Translating an audio description track is a distinct task from translating ordinary movie dialogue. An audio description narrator explains important visual information during a video, often for audiences who are blind or have low vision. The description might say that someone opens a letter, that a warning appears on screen, or that a character silently leaves a room. A careless translation can add motives, emotions or visual details that neither the source description nor the image establishes.
If you are searching for how to translate audio description, how to localize described video, or how to translate visual narration for accessible films and educational videos, the practical rule is to preserve what can be observed and when it becomes useful. Compare the original description with the video where appropriate, keep description separate from dialogue, and check that the translated words fit the moments when viewers need them. Translation should increase access to information, not manufacture a new account of what the screen means.
Did you know that a video can have accurate subtitles and still be inaccessible to someone who cannot see its essential visual content? Subtitles primarily represent audio in text, whereas audio description communicates meaningful visual information through sound. They solve different access problems. A trustworthy multilingual workflow may therefore require both translated dialogue and a separately translated description track.
The quick answer: translate the described evidence, not an imagined story
Begin with the original audio description script, the corresponding video and, if available, the dialogue script with timecodes. Mark each description cue as a distinct narrative source. For every cue, identify what is directly depicted, what words are spoken, and whether the description supplies an interpretation. Then translate the cue in a natural target-language form while preserving what is observable, uncertain or intentionally undisclosed.
Test the translation in context. Description often occupies pauses between dialogue, so the new language must be intelligible without covering important speech or sound. Do not solve a timing problem by removing the one detail the audience needs. When the target language requires more time, revise the syntax carefully, negotiate space with the production team, or consider an authorised extended-description format.
The final check is not simply “Does the sentence sound fluent?” It is “Can a listener understand the significant visual event, distinguish it from dialogue, and follow the story without claims the image did not support?” That is the standard to work toward.
What audio description is and what it is not
Audio description is a spoken account of visual information needed to understand video content. It can describe actions, settings, scene changes, visible text, expressions and relationships when those details matter. It is not a replacement for the complete soundtrack, a constant commentary on every object, or a licence to explain everything the creator left ambiguous.
A subtitle track and an audio description track may both use language, but their starting evidence differs. Subtitles usually begin with speech and relevant sounds. Description begins with visually presented information that has no adequate spoken equivalent. A descriptive transcript may combine those streams in text, but good labelling preserves which information was heard and which was seen.
The W3C Web Accessibility Initiative explains that essential visual information can be integrated into a video’s main audio or provided through separate description. That distinction matters during translation: some videos have a separate narrator, while others were designed so presenters explain their own visuals.
Translating an existing track versus writing a new one
A translator of an existing audio description track has a source script and a target-language task. Their job is to carry across the source’s selected visual information and pragmatic effect, checking for accuracy and timing. A writer creating audio description from a previously undescribed video chooses what visual information to describe in the first place. The two jobs overlap, but they involve different editorial decisions.
If an existing source cue says, “She places the red folder under the desk,” a translator should not quietly rewrite it as “She hides a confidential file.” The second sentence introduces an interpretation of secrecy and contents. If the picture clearly contradicts “under the desk,” flag the source issue rather than concealing the change inside translation.
When you are commissioned to produce a new description from the video, follow the appropriate descriptive-writing brief, not merely the existing dialogue script. Establish who can approve additions, whether the new wording must match a reference version, and how corrections to source errors are documented.
Make a source matrix with three columns of meaning
Use a working matrix with visual evidence, original description and target-language description. Visual evidence records what appears on screen and at what moment. Original description records exactly what the established narrator says. The target column contains the translated version and brief decisions about grammar, terminology and timing.
This matrix reveals mismatches that ordinary sentence-by-sentence translation misses. The original narration might compress several movements into “He leaves,” while the video shows the character walking through a door after a long pause. A new language may need a subject explicitly named, but it should not include an imagined destination. The observed departure is the defensible core.
A fourth optional column can record uncertainty. “Handed to an unidentified person,” “text partially obscured,” or “emotion inferred by source script” are useful reviewer flags. The finished narration should not expose production shorthand; the notes keep editors honest about evidence and interpretation.
Decide which visual information matters to comprehension
Description must be selective because time and listener attention are limited. The aim is to convey visual information that contributes to understanding the action, relationships, place or narrative development. A background decoration may be unnecessary, while a character discreetly placing a key in a drawer may be central to what happens later.
The translator should preserve that priority. A short sentence can be accurate but unhelpful if it removes the only plot-relevant detail. Conversely, a beautiful target-language passage can overload the soundtrack with descriptive flourishes that prevent the listener from hearing the actual dialogue. Accuracy includes making information available at a useful moment.
Ask what a listener would miss without the cue. If the answer is “the identity of the person entering,” the name or distinguishing feature may be important. If the answer is merely “the colour of a distant wall,” the detail may be optional unless colour is itself relevant to the story or lesson.
Objective observation versus interpretation
The difference between “a person frowns” and “a person realises they have been betrayed” is substantial. The first describes a visible expression; the second reports an internal conclusion. The conclusion may be plausible, but plausibility is not enough to present it as observed fact. Translation must not silently move from a visible action to a psychological diagnosis.
Similarly, “she grips the railing” is not automatically “she panics.” “He looks at the door” is not “he plans to escape.” “They exchange glances” does not necessarily mean they agree. A skilled description can make emotional information accessible by describing expressive cues without making speculative claims about their cause.
Some source descriptions themselves contain conventional interpretive language. When translating them, assess the commissioning style guide and the audiovisual evidence rather than substituting personal interpretations. If an important source claim exceeds what can be supported, raise an editorial query before final delivery.
Keep the description voice separate from the people in the scene
An audio description narrator usually has a different communicative role from the characters. Their voice may describe what someone does while a character is silent, and it may pause for character dialogue. If the target script does not distinguish these roles, listeners can mistakenly attribute the narrator’s words to a character.
Use script labels during production: “AD narrator,” “Character A,” “Character B,” “voice-over in scene,” and “source soundtrack.” Do not assume every voice-over is audio description. Documentary narration may be part of the programme itself; an accessibility description track supplies additional visual information. In some educational videos, description is integrated into the speaker’s own words.
When multiple voice tracks exist, verify which one the audience will hear in the target-language version. A voice actor assigned to the description should not accidentally read character dialogue from the same combined transcript. Source separation is a foundational translation skill, not merely an audio mixing preference.
Work with silence, pauses and the original soundtrack
An audio description cue is meaningful partly because of its timing. A brief description during a pause allows the audience to follow a movement without missing dialogue. If the translated sentence becomes longer, inserting it over a crucial exchange can make the soundtrack less accessible than before. Cutting words indiscriminately can cause the opposite problem by removing the action’s essential information.
Measure the available pause in the actual edit and read the target line aloud at a comfortable pace. Remember that different languages have different syntax and lexical length; a word-for-word character count is a poor predictor of listening time. Allow room for comprehension, not just for the last syllable to finish.
When the pause is genuinely too short, consider a tighter but equivalent sentence, an approved change in cue placement, or an extended-description version if the production supports one. W3C’s planning guidance for accessible media describes how separate or integrated visual descriptions fit alongside the rest of a programme.
Translation length: compression without distortion
Suppose the source says, “On the desk, beside the unopened envelope, a small metal key catches the light.” A concise translation may preserve “a small metal key lies beside an unopened envelope” if the lighting detail is not needed and the brief permits compression. But changing it into “the key to the safe lies on the desk” introduces an unidentified purpose.
When shortening a sentence, preserve actor, action, object and any distinction necessary to follow later events. A sentence can lose an ornamental adjective without losing the plot, but losing “unopened” may matter if the next scene turns on whether anyone has read the letter. The right compression is semantic, not mechanical.
Read the compressed sentence next to the image and the next two cues. A choice that looks safe locally may remove a detail that becomes important in the following sequence. Description is a connected narrative, not a list of independent visual labels.
Translate spatial relationships precisely
Audio description frequently uses prepositions, perspective and movement. “Behind the chair,” “to her left,” “toward the door” and “across the table” place objects in relation to other objects or viewers. Different languages structure these relationships differently, and some prefer relative directions while others rely on contextual reference.
Check whether “left” means screen left, character left or the position of an object relative to another person. A literal translation that swaps frames of reference can mislead a listener. If the original is ambiguous, the video may help resolve it; if the video does not, do not invent certainty.
Spatial distinctions become especially important in educational and practical demonstrations. “The red button sits below the screen” differs from “the button on the lower left.” When somebody must identify an object safely, the translation should retain the clearest grounded relation rather than decorate the sentence with extra directional words.
Names, pronouns and late revelations
Many stories delay identifying a person or object. A camera may reveal only a silhouette before showing a familiar character. An audio description should respect deliberate withholding when possible, using an appropriate descriptor until the identity becomes clear. Translating “a figure” as a named character because the translator knows the ending may spoil the reveal.
Pronouns also require care. A target language may force a grammatical category that the source leaves open. Investigate whether an accurate neutral construction is possible, and consult the editorial brief when it is not. Do not assign age, gender, ethnicity or identity from appearance alone, especially where the source intentionally avoids doing so.
Keep a character and term sheet across episodes or modules. The same person should not be described by contradictory names or titles merely because different translators worked on different sections. Consistency makes the audio world easier to follow.
Colours, clothing and other descriptive detail
A source audio description may use colour to distinguish objects or people. If the colour later helps identify a character, preserve it accurately. If the source uses a culture-specific colour term, choose an ordinary target-language equivalent that matches the visible hue or the production glossary. Avoid imposing symbolic meanings such as luck, danger or mourning unless the work explicitly establishes them.
Clothing can also indicate action, role or continuity, but clothing is not proof of an occupation or identity. “A person in a white coat” is evidence of clothing; “a doctor” may be a role requiring independent verification. Similarly, a formal suit does not establish that somebody is a lawyer or executive.
Description should give listeners usable distinctions. Short recurring identifiers often work better than elaborate repeated portraits. Translators can help make those identifiers consistent across scenes without rewriting the film’s intentions.
Scene changes and visual transitions
A video can move from a classroom to an outdoor setting, from present action to a flashback, or from a demonstration to a chart. If the original description marks that shift, the translated version must help the listener understand it at the correct moment. A transition phrase can be essential even when it contains little dramatic vocabulary.
However, avoid naming a time shift without evidence. A dimmer colour palette does not by itself prove that a scene is a memory. An establishing shot does not always indicate a new city. If the source describes “another room” rather than a particular place, preserve the observable relation unless the programme confirms the location elsewhere.
When timing is tight, a concise transition can carry a great deal of meaning: “Later, in the workshop” is useful only if “later” is established. A more cautious “In the workshop” may be the correct translation when temporal order is unknown.
On-screen text and captions in the image
Video text can carry names, instructions, figures, messages or warnings that are not spoken. An audio description may read some of that information aloud, paraphrase it, or identify that text appears. The translator should check whether the same wording is already present in the dialogue or another audio track, so that listeners are not overwhelmed by unnecessary duplication.
Critical text such as legal disclaimers, safety warnings and numerical values may need exact treatment according to the brief. Decorative posters or incidental labels may not. A subtitle already visible to sighted viewers does not automatically reach someone relying on an audio-only presentation.
For the neighbouring translation problem of text appearing on screen, read the eduKateSG guide to on-screen text, lower thirds and graphics. In audio description, the key issue is deciding what visual text must become audible and how to fit it into the soundscape.
Charts, diagrams and educational explanations
An educational video may show a graph while the narrator says, “The line moves upward.” That may be insufficient for a listener who needs the axes, variables and numerical comparison to understand the lesson. When translating an existing description, check whether those essential details are included. If not, flag the access gap rather than pretending that a new adjective has solved it.
For example, “the chart rises” does not tell us whether a price, population, temperature or probability is increasing. An accurate description should identify the variable when known. “The rate increases from two to five units per minute” conveys more actionable content than “the line climbs sharply,” although the available data must support those figures.
The educational purpose is to help a learner reconstruct relationships. The eduKateSG mathematics learning guide demonstrates the broader idea that understanding a representation requires more than naming what it looks like. Translated audio descriptions should likewise help people understand the structure of information.
Numbers, quantities and labels must remain attached to their referents
If a source track says, “Three boxes remain on the shelf,” do not translate it as “a few boxes remain.” The count may be significant. If the screen shows two boxes while the script says three, document the discrepancy rather than choosing whichever version happens to seem more plausible.
Descriptions often use numbers with measurements, directions, prices or counts. Verify the unit and context, especially when the target language changes plural forms or number agreement. Do not convert measurements casually unless the brief requests a conversion and the converted value is accurate, appropriately rounded and clearly presented.
A common editing error is preserving a figure but attaching it to the wrong object. For more systematic treatment, see translating spoken numbers, measurements and quantities. The same discipline applies when those figures originate in visual description.
Cultural adaptation without rewriting the scene
A source description might identify an object or gesture that is unfamiliar to a new audience. Translation can sometimes use a concise functional description, but should not replace the depicted object with an unrelated local equivalent. A viewer needs to imagine what the camera actually shows.
Suppose the video displays a particular style of paper lantern. Calling it simply a “lamp” may remove culturally significant detail, while inserting a long history lesson may break timing and change the tone. A good target-language phrase balances recognisability and fidelity. Additional cultural context can go in a companion note or descriptive transcript when needed.
Gestures are especially sensitive. A raised hand can signal greeting, interruption, agreement or simply movement depending on the scene. Translate the visible action first, and avoid asserting its social meaning unless the narrative supplies evidence.
The ethical boundary: do not diagnose or stereotype from images
Audio description should not speculate about a person’s health, mental state, identity or social position from appearance alone. Translators inherit the responsibility to avoid strengthening unsupported labels. A person leaning on a cane is not automatically unwell at that moment; somebody who avoids eye contact is not automatically dishonest.
When a source script uses interpretive terms that could stigmatise or misidentify a person, follow an appropriate editorial review process. Translators may raise questions, propose evidence-based wording and document decisions without unilaterally transforming the work into their own commentary. The aim is responsible access to the original, not an invented moral account of the scene.
Accuracy also requires respect for tone. A comedy, documentary, drama and safety-training video may call for different levels of detail. Yet genre does not excuse inventing a person’s inner thoughts or altering observable actions.
Worked case: the envelope that is not yet opened
In a fictional drama, the original description reads, “A sealed envelope rests beside the empty cup.” The next scene shows the character placing it in a drawer without opening it. A translator writes, “She hides the secret message.” That sounds dramatic, but the contents of the envelope are not yet known, and the act of placing it in a drawer may not amount to concealment.
A faithful description preserves the visible facts: the envelope is sealed, positioned near the cup, and later moved to the drawer. If the target language allows a simple phrase for “sealed,” choose it even if the word is less poetic. The envelope’s unopened status may later be crucial evidence in the plot.
The learner’s diagnostic question is, “Which part of my sentence can the viewer actually see or the source script explicitly establish?” Any unsupported phrase should be removed, clarified as speculation or routed to editorial review.
Worked case: a gesture without a confirmed motive
In a fictional interview, a guest briefly looks away and folds their arms. The source description says, “The guest looks down and crosses their arms.” The translated version says, “The guest becomes defensive.” That is an interpretation, not a visual event. The guest might be thinking, uncomfortable, cold or simply changing posture.
A more accurate translation names the movements in natural language. If the programme’s context explicitly establishes an emotion, the script may be able to communicate it, but the interpreter should know which evidence supports that choice. A gesture does not carry the same meaning in every culture or circumstance.
This case makes an important cross-linguistic point: a target-language adjective that seems like a tidy substitute for two verbs may change the epistemic status of the sentence. Fluency does not justify unwarranted certainty.
Worked case: a technical diagram with too little description
An instructional video shows a process flow with three boxes connected by arrows. The description says, “A diagram appears.” The viewer listening without the screen cannot know the sequence being taught. Translating this sentence perfectly does not fix the missing accessible information.
The translation team should flag that the source description is insufficient and ask whether the brief permits supplementary description. If it does, the new wording must be grounded in the visible diagram: “Three boxes, labelled Input, Check and Output, connect from left to right,” provided those labels and direction actually appear. The names must not be guessed from the lecture topic.
For education, access includes relationships, not merely object categories. A learner needs to know which component feeds the next and what each label means. Description is successful when the process becomes understandable.
Worked case: a silent location change
A character closes a door; the next shot shows the same character standing outdoors in rain. The original audio description says, “Outside, rain falls around her.” The translator adds “The next morning,” imagining that a night has passed. Yet the editing may depict an immediate move to the building exterior.
Translate the location and weather, but do not add a temporal marker without confirmation. If the film establishes morning through another reliable cue, that can inform the description; if not, uncertainty should remain. The user of the description deserves the same narrative suspense as the viewer of the image.
A seemingly helpful time word can alter cause and effect. Translators should be particularly cautious with “later,” “meanwhile,” “already” and “again,” because these small words shape the timeline.
Worked case: audio description over music
A documentary montage shows workers arranging equipment while instrumental music plays. The audio description must convey an important visual step in a brief opening between musical phrases. In the target language, the initial draft is twice as long and covers the following interview. Simply speaking faster would make the accessible track harder to process.
Identify the essential action and remove only nonessential descriptive weight. A compact sentence such as “Workers fasten the protective cover” may preserve the instructional point if those details are on screen. If the original cue includes a crucial warning symbol or number, it should not be sacrificed merely to match the music’s rhythm.
Test the revised line with the actual sound mix. The translator needs to hear whether the words remain clear, whether any dialogue is masked, and whether the visual action is still understandable at the moment it occurs.
Worked case: a character’s identity is intentionally concealed
At the beginning of a mystery, a hooded figure enters a room. The source description says, “A hooded figure places a small box on the table.” The translator, who has seen the full episode, names the figure because they know the reveal. That destroys information the work deliberately withholds.
The correct translation keeps a nonidentifying descriptor until the scene provides a reliable identity. If the target language requires a gendered form, search for a neutral alternative where natural. Do not claim identity from the shape of the figure or the sound of footsteps alone.
Description gives access to the film’s intended information, including its uncertainty. Withholding an unconfirmed name is not a failure to be helpful; it is part of preserving the story.
Worked case: a displayed warning that is not spoken
A laboratory demonstration includes an on-screen safety warning while the presenter continues explaining a procedure. The existing dialogue translation is accurate, but it never communicates the warning to someone who cannot see the screen. The audio description source says only, “Text appears in the corner.”
If the warning contains safety-relevant instructions, the omission should be flagged. The team may need a revised description, integrated narration or another accessible text alternative. Any resulting translation must carry the actual warning accurately, including negation, sequence and units, without inventing new safety guidance.
The distinction is crucial: translating a weak source description is not the same as providing an adequate accessible experience. Responsible localisation identifies the gap rather than hiding it.
Worked case: a chart with an uncertain scale
A financial explainer displays a line graph with a steep visual slope but an unlabeled vertical axis. The description says, “The line rises steeply.” A translator upgrades this to “Revenue doubles.” That is unsupported; an angled line without a confirmed scale cannot establish a doubling.
A faithful translation describes the visible trend, while a reviewer can request additional chart data if the learning goal requires precise quantities. If the presenter supplies numbers aloud, those numbers must be handled as spoken evidence and linked to the graphic only when the relationship is explicit.
This example demonstrates why description and inference must remain separate. The listener needs access to the same evidence, not stronger claims than the image can sustain.
Test a translated track with its intended users
A language expert should verify the target wording, but a final accessibility review also benefits from people who use audio description. Ask whether participants can follow who moves, what changes, which objects matter and whether the timing permits attention to the dialogue. Avoid leading questions that reveal the story before the listening test.
A useful evaluation includes both factual comprehension and listening comfort. Did the translation omit an important movement? Did the audio crowd the character’s words? Was a new name introduced before the film revealed it? Was a technical relationship left unexplained? These are different error categories with different remedies.
When user testing is not available, perform a structured audio-only review and document its limits. A sighted editor can listen without watching, but that exercise does not substitute for the full range of perspectives from people who regularly rely on accessible media.
A compact quality-control matrix
Review each cue for six dimensions: visual grounding, narrative relevance, source fidelity, speaker/source separation, language clarity and synchronisation. Mark a cue for correction whenever one fails. It is possible for the translation to be fluent and grammatically perfect yet weak in two or three of the other dimensions.
For visual grounding, ask which depicted details support each noun, verb and descriptive modifier. For narrative relevance, ask whether the cue gives the audience information they need. For fidelity, compare it with the authorised source description. For timing, listen to the actual mix. For source separation, confirm that the description does not look like character speech in transcripts or subtitle exports.
Keep reviewer notes specific. “01:05: ‘furious’ is unsupported by the visual source; replace with the observable action” is more actionable than “tone feels wrong.” The goal is repeatable correction, not a vague impression of professionalism.
A practical learning route for translators
Start with a silent thirty-second sequence containing two or three important visual actions. Write a factual description using only what can be seen. Separately write an interpretive summary, then compare the two. Notice how often the summary adds purpose, emotional judgment or imagined background. Translating the factual description first trains restraint.
Next select a short existing audio description track that you have permission to use. Translate three cues, speak them aloud against the video and mark where the timing fails. Revise by removing unnecessary structure rather than deleting key visual facts. Keep a terminology sheet for recurring objects and characters.
Finally, let another person listen to the translated version without seeing the image. Ask them to describe the action and sequence. Compare their account with the actual video. If they confidently report something unsupported, examine whether the translation created that certainty.
Frequently asked questions
How do I translate an audio description track into another language?
Use the source description script, the corresponding video and a time-coded working file. Translate each cue with attention to visual evidence, natural language and available pauses. Keep the description narrator distinct from dialogue and perform an audio-only comprehension check on the finished result.
Is translating audio description the same as translating subtitles?
No. Subtitles primarily render speech and relevant sounds in text. Audio description communicates important visual content through speech. A multilingual video may need both, and a descriptive transcript can combine them in written form when the sources are clearly identified.
Can I add visual details that the original description leaves out?
Not silently as though they were already in the source. If the task includes creating or improving description, additional observed details may be appropriate within an authorised editorial process. If the task is strict translation of an existing track, flag significant omissions and obtain guidance before changing the informational scope.
Should audio description explain how a character feels?
Only to the degree justified by the source and the applicable description policy. Observable movements and expressions are safer than invented psychological motives. “She covers her face” and “she is ashamed” are different claims and should not be treated as interchangeable.
What if the translated cue no longer fits between lines of dialogue?
Rewrite the target sentence more concisely while preserving essential visual information, review cue placement with the production team, or use an approved extended-description method where necessary. Speaking unnaturally fast or covering important dialogue is not a satisfactory solution.
How should I translate names before a character is revealed?
Preserve deliberate uncertainty. Use an appropriate description such as “a figure” when the source withholds identity, and do not insert a name merely because you know the later reveal. Maintain a consistent character glossary once identity is confirmed.
Can automated translation tools handle audio description?
They may help create a draft, but they cannot reliably verify whether added details are visible, whether a gesture has the claimed meaning or whether the spoken cue fits between soundtrack events. Review against the video and the target audience’s accessibility needs is essential.
Do diagrams and charts need more than a general audio label?
Often they do. If a visual relationship or number is necessary for understanding, “a chart appears” is insufficient. The team should provide a grounded description of axes, values, labels or process steps as appropriate, without inventing unseen data.
Does the same method work for documentaries and lessons?
The foundational method does, but information priorities vary. A drama may depend on character actions and suspense; a scientific lesson may depend on graph relationships and measurable details. Translate according to the source evidence and the audience’s purpose rather than imposing one fixed style.
Where can I study the adjacent skills?
Read the guide to environmental sound cues in accessible captions for the reverse accessibility direction, and consult eduKateSG’s vocabulary routes for precise spatial terms, connotation and word meanings. The W3C accessible media planning guide is a useful standards-oriented companion.
The principle to keep
Audio description translates visual access into spoken information. Translating that description into another language adds a further layer of responsibility: every sentence must still belong to the real image, the real sequence and the time in which it can be understood. Clear wording is valuable, but evidence-based description is what keeps the audience informed rather than misled.