the next line defeats the source’s deliberate suppression. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep the redaction visibly intact in the translated transcript and apply the same access restriction. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Could a reader discover the withheld information only because the translation has made it explicit? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
23. A narrator quoting a slogan
Illustrative source moment: ‘They shouted, “Bread and work!” over and over.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Replacing the historical slogan with a modern political motto may alter cultural and historical specificity. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Translate the slogan’s meaning and record its original form if archival research depends on exact words. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the quoted phrase be traced to the source without conflating it with a later slogan? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
24. A sentence abandoned mid-thought
Illustrative source moment: ‘If I had told the manager, then … anyway, we stayed.’. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Completing the conditional invents an outcome and a motive that the narrator did not state. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Use established interruption notation; preserve the transition to a new topic rather than finishing the thought. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the target retain the distinction between an unfinished possibility and the described decision? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
25. An interviewee amending the transcript
Illustrative source moment: After recording, the speaker changes ‘three brothers’ to ‘two brothers’ on the approved corrected transcript.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Silently mixing the audio version with the corrected transcript creates an undocumented hybrid. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Translate the designated authorised version and log any correction against stable line numbers. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Can the archive recover exactly which version formed the basis of the published quotation? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
26. Words spoken in anger
Illustrative source moment: ‘You people never listened,’ the narrator recalls telling a council officer.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Softening this to ‘I felt unheard’ changes an accusation into a private feeling and removes the original address. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Keep second-person address and intensity while choosing natural target-language syntax. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Does the translated sentence preserve both who was confronted and the nature of the complaint? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
27. Culturally specific kinship language
Illustrative source moment: The speaker calls an older family friend ‘Auntie’ throughout although no biological relationship is claimed.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: A literal family relation may falsely imply kinship, while stripping the title loses social respect. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Preserve the social function through a context-aware equivalent or retain the address with a brief gloss. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Could the reader wrongly create a family tree from the translated form? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
28. A short excerpt selected for publication
Illustrative source moment: An editor wants the sentence ‘We had enough’ as a stand-alone quotation about life in the camp.. This is part of a live conversation, so the translator must interpret the line in relation to its preceding question, the identity of the speaker and the source’s transcription conventions. A target sentence that reads naturally in isolation may still be historically false if it changes one of those relationships.
Where translation goes wrong: Without surrounding turns, ‘enough’ might mean food, space, money or endurance; the isolated excerpt is ambiguous. The problem is not merely a choice of vocabulary; it is a change to what a future reader would reasonably believe about the original event or exchange.
Better editorial decision: Check the surrounding question and supply context outside the quotation, not as silent additions inside it. Preserve the minimum context needed to understand the utterance, but do not invent missing words or remove evidence of uncertainty simply to make the target-language passage elegant.
Verification: Would a reader interpret the excerpt the same way after seeing the full exchange? Use the source turn identifier or audio timestamp when checking the result. If two reviewers cannot resolve a consequential difference, document the unresolved question rather than hiding it in the published quotation.
An independent end-to-end translation workflow
Stage one — intake. Confirm whether the approved source is audio, a verbatim transcript, a corrected transcript or a public excerpt. Check consent and access requirements before transfer to any translation workspace. Record the interview’s language varieties, speaker list, transcription symbols and desired readership. The delivery agreement should say whether a readable version or a research-grade transcript is expected, because this determines how pauses and spoken repairs are represented.
Stage two — structural markup. Divide the source by speaker turn, assign stable identifiers and mark statements that need special care: contested claims, proper nouns, historical dates, dialect words, quotations, corrections and redactions. Preserve time alignment separately from sentence segmentation. Where source material comes with editorial brackets or footnotes, tag those layers explicitly so a translator cannot accidentally place the editor’s explanations in the narrator’s mouth.
Stage three — drafting. Translate complete turns rather than decontextualised single words. Read back to the previous question whenever the pronoun or level of agreement is unclear. Build and maintain a small terminology list for recurrent places, organisations and culturally specific expressions. When a question has no clear answer, enter a query with the exact turn number. Do not resolve uncertain audio through fluent guesswork.
Stage four — fidelity review. Compare all assertions with the source, checking negation, agency, chronology, evidential phrases, degrees of certainty, quotations, redaction markers and meaningful omissions. A second reviewer should identify the strongest historical claims and verify that they still belong to the same speakers. Corrections should be made in the target manuscript without silently overwriting the history of editorial decisions.
Stage five — publication review. Check grammar and readability without erasing spoken character. Validate spelling of approved names, timestamp anchors, citation format, archive metadata, consent labels and excerpt boundaries. If the collection later issues an authorised correction, publish a version note or replacement according to the archive’s practice. Translation is ready when a reader can understand the conversation and researchers can audit the evidence.
Practical exercises for editors, students and translators
The quotation provenance drill
Choose six striking lines from a long interview. For each, record the speaker, question prompting it, timestamp, source version and any nearby qualification. Only then create a translated quote. This prevents the popular but damaging practice of extracting an elegant sentence while losing the exchange that gave it meaning. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The certainty ladder
Collect a speaker’s statements that use ‘I know’, ‘I remember’, ‘I think’, ‘perhaps’, ‘someone told me’ and ‘I don’t know’. Translate them in order of decreasing certainty without smoothing them into one register. Ask a second reviewer to rank the target statements independently. Significant rank differences reveal semantic drift. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The provenance split
Highlight interviewee speech in one colour, interviewer prompts in another and editorial additions in a third. Apply the same scheme to the translation. If any target sentence changes colour, investigate immediately: a framing question may have become a witness claim or a bracketed explanation may have become invented speech. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The memory chronology
Write a timeline using only claims in the transcript, marking approximate intervals and corrections. Build a separate timeline from the translation. Do not repair contradictions. The two timelines should agree not because every date is neat but because every uncertainty remains visible. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The privacy test
Review every retained proper noun, workplace detail, place and kinship reference alongside the collection’s permissions. Translate access limitations as carefully as spoken words. Pay special attention to details that seem harmless in one language but become direct clues when explained for another audience. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The audio spot-check
Select the first minute, a passage involving several speakers, a difficult proper noun and the strongest quotation in the article. Listen with the translation. Check turn order, inaudible passages, tone and timestamp alignment. This targeted audit is useful but must supplement, not replace, a complete fidelity review. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The ambiguous pronoun audit
Mark every ‘he’, ‘she’, ‘they’, ‘it’, ‘there’ and ‘that’ that could refer to more than one entity. Where target grammar forces specificity, record the decision and supporting context. If evidence is inadequate, preserve ambiguity through an approved construction or annotation. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
The reader reconstruction test
Give the translated excerpt to someone who has not seen the source. Ask them to write down who spoke, what each person knew, what remained doubtful and whether any facts were externally verified. Compare those notes with the source. Misreadings show where the translation needs clarification without falsifying speech. Write a short justification for each consequential choice, and compare the justification with the original utterance rather than with the elegance of the finished paragraph. The goal is to train evidence-preserving decisions that remain defensible when the translation is quoted elsewhere.
Questions readers ask about interview and oral-history translation
Should every ‘um’ and ‘uh’ be translated?
Only according to the agreed transcript level. Sound-by-sound documentation may be appropriate for conversation analysis, while a readable transcript may omit nonmeaningful disfluencies. Keep hesitation when it changes certainty, turn-taking or emphasis.
Can I correct a remembered date?
Do not change what the speaker said. Put corrections in a clearly identified editorial note with a source, or use an authorised amended transcript. The speaker’s original claim and the established date remain distinguishable.
What if the recording disagrees with the transcript?
Treat it as a version-control question. Determine which document is the approved translation source, flag the discrepancy and request a ruling from the archive or editor. Do not combine convenient pieces silently.
How should a laugh be shown?
Use the recording’s or institution’s standard annotation where laughter is meaningful. Avoid inventing an emotional diagnosis or using exaggerated punctuation. A laugh can show humour, discomfort or simply social pacing.
Should I translate a speaker’s accent into nonstandard spelling?
Usually not. Respectful transcription captures words and meaningful features without caricature. If phonetic evidence is central to the study, use a documented linguistic convention designed for that purpose.
Is machine translation safe for oral histories?
It can generate a working draft, but it cannot certify speaker attribution, uncertainty or permissions. Use an approved system for sensitive data and review every turn against the source with a competent human reviewer.
How do I handle an unknown name?
Check the audio and collection metadata first. If identity remains uncertain, preserve the uncertainty rather than choose a plausible name. Make any editor-supplied identification explicit.
What is the best final check?
Read the target alone, reconstruct a turn-by-turn account of who asserted what and at what level of certainty, then compare it to the source. Audit redactions, notes, timestamps and version history separately.
Final translation quality checklist
Before a transcript leaves the editorial workflow, verify the speaker map; question-and-answer sequence; source version; timestamps; names and aliases; approximate dates; repairs and corrections; relevant hesitations; quoted speech; dialect and code-switching; indirect evidence; ambiguities; editorial brackets; redactions; permissions; target-language readability; and traceability to the recording. Recheck the most quotable sentences because they are most likely to circulate without surrounding context.
The deepest test is not whether the translated narrator sounds polished. It is whether the narrator still sounds like the same accountable witness: remembering some things clearly, doubting others, responding to a particular interviewer and speaking under specific documentary conditions. Preserve that structure and translation becomes a dependable bridge between readers and the historical record.
Continue learning: Master Art of Translation · Translate Series Hub · Complaints and Grievance Responses.