The Tension at the Centre
News video translation sits between two requirements that pull against each other.
Speed is intrinsic to the product. A translated report published two days after the event has limited value, and in fast-moving stories it may be superseded before it appears.
Accuracy is intrinsic to the product's legitimacy. A news organization's standing rests on the accuracy of what it publishes, and a translated version carries the same masthead as the original.
Most content categories can resolve this tension by taking more time. News cannot. The resolution has to come from process design — deciding in advance what checks are non-negotiable regardless of deadline, what can be deferred, and what the honest labelling is for content published at each level of verification.
Organizations that do not make these decisions in advance make them under deadline pressure, which reliably produces the wrong answer.
Quotes Are Not Ordinary Text
The single most important distinction in news translation is between reported content and quoted speech.
When a news organization places words inside quotation marks, or presents dubbed audio as a person's speech, it asserts that this is what the person said. A translation that renders a quote loosely has produced a misattribution — words the subject did not say, presented as theirs.
This has practical consequences for how quotes should be handled:
Translate quotes conservatively. The latitude appropriate for narration is not appropriate for quoted speech. Where a natural translation would smooth or clarify, the quote should preserve the source's actual construction, including hedging, qualification, and imprecision.
Preserve ambiguity. If a speaker said something that could be read two ways, the translation should not resolve it. Resolving ambiguity in a quote makes an editorial judgment and attributes it to the speaker.
Mark ellipsis and editing. Where a quote is shortened, the convention that marks the omission in text applies equally in translation.
Retain the original where it matters. For contested or highly consequential statements, showing the original alongside the translation lets the audience see what was actually said. This is standard practice in serious international reporting and is worth preserving in video through subtitles that show the original audio's words.
Get a second reading on consequential quotes. A quote that will be the story's headline deserves review by a second qualified speaker, regardless of deadline. This is the check that should not be deferred.
Attribution and Sourcing
Translation can obscure sourcing in ways that matter editorially.
The distinction between what a reporter observed, what a source stated, what a document showed, and what is being alleged is carried by specific verbs and constructions, and those constructions do not map uniformly across languages. A verb that implies confirmed fact in one language may have a hedged equivalent in another, or vice versa.
Attribution verbs deserve explicit attention in the terminology base: how to render said, claimed, alleged, confirmed, denied, and their gradations in each target language, with the distinctions preserved.
Anonymous sourcing conventions differ by journalistic culture. The phrasing used to describe an unnamed source carries information about their proximity and reliability, and that information should survive translation.
Legal exposure varies by jurisdiction. Defamation standards, privacy law, and reporting restrictions differ substantially between countries, and content that is publishable in the source market may not be in the target. A translated report distributed into another jurisdiction may be subject to that jurisdiction's law.
For sensitive reporting, legal review in the target jurisdiction is warranted, and this is a real constraint on how fast translated versions can be published.
Synthetic Voice Disclosure
Using generated voice in news content raises a disclosure question that other categories do not face as sharply.
When a translated report uses synthetic audio to render a person's words, the audience is hearing something the person did not say — not in the sense that the words are wrong, but in the sense that the voice is generated. Whether and how to disclose this is an editorial decision with real weight.
Prevailing practice in serious journalism leans toward disclosure. Common approaches include an on-screen note that the audio is a translation, retaining the original audio faintly beneath the translated track so the audience hears that translation is occurring, or using a clearly distinct narrator voice rather than one matched to the speaker.
The voiceover convention used across several media traditions — translated narration over audible original audio — has the incidental virtue of being self-disclosing. The audience hears both and understands immediately what they are receiving.
Cloning a subject's voice to make them appear to speak another language is the case requiring most care. For a news organization's own presenters, where consent is clear and the practice is disclosed, this is straightforward. For subjects of reporting, generating their voice saying words in a language they did not speak raises questions that most editorial standards would answer conservatively.
Set the policy at the organizational level rather than leaving it to individual producers, and state it publicly. Audiences respond better to a disclosed practice than to a discovered one.
Speed Without Abandoning Standards
A tiered approach lets speed and accuracy coexist by being explicit about which is being prioritized.
Tier one: immediate. Machine-generated subtitles published quickly, clearly labelled as automatically translated and unverified. This serves audiences who need the information now and are able to weigh it accordingly. The labelling is what makes it honest.
Tier two: verified. Human-reviewed translation with quotes checked, attribution verified, and terminology confirmed, published within hours. This becomes the version of record and replaces the tier-one version.
Tier three: full production. Dubbed audio, localized graphics, and full editorial treatment for stories that warrant it.
Critically, some checks should never be deferred to a later tier: verification of consequential quotes, correct attribution, and legal review where the content carries risk. If these cannot be completed, the story should not publish in that language rather than publishing unverified.
Deciding this list in advance, and giving editors the authority to hold a translated version, is what prevents deadline pressure from eroding the standard.
Terminology Under Deadline
News terminology cannot be built per story, because there is no time. It has to exist in advance.
A standing terminology base should cover: institution names in each language, official titles and how to render them, place names including contested ones, recurring political and legal vocabulary, and the attribution verb gradations discussed above.
Contested naming is the hardest category. Place names, group designations, and event descriptions carry political weight, and the choice between competing terms is an editorial position rather than a translation decision. These should be decided at the standards level, documented, and applied consistently rather than resolved by whoever is translating at the time.
Numbers, dates, and units need conversion rules established in advance, since news content is dense with figures and errors here are both easy and consequential.
The base should be maintained continuously rather than assembled reactively, with new terms added as stories develop rather than after they break.
Corrections Across Versions
When a story is corrected, every language version needs correcting, and this is where multi-language news operations most commonly fail.
The correction workflow should treat translated versions as part of the story rather than as derivatives. A correction issued on the source version that does not propagate leaves incorrect content published under the same masthead in another language.
This requires knowing what exists. A record linking each translated version to its source, with version identifiers, is what makes propagation possible. Without it, finding every affected version after a correction is manual and unreliable.
Correction notices should appear on translated versions in that language, following the same standards as the source. A correction notice in English on a Spanish-language version serves nobody.
Where a story is withdrawn, translated versions must be withdrawn simultaneously. Orphaned translations outliving a retracted source are a recurring failure and a serious one.
Timeliness matters as much as completeness. A correction that reaches translated versions days later has left incorrect content published throughout that period.
Live and Near-Live Content
Live news translation is a different technical problem with its own constraints.
Real-time translation operates without the review step, which means it is inherently tier-one content and should be labelled accordingly. Audiences generally understand that live translation is approximate, provided it is presented as such.
Latency is a genuine constraint. Any real-time system introduces delay, and in live content that delay accumulates against the source.
Accuracy degrades on exactly the material that matters most: rapid speech, overlapping speakers, unfamiliar proper nouns, and emotionally charged delivery. These are common in breaking news.
For consequential live content, a reasonable practice is to publish the live translation with clear labelling, then follow with a reviewed version once the event concludes, and treat the reviewed version as the record.
Archives and Long-Term Accuracy
News content persists, and translated archives carry the same accuracy obligation as source archives.
Content that was accurate when published may become misleading without context — a report on a developing situation reads differently once the situation resolved. Where source archives carry contextual notes, translated versions need them too.
Search and discovery of archived translated content requires localized metadata, and archives assembled without it become effectively inaccessible in the languages they were translated into.
Retention and takedown obligations may differ by jurisdiction, and translated content distributed into another market may be subject to that market's requirements around removal and correction.
Graphics, Chyrons, and On-Screen Text
News video carries a high density of on-screen text, and it is frequently the element left untranslated.
Lower thirds identifying speakers, their titles, and their affiliations carry information the audio may not repeat. A viewer who cannot read the name and role of the person speaking has lost context that the source audience received automatically.
Data graphics — charts, maps, comparison tables — often carry the substance of a report. Financial and statistical journalism in particular can be largely unintelligible without the graphics, and translating the narration alone delivers a fraction of the content.
Location and date stamps orient the viewer and should follow target-market date conventions rather than source ones.
Headlines and tickers, where present, are editorial content and need the same care as the script.
Maps deserve specific attention. Place names, borders, and territorial labelling carry political weight, and conventions differ by market. A map that is unremarkable in one country may be contentious in another, and this is an editorial standards question rather than a translation one.
Where full graphic localization is not achievable under deadline, subtitling the on-screen text is a workable fallback, applied consistently rather than selectively.
Building the Practice
Set the standards first: what checks are non-negotiable, what the synthetic voice disclosure policy is, and who has authority to hold a translated version.
Build the standing terminology base before it is needed, including the contested naming decisions, and maintain it continuously.
Establish the tiering explicitly, with honest labelling at each level, so that speed does not silently become a quality compromise.
Secure named reviewers per language with realistic availability, since a review requirement without a reviewer is a requirement that will be waived under pressure.
Build the correction propagation mechanism before the first correction is needed, including the version records that make it possible.
Train the desk. Editors making translation decisions under deadline need to know the standards without looking them up, which means the policy has to be taught rather than merely published. The failure mode is not usually disagreement with the standard; it is not knowing it existed at the moment the decision was made.
News is the category where the cost of a translation error is highest relative to the size of the error. A single mistranslated quote can become the story. Organizations that treat translated versions as fully part of their editorial output — subject to the same verification, correction, and standards as the source — produce work they can stand behind. Those that treat translation as a distribution step eventually discover the difference publicly.
