Multimodal Translation: Speech, Subtitles and Visual Meaning

How to translate video and multimedia without losing the connections between words, pictures and time

How do you translate a video with subtitles, spoken dialogue, text on screen, gestures and sound effects without making the different channels contradict one another? The reliable method is to translate the meaning of each scene as a coordinated system, not as independent text files. A spoken warning can be correct in translation while arriving after the relevant action, a subtitle can be accurate but attached to the wrong speaker, and an on-screen button can be translated differently from the voice that tells a viewer to press it. This complete guide to multimodal translation explains how to preserve those relationships.

The hidden problem in audiovisual translation is alignment. Human audiences combine sound, speech, images, camera viewpoint, written labels, timing and social cues as they interpret an event. Removing a pointing gesture from its spoken reference, changing the order of two captions, or translating a visual instruction without checking the footage can create a new message that no single component intended. Translators must therefore identify which channel supplies each part of the meaning, who or what it refers to, and when the reader or viewer receives it.

If you are searching for how to translate subtitles and on-screen text together, how to localize voice-over with captions, how to synchronize translated dialogue with visual actions, or how to quality-check multilingual video for accessibility, start with a shared scene ledger. This article gives a repeatable translation workflow, precise failure diagnostics, cross-industry worked examples and transfer tests for captions, transcripts, audio description, graphics, product demonstrations, meetings and recorded lessons. It is world-facing: the method depends on the medium and audience, not a particular location.

The central rule is to ask which relationship a translation must preserve. A hand points to a left-hand switch while the voice says ‘this one’; the phrase depends on a visual referent. A camera cuts to a second speaker as subtitles continue from the first; attribution depends on time. A siren sounds before a visible warning sign appears; the sequence creates an instruction and an expectation. In each situation, translating the spoken string alone cannot guarantee communicative equivalence. The unit of review must be the meaning-bearing scene and its linked assets.

This chapter is not a substitute for the existing specialist eduKate guides on timed-text markup, subtitle format validation, accessibility or speech synthesis. Those resources explain the rules and technologies of particular channels. Our narrower editorial ownership here is the cross-channel relationship: what happens when several accurate assets become misleading when assembled together. Every example is an invented teaching demonstration, not a report about a real media production.

Quick method: inspect the scene, build the alignment map, then translate

Before opening a translation memory, obtain a stable source-media cut with frame or time references. List the assets that a viewer perceives: spoken dialogue, sound effects, visible text, important images, gestures, captions, charts, speaker identifiers and any optional description track. Identify the event or claim conveyed by their combination. Record dependencies such as ‘the word this points to the red lever at this moment’ or ‘the warning must be available before the action shown at the next cut.’ Translate the speech and screen text consistently, preserve cue-to-event timing, and validate the assembled target by watching and listening rather than proofreading files in isolation.

Use at least two review modes. First, read the translated captions without sound to discover missing speakers, unreadable speed and ambiguous reference. Second, watch the target with the soundtrack and picture to catch timing clashes, wrong gestures, visual contradiction and narration that arrives too late. Where accessibility is part of the brief, independently assess caption coverage of meaningful non-speech audio, audio description of important visual information and the descriptive transcript. None of these checks substitutes for the others.

What authoritative specifications do and do not establish

The W3C Web Accessibility Initiative distinguishes same-language captions from translated subtitles for explanation, while acknowledging that common terminology varies internationally. Captions cover relevant speech and meaningful nonspeech audio, typically in synchrony with video. Translated subtitles frequently focus on dialogue and may not provide everything a deaf or hard-of-hearing user needs. The words captions and subtitles cannot safely be treated as interchangeable requirements without a defined delivery brief and accessibility test.

WCAG 2.2 covers prerequisites for accessible prerecorded and live synchronized media through several separate success criteria, including captions, audio description and media alternatives. A translation project should distinguish legal or contractual conformance claims from its editorial goal of making the assembled scene understandable. Captions alone do not describe every visual fact, and an audio description cannot automatically replace a full set of captions. The appropriate deliverables depend on the medium and the requirements applicable to the product.

WebVTT describes time-aligned text cues and cue settings. W3C’s IMSC Text Profile 1.3, published as a Recommendation on 21 May 2026, is a modern text-only TTML2 profile intended for global subtitle and caption delivery. It adds support for superscript and subscript and introduces authoring guidance for Japanese-language content compared with the previous text profile. A technically valid timed-text file can nevertheless be semantically wrong if its cues identify the wrong speaker or obscure the very action the words refer to.

Provider-specific requirements deserve explicit versioning. Netflix’s English (USA) timed-text style guide, for example, states a 42-character line limit and adult English reading-speed limit of up to 20 characters per second, with different expectations for children. These figures are not universal laws for every language, audience, device or client. Production should use the actual current delivery specification for each target track, while keeping underlying meaning, audience comprehension and accessibility as separate review goals.

The scene ledger: one representation for many languages and output formats

An effective alignment ledger can be a table or structured data record. Each row represents a meaning-bearing beat, not necessarily one sentence or one subtitle cue. Store a stable beat identifier, source video version, start and end time, speaker or actor, source speech, scene action, on-screen text, important sound, communicative intention, required audience inference, translation variants, timing window, visual exclusion zones, accessibility notes and quality-assurance status. Records may reference several media assets at once.

A ledger is useful only if its dependencies are explicit. For instance, a beat may have ‘gesture points toward the upper green control’ and ‘spoken pronoun refers to that control.’ The translator can then make a language-appropriate decision: replace the pronoun with the control name, maintain a natural deictic expression if the image is clear, or query the source owner if the footage is ambiguous. The decision should be tested with the final picture, because the correct answer depends on what the viewer actually sees.

The ledger must also record the media’s intended mode. A silent autoplay clip with captions, a narrated tutorial, an audio-described programme and a live video meeting are not equivalent delivery conditions. A caption that works with sound on may be insufficient when the main evidence is a gesture. A transcript that works as a linear article may omit meaningful timing. Distinguish invariant semantic relationships from channel-specific representations rather than forcing every output to contain exactly the same words.

1. Cross-channel referent

A pronoun or pointing expression refers to an object visible only during a specific shot. Consider this fictional scene: A narrator says ‘move that there’ while a hand places a cartridge into an upper slot. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

The source analysis should name the spoken content, visible referent, relevant moment, and expected viewer inference. If those do not agree, it is a source-authoring problem rather than something to conceal with an elegant translation. A useful distinction is whether the words merely accompany the scene or depend on it. Dependent meaning requires a link to a frame, speaker, graphic or action in the alignment ledger.

Repair method: Translate the referent in context; if the target needs a noun phrase, name the correct object rather than guessing. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Mute the video and let a target-language reviewer identify the object and destination from captions plus images. Acceptance condition: The viewer must choose the same item and destination as the source viewer. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

2. Visible label and spoken name

A voice names a control that has an independently localised interface label. Consider this fictional scene: The audio says ‘press Continue’ but the displayed button in the target says ‘Next’. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Agree a single concept and verify whether the screen text is being replaced, mirrored or left in source language. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Play the final clip and check that a user can locate the control without translating back mentally. Acceptance condition: The user acts on the intended screen element. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

3. Caption to speech onset

The time a subtitle appears influences which person and utterance it represents. Consider this fictional scene: A caption appears while the next speaker is already visible. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Retarget the cue to the audible line and preserve shot-change context, rather than simply shifting all captions by one offset. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Switch between sound-only and mute playback to test speaker identity. Acceptance condition: The attributed statement belongs to the person who actually said it. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

4. Gesture-speech polarity

A nod, head shake or hand movement can affect the interpretation of spoken agreement and refusal. Consider this fictional scene: A speaker says ‘sure’ sarcastically while shaking their head. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Preserve pragmatic intent without asserting that gesture alone universally encodes the same convention in every culture. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Ask a viewer to interpret the full scene and explain the evidence for that interpretation. Acceptance condition: The translation does not turn a refusal into sincere consent. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

5. Sound-before-visual warning

An alert sound precedes a sign or spoken instruction and creates urgency. Consider this fictional scene: A beep sounds two seconds before the display shows ‘Stop’. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Make meaningful sound available through captions where required; keep the instruction in the right sequence. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Test what a viewer understands at the moment action is required. Acceptance condition: The warning arrives before—not after—the depicted hazard. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

6. Off-screen speaker

Voice identification depends on whether the speaking person appears in the frame. Consider this fictional scene: Two people talk while the camera shows only a product close-up. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Use a consistent voice or speaker label where the target medium needs it. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Watch without sound and ask which sentence belongs to which person. Acceptance condition: No opinion or action is falsely attributed to the visible person. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

7. Overlap and interruption

Two voices may speak simultaneously, with interruption carrying a social or factual consequence. Consider this fictional scene: One speaker starts to say ‘not yet’ while another announces completion. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Prioritise the meaning of both voices according to the brief, marking interruption rather than synthesising a single consensus. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Compare annotated source audio with the target sound mix and captions. Acceptance condition: The viewer understands the interrupted claim and its status. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

8. Slide-to-narration link

A presenter refers to a chart label or colour while speech continues. Consider this fictional scene: The voice says ‘the dashed line’ as the slide shows a blue dashed series. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Identify the actual graph item and ensure translated references match visible labels and legend. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Pause at the referenced frame and read the target speech beside the target slide. Acceptance condition: The same data series is selected. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

9. Diagram sequence

Spoken instructions correspond to stages of an animated process. Consider this fictional scene: The narration says ‘first’ just as the second valve begins moving. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Bind translated text to scene beats rather than relying on sentence counts. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Trace the depicted action order with narration hidden and then restored. Acceptance condition: The reader reconstructs the correct procedure. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

10. Camera cut and location

A camera transition can relocate a deictic phrase such as ‘here’ or ‘behind us’. Consider this fictional scene: A guide says ‘here’ as footage cuts from an entrance to a storage room. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Resolve the intended location before translating and retime captions when necessary. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Ask viewers where the instruction applies using the actual target edit. Acceptance condition: The translated phrase names the correct location. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

11. On-screen number versus speech

A voice states a quantity that appears as a figure in a visual. Consider this fictional scene: The audio says ‘about thirty’ while a counter reads 28.7. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Preserve any legitimate approximation and query unexplained contradictions; do not edit one channel independently. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Compare spoken, captioned and graphic quantities in the published version. Acceptance condition: No artificial precision or numerical conflict is introduced. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

12. Unit displayed in graphic

A unit in a visual constrains a verbally translated value. Consider this fictional scene: A chart labels temperature in Fahrenheit but a voice-over gives Celsius numbers. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Record conversion decisions for both channels and preserve scale, rounding and units. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Check the axis, subtitles and narrated figures simultaneously. Acceptance condition: The audience interprets the same magnitude. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

13. Text appearing before action

Instruction text may be present before, during or after the relevant action. Consider this fictional scene: A safety overlay remains visible while a machine starts. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Translate the text within the visibility period needed for decision-making. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Review at playback speed and identify the first frame the instruction is actionable. Acceptance condition: The viewer has enough information before acting. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

14. Visual contradiction

The footage itself may conflict with narration or visible labels. Consider this fictional scene: A voice says ‘the lower latch’ while an arrow highlights the upper one. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Raise a source-content query rather than silently choosing a side. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Document source contradiction and test the approved resolution in the final cut. Acceptance condition: The translator does not disguise a source error as a translation choice. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

15. Speaker identity across cuts

A voice can continue over shots of another person or object. Consider this fictional scene: An interviewer asks a question while the camera shows the interviewee preparing to answer. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Track speaker turns in audio, not only faces visible on screen. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Cross-reference each target caption with actual voice onset and speaker. Acceptance condition: The reader never believes a question was answered by the wrong party. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

16. Reused footage and changed script

A production may update narration without changing all the visual assets. Consider this fictional scene: An updated product clip has a new menu label but old screen captures. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Maintain version identifiers and resolve content mismatches before translation release. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Compare approved script revision with footage build and all target overlays. Acceptance condition: The translated release reflects a consistent source version. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

17. Meaningful silence

A pause can signal hesitation, suspense, refusal or the time for an action. Consider this fictional scene: The speaker stops after ‘we could…’ while looking at a closed door. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Do not fill silence with invented explicit content; preserve the intended pause when practical. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Watch for a target voice-over or caption that closes the hesitation prematurely. Acceptance condition: The same uncertainty or dramatic beat remains available. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

18. Laughter and audience reaction

Laughter or gasps can change how a sentence is understood. Consider this fictional scene: A speaker says ‘excellent’ while others laugh after a visible mistake. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Decide whether captions require nonspeech information and keep tone readable without editorialising. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Review as a viewer who cannot hear the reaction and another who cannot see facial cues. Acceptance condition: The audience can distinguish praise from irony when the original supports it. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

19. Music as information

Music may signal context or mood rather than merely decorate a scene. Consider this fictional scene: A short motif returns when a character recognises an object. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: In accessibility tracks, identify meaningful sound without imposing unsupported narrative claims. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Compare the purpose of sound labels against the story and captions. Acceptance condition: A key storytelling cue is not silently lost. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

20. Audio-description collision

A spoken description of visual content can compete with dialogue for the same time slot. Consider this fictional scene: A description of a gesture would overlap a safety instruction. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Allocate description intelligently and consider an alternate extended version if necessary. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Listen to the full target mix with dialogue, description and effects. Acceptance condition: Important information remains perceptible rather than buried. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

21. Transcript versus timeline

A linear transcript can omit the temporal dependencies that carry meaning in video. Consider this fictional scene: The phrase ‘before this’ occurs while an arrow appears for two seconds. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Add scene descriptions, timestamps or linked references as appropriate. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Have a transcript-only reader reconstruct the intended order and action. Acceptance condition: The transcript retains necessary temporal meaning. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

22. Captions versus subtitles

Same-language accessibility captions and interlingual dialogue subtitles can serve different readers. Consider this fictional scene: The translation retains dialogue but omits a crash that changes the plot. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Deliver the correct track type; include meaningful non-speech cues in caption deliverables. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Review tracks separately according to their stated use cases. Acceptance condition: A user choosing a caption track receives necessary audio information. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

23. Forced narrative

A brief foreign-language sign or dialogue fragment may require special treatment within an otherwise localised video. Consider this fictional scene: A warning sign is visible while an English voice-over continues. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Decide if a forced subtitle or translated graphic is needed without duplicating or obscuring other text. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Watch in each track combination, including when general subtitles are off. Acceptance condition: The user does not miss essential non-primary-language information. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

24. Line-break syntax

Subtitle breaks can change the apparent subject, negation or scope of a line. Consider this fictional scene: A line ends with ‘do not’ and the next caption begins with ‘press’. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Group syntactic units for comprehension while meeting the project reading-speed and line-limit rules. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Read captions at real playback speed without pausing. Acceptance condition: The audience receives the intended instruction without a temporary opposite interpretation. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

25. Reading-speed pressure

Target language expansion can make complete written content unreadable within a fixed display window. Consider this fictional scene: A four-second demonstration contains a long compound instruction. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Prioritise meaning-preserving condensation, adjust timing where permitted or modify the edit rather than dropping the critical condition. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Use a human reading test and project-specific numerical thresholds. Acceptance condition: The viewer has time to understand the action and its safety limit. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

26. Caption occlusion

A subtitle may cover the object, control, face or evidence to which it refers. Consider this fictional scene: The bottom-positioned caption obscures a small warning label. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Use appropriate cue region or edit layout while preserving safe-area and style requirements. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Inspect the smallest supported device view as well as desktop. Acceptance condition: The subtitle and the relevant object are simultaneously perceptible. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

27. Safe-area variation

Different frame crops can remove edge text, hands or speaker labels. Consider this fictional scene: A square social crop removes a pointing finger that gives ‘this’ its meaning. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Audit each aspect ratio as its own scene composition. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Test target captions and visual references across declared delivery formats. Acceptance condition: No crop deletes a referent required to interpret words. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

28. Right-to-left mixed content

Arabic or Hebrew scripts mixed with dates, codes and left-to-right UI labels may display confusingly. Consider this fictional scene: An RTL caption places a model number beside a directional arrow. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Use language- and direction-appropriate rendering controls and test the player, not just stored text. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Verify numerals, punctuation, speaker tags and screen labels in actual playback. Acceptance condition: Visual order matches the intended reading and action. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

29. Ruby and vertical writing

Some languages need language-appropriate layout beyond plain horizontal glyph replacement. Consider this fictional scene: A term carries pronunciation guidance or appears in a vertical panel. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Select supported presentation technology and validate typography without changing the relation to the scene. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Inspect actual fonts and line layout on supported devices. Acceptance condition: The same term and referent remain readable. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

30. Cultural interpretation of signs

A gesture or visual symbol may not carry a universal social meaning. Consider this fictional scene: An actor uses a hand gesture conventional in the source production. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Research the intended gesture in context; use appropriate verbal or descriptive support instead of stereotype. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Have target-audience reviewers explain what they infer and compare with source intent. Acceptance condition: The translation avoids accidental insult or inversion of meaning. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

31. Speaker naming and pronouns

Pronoun choice and caption labels can interact with visible speaker identity. Consider this fictional scene: A target language requires a distinction not specified by the original script. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Seek authoritative character or speaker information instead of inferring identity from appearance. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Review all occurrences for consistency and respect source-author preference. Acceptance condition: Unstated identity information is not invented. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

32. Visual joke and wordplay

An apparent dialogue pun may depend on an object or action in the frame. Consider this fictional scene: A character says ‘that’s a bright idea’ as a lamp flashes. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Preserve the functional humour or double meaning without falsely translating the image. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Test whether the target audience notices the intended joke at normal speed. Acceptance condition: The joke or informative contrast reaches the viewer. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

33. Chart provenance

Visual data can be preliminary while narration sounds definitive. Consider this fictional scene: A draft chart has a faint ‘illustrative data’ watermark. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Preserve the illustrative or provisional status in translated speech and display. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Check overlays and slide footnotes together. Acceptance condition: An example is not recast as observed evidence. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

34. Recorded versus live

Timing and correction options differ between live and prerecorded captions. Consider this fictional scene: A live automatic caption misrecognises a word during a public meeting. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Identify the live status and record correction approach for any published replay. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Compare live text and edited archive without claiming they were identical. Acceptance condition: The audience can tell what is provisional and what was corrected. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

35. Audience task divergence

A viewer may need translation to understand or to act immediately. Consider this fictional scene: A silent training video demonstrates an unfamiliar tool assembly. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Define whether the target must support passive comprehension or real-time task execution. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Ask a user to perform the intended task from translated media alone. Acceptance condition: Important prerequisites arrive before the corresponding action. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

36. Multi-version approval

Different language tracks may be updated at different times. Consider this fictional scene: The source video cut changes but two target subtitle files retain old timecodes. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Require explicit source cut IDs and regression checks for each language deliverable. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Verify media build, target text revision and quality-assurance signature together. Acceptance condition: No outdated target track ships against the wrong footage. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

37. Camera direction and pronoun reference

A scene can have two competing frames of orientation. Consider this fictional scene: A presenter says ‘on your left’ while looking at the viewer across a desk. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Preserve whether the phrase uses viewer-left or actor-left rather than literal left alone. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Show the translated line to a viewer who has not seen production notes. Acceptance condition: The viewer reaches the correct side of the scene. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

38. Emphasis and contrast

Stress, pitch or graphics can highlight which alternative is intended. Consider this fictional scene: The narrator stresses ‘blue’, not ‘green’, while two boxes are displayed. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Choose natural target focus marking and verify graphic emphasis. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Have a reviewer identify the selected alternative without explanation. Acceptance condition: The translation keeps the same contrast set. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

39. Sound-source location

A sound may be off-screen and make location relevant. Consider this fictional scene: A knock occurs behind the camera while the subject looks right. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Do not add a visible source or direction not established; identify source carefully if captioning requires it. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Ask where the viewer believes the sound originated after using captions alone. Acceptance condition: The translation preserves supported location and uncertainty. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

40. Permission and gesture

A spoken approval may be overridden or conditioned by visible action. Consider this fictional scene: A supervisor says ‘go ahead’ but holds up a stop palm to a different worker. The precise failure lies in the relationship between channels, not necessarily in the dictionary meaning of any isolated word. A translator should first describe the depicted event and its communicated effect in plain language, then record which channel supplies each part of that meaning.

Repair method: Segment participants and targets of instructions instead of assuming one addressee. Different language pairs can distribute information between clauses, word order, typography and context in different ways. Do not assume a short target-language equivalent will automatically carry a gesture’s reference, a sound cue’s urgency or a diagram’s distinction. The aim is semantic consistency between channels, not mechanical repetition of one sentence.

Independent test: Review who is permitted to do what at the time each gesture occurs. Acceptance condition: No authorisation transfers to the wrong person. Write down the exact source-media version and time range used for this decision. If the clip is recut or a screen label changes, this record identifies which target subtitles, narration segments, transcript passages and accessibility descriptions require renewed review.

The twenty-stage method for cross-channel translation and release

Stage 1: Define the audience and task

State what viewers must understand, identify, feel or do; distinguish entertainment comprehension from safety-sensitive task execution.

Treat this stage as a decision with evidence. A reviewer should name the exact source timestamp or visual element behind the choice, the target wording or display state, and the risk if the choice is wrong. This is more informative than checking a box marked ‘translated’, because an accurate text file can still fail when attached to the wrong scene or incompatible delivery format.

A practical passing result answers three questions: what did the source viewer understand at this moment, what can a target viewer understand now, and what difference remains? If the target viewer could plausibly choose another control, misidentify a speaker or miss an essential warning, document a concrete edit rather than treating the mismatch as an unavoidable consequence of translation.

Stage 2: Freeze the source-media version

Obtain the correct audio and video master, frame rate, duration, source-language script, on-screen graphics and change history.

Stage 3: Review the entire sequence

Watch the asset once without translating to understand plot, event order, speaker relationships and recurring visual references.

Stage 4: Inventory the channels

Identify dialogue, non-speech sound, on-screen writing, signs, gesture, facial expression, diagrams, timing, caption tracks and optional descriptions.

Stage 5: Segment by meaning beat

Use a beat boundary where the intended inference changes, even when a sentence or subtitle continues across the cut.

Stage 6: Build the reference graph

Connect deictic phrases to visible objects, pronouns to speakers, audio cues to events and labels to UI or diagram controls.

Stage 7: Assign source ownership

Check whether an instruction belongs to a character, narrator, visible presenter, external voice or on-screen system.

Stage 8: Record temporal constraints

Mark when information first becomes necessary, when it is visible or audible and when the associated action takes place.

Stage 9: Detect source contradictions

Flag mismatches between script, soundtrack, visuals or numerical overlays; do not resolve production errors through guesswork.

Stage 10: Define track deliverables

Decide which languages and modes need translated subtitles, same-language captions, transcripts, voice-over, audio description and graphics replacement.

Stage 11: Lock high-impact concepts

Agree names for controls, objects, character roles, measurements, warnings and technical actions across all target assets.

Stage 12: Draft natural target meaning

Translate for the target audience and genre while preserving conditions, referents, source intent and causal or procedural order.

Stage 13: Reconcile length and timing

Check reading speed, exposure duration, segmentation and permitted editorial condensation against the actual delivery brief.

Stage 14: Check layout and occlusion

View target captions and graphics over the final footage across relevant aspect ratios and supported devices.

Stage 15: Validate accessibility separately

Test meaningful sound cues, speaker identification, visual description and transcript completeness according to their own purposes.

Stage 16: Inspect timing as a viewer

Watch the finished clip at playback speed, listening to dialogue and observing whether written information arrives in time.

Stage 17: Test without one channel

Use mute, audio-only and transcript-only review to expose dependencies hidden by the full presentation.

Stage 18: Review target-language pragmatics

Check focus, humour, politeness, gesture interpretation, contrast and cultural assumptions with competent target readers.

Stage 19: Run regression on every edit

When the media cut, screen label or script changes, identify every connected translation segment and review the links again.

Stage 20: Record the acceptance evidence

Store source cut ID, target asset version, test devices, reader feedback, outstanding ambiguities and approval status.

Failure modes: thirty-six predictable cross-modal translation errors

The strongest way to diagnose a multimedia translation is to identify which relationship the reader loses. A good-looking sentence may have the wrong source speaker, time, referent, status or readable duration. The thirty-six cases below are invented illustrations for quality control: each defines a specific failure, a targeted repair and a practical test. For accessibility deliverables, use the relevant requirements rather than assuming a translated dialogue track performs the same job as captioning or audio description.

Failure 1: Wrong speaker on cut

Symptoms: A subtitle from an earlier voice appears over the next speaker’s face. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Retarget the cue to the actual audio turn and preserve speaker labels. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Who does a silent viewer think spoke? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 2: Pronoun without object

Symptoms: A translated this appears after the camera cuts away from the object. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Name the referent or adjust cue timing based on the real shot. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the viewer identify the item? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 3: Button-name drift

Symptoms: Narration says Submit but a visible target button says Send. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Use the approved current interface label across voice and picture. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can viewers find the intended control? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 4: Late warning

Symptoms: The caution appears after the dangerous action begins. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Expose the warning before the relevant decision or change the edit. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does the text arrive in time? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 5: Missing alert sound

Symptoms: A critical warning beep has no counterpart in accessibility captions. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Include a meaningful nonspeech cue when the track is specified as captions. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can a viewer without hearing detect the alert? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 6: Unexplained silent action

Symptoms: A demonstrator performs an important step never described outside the video. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Provide the action in a suitable descriptive transcript or audio description. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can an audio-only user reconstruct the action? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 7: Gesture overconfidence

Symptoms: An unfamiliar gesture is treated as universal agreement. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Resolve intended meaning through performance and audience testing. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does a viewer infer the source speech act? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 8: Number overlay mismatch

Symptoms: The narration and graphic show different values after target-language rounding. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Resolve numeric source discrepancies before finalising either channel. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Do viewers hear and see consistent values? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 9: Covered visual evidence

Symptoms: A caption sits over the very object used to support the claim. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Move the cue region or change composition to protect the reference. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the object and caption be seen together? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 10: Unreasonable reading speed

Symptoms: A long target sentence is given too little time for comprehension. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Condense safely or adjust approved timing and edit length. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can target readers understand at normal playback? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 11: Misleading line break

Symptoms: A negation appears only in the second cue after an affirmative first cue. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Keep negation and predicate together when segmentation is consequential. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Could the first cue invite a wrong action? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 12: Wrong audio master

Symptoms: Captions are timed to a rough track while final dialogue has different pauses. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Reconcile all cues with the approved final source soundtrack. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Which media version was checked? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 13: Stale interface visual

Symptoms: A menu label changed after the narration script was approved. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Synchronize script revision and approved screenshots. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Is the named control visible? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 14: Interrupted speech smoothed

Symptoms: A disagreement is dubbed as if both speakers agreed. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Preserve turn-taking and outcome rather than deleting conflict. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does the exchange remain contested? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 15: Forced narrative omitted

Symptoms: An essential foreign-language sign appears with no translated text. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Provide an appropriate graphic replacement or track treatment. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the audience interpret the sign? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 16: Accessibility promise missed

Symptoms: An ordinary dialogue subtitle file is labelled as full accessibility captions. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Review and deliver meaningful sound and speaker information when required. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does the selected track match its stated purpose? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 17: Description too late

Symptoms: A visual description arrives after the described object disappears. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Align the description with the image and avoid drowning dialogue. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the listener locate the event in time? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 18: Graph colour conflict

Symptoms: Speech says red line while updated chart uses orange. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Check series identity by label and data rather than colour alone. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can viewers identify the same series? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 19: Lower-third not translated

Symptoms: A speaker’s title supplies authority but never appears in target text. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Translate or describe identity information as appropriate to the medium. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can viewers identify the speaker’s role? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 20: Invented identity

Symptoms: The translator assigns a name based only on appearance. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Use verified credits or request authoritative identification. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: What evidence establishes the name? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 21: Bidirectional numbers disrupted

Symptoms: An RTL subtitle reorders product codes next to punctuation. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Validate language direction and numerals in the real player. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can numbers and labels be interpreted safely? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 22: Crop removes gesture

Symptoms: A vertical export cuts off the finger that establishes the referent. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Reframe the shot or make the referent explicit in language. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does every aspect ratio preserve meaning? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 23: Timebase conversion error

Symptoms: Cue timings drift because the frame-rate interpretation changed. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Recalculate against the final media and inspect critical cuts. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Are cues synchronized end to end? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 24: Invented sound source

Symptoms: A caption credits a visible person with an offscreen crash. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Describe only what is perceptible and avoid unsupported attribution. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Is source direction or actor actually established? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 25: Wordplay detached

Symptoms: A pun depends on a visible lamp but the target loses the relation. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Seek an idiomatic scene-compatible creative solution. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does a viewer notice the intended association? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 26: Transcript from wrong cut

Symptoms: A readable transcript reflects an older edit of the video. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Rebuild from approved content and preserve asset version references. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Do transcript and video tell the same story? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 27: Unusable overlay typography

Symptoms: Target-language screen text becomes unreadable over moving footage. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Test contrast and legibility with real footage and fonts. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can viewers read it at speed? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 28: Approval invented from gesture

Symptoms: A dubbed yes plays over visible hesitation or a refusal cue. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Retain the source speech act and check addressee. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does the scene imply consent that was not given? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 29: Duplicate competing instruction

Symptoms: A graphic and caption repeat the same command at incompatible times. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Reconcile timing and avoid contradictory cognitive demands. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: What action does the viewer expect now? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 30: Speculative audio description

Symptoms: Description says a character is angry based only on an ambiguous expression. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Describe observable facial movement or posture instead of invented motives. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the interpretation be justified by what is shown? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 31: Scene leakage

Symptoms: A caption from an earlier segment remains visible during a different action. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Retarget out-time and inspect narrative boundaries. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does text attach to the correct scene? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 32: Screen text not replaced

Symptoms: A source-language graphic stays onscreen while translated instructions assume localisation. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Specify final visual treatment and name the label actually displayed. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the viewer find the real sign? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 33: Inconsistent voice

Symptoms: The same character has sharply different status or politeness in dub and subtitle. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Use a shared character-intent sheet across outputs. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Does the character’s relationship remain coherent? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 34: Machine transcript not corrected

Symptoms: Live ASR guesses a proper name and the replay retains the error. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Verify spelling and annotate corrections in the archival version. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Is the cited person correctly identified? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 35: No recut regression

Symptoms: An edit moves the gesture but target captions retain former timecodes. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Rerun cue-to-scene verification after every source edit. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Which dependencies changed? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Failure 36: Ambiguous chart key

Symptoms: Two translated categories collapse into the same short label. The target may read naturally when separated from the footage, but it changes what a reasonable viewer can conclude when speech, picture and time are combined. The error should be classified under its real dependency rather than logged vaguely as ‘bad translation’. Ask whether the audience is likely to identify an incorrect object, misunderstand who spoke, follow a wrong instruction, misinterpret a number or miss information carried only by another channel.

Repair: Preserve different concepts and adjust visual space. Locate the approved master and identify the relevant source time range, affected visual element and intended viewer inference. Then change the smallest set of connected assets necessary to restore that relationship. If the source contradicts itself, open an author query instead of silently deciding which channel is correct. Do not invent evidence, safety clearance or character intent while smoothing the prose.

Verification: Can the audience distinguish the categories? Test the target as actual playback with realistic reading conditions, then repeat a relevant access-mode test such as silent video, audio-only listening or transcript-only reading. A failure is resolved only when the target audience can reconstruct the source relation, not merely when the subtitle file passes a technical parser. Record the result with the media version and cue identifier so later revisions are checked again.

Thirty-six worked examples: diagnose a scene, repair the dependency, retest the viewer

Every example in this chapter is fictional. The source scene describes an intended meaning; the erroneous rendering describes what an otherwise fluent translation might accidentally communicate; and the repair is a semantic target for competent translators, not a ready-made phrase for every language. Timecodes, visual identities and version choices are deliberately specified when they affect the answer. A good editor should compare the approved media with the target audience’s interpretation, not merely compare sentence strings.

Scene laboratory: Product tutorial

In product tutorial, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 1: product tutorial — scene relationship 1

Source beat: A presenter points to the upper of two nearly identical sockets and says insert the cartridge here. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: A late subtitle appears when the camera has switched to the lower socket. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The upper socket must be identified at the exact insertion moment. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Show silent playback and ask which socket viewers select. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 2: product tutorial — scene relationship 2

Source beat: The video demonstrates a clockwise turn while narration calls it tightening. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: Target narration calls it an anticlockwise turn because the translator follows the camera mirror. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The direction must be expressed from the viewer’s intended frame and match the demonstrated motion. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a novice to demonstrate the turn without extra instruction. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 3: product tutorial — scene relationship 3

Source beat: A green completion light appears just as the presenter says wait for this. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target translation calls it a red fault light. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The completion indication must align with the actual green light and not a different device state. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Pause the clip and ask when setup is complete. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Software walkthrough

In software walkthrough, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 4: software walkthrough — scene relationship 1

Source beat: The source cursor selects a button labelled Continue while the dubbed voice gives the next step. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target screen reads Next but subtitles instruct pressing Submit. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The viewer needs one consistent button identity across visible UI and spoken instruction. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a user to locate the control using only target media. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 5: software walkthrough — scene relationship 2

Source beat: An onscreen warning says data will be lost if the deletion option is chosen. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translated narration calls the action reversible despite the caution. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The irreversible consequence and condition must be stated consistently across screen and voice. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask the viewer what happens after selecting Delete. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 6: software walkthrough — scene relationship 3

Source beat: A toast confirms that saving has finished just before the speaker starts a new topic. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target caption arrives after the toast has disappeared. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The saved status should be legible and correctly aligned with the relevant narration. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Watch once without pausing and ask how the user knows the save succeeded. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Science lesson

In science lesson, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 7: science lesson — scene relationship 1

Source beat: A teacher circles the dashed line on a chart and says this is our comparison series. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target subtitle says the solid line was the comparison. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The dashed series must remain the referent and its label must match the graphic. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a learner to point out the compared data series. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 8: science lesson — scene relationship 2

Source beat: A labelled animation shows a gas flowing from a reservoir into a vessel. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translated narration reverses the flow direction. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The movement goes from reservoir to vessel and the direction matters to the explanation. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask the learner to redraw the flow after seeing the target. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 9: science lesson — scene relationship 3

Source beat: A slide illustrates a hypothetical model with a visible simulation watermark. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target voice-over states that the animation is direct video evidence. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The model must remain labelled a simulation rather than an observation. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a viewer what kind of evidence the visual provides. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Safety training

In safety training, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 10: safety training — scene relationship 1

Source beat: An alarm sounds and a person stops approaching a moving component. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translation caption omits the alarm and gives the stop cue only after the person has stopped. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The warning must be perceptible before the depicted decision to stop. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Run silent playback with captions and test recognition of the alert. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 11: safety training — scene relationship 2

Source beat: The narrator says do not remove the cover while power is connected. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The first target subtitle reads remove the cover and the negation appears after a cut. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The prohibition and its condition must remain a single intelligible instruction. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a reviewer what they would do after reading the first cue. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 12: safety training — scene relationship 3

Source beat: A demonstrator tests for zero energy before beginning work. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translated transcript places the test after the work begins. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The verification precedes the work and must remain in the correct sequence. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a user to list prerequisites from the target transcript. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Museum video

In museum video, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 13: museum video — scene relationship 1

Source beat: A guide refers to a small ceramic vessel on the visitor’s right. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target describes the larger vessel beside it. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The identified object must remain the small right-hand vessel in the displayed scene. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Have a viewer point to the referenced exhibit. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 14: museum video — scene relationship 2

Source beat: An actor reads a historical letter and a curator then offers an interpretation. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The caption track attributes the curator’s interpretation to the historical letter writer. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: Source quotation and present-day interpretation must remain separate voices. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask which statements come from the original letter. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 15: museum video — scene relationship 3

Source beat: A visual map changes period while the narrator explains an earlier route. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target subtitle stays visible into the later map and makes the historical period ambiguous. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The earlier-route claim belongs to the earlier map and its stated date. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask a learner when the depicted route existed. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Interview film

In interview film, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 16: interview film — scene relationship 1

Source beat: A witness says a colleague told them an entrance had been closed. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translation says the witness personally saw the entrance locked. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: A reported statement must not become an eyewitness claim because illustrative footage is shown. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask what the interviewee could personally confirm. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 17: interview film — scene relationship 2

Source beat: Two participants overlap in speech with opposing interpretations of one event. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The subtitling compresses both voices into a single unqualified agreement. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The audience must retain the disagreement and identify both speakers. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask the viewer which participant disputed the conclusion. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 18: interview film — scene relationship 3

Source beat: The camera shows an interviewee as a narrator asks a question from offscreen. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: A caption attributes the question to the visible interviewee. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The question belongs to the offscreen narrator and the answer to the interviewee. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Test speaker attribution with volume muted. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Retail demo

In retail demo, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 19: retail demo — scene relationship 1

Source beat: A presenter holds a smaller model and calls its capacity a certain value. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The localised graphic shows the capacity of the larger model. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The product variant and capacity must be a matched pair. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask the buyer which model offers the stated capacity. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 20: retail demo — scene relationship 2

Source beat: A discount price flashes onscreen with a minimum-purchase condition. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: Voice-over translation drops the condition. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The price and eligibility condition must remain connected and legible. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask the buyer what is required to qualify. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 21: retail demo — scene relationship 3

Source beat: The visible package says a feature is optional while the narration says available on request. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target subtitle renders the feature as included by default. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The buyer must understand it is optional and requires a request. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask whether the feature is automatically included. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Sports clip

In sports clip, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 22: sports clip — scene relationship 1

Source beat: A player passes toward the left just before the camera cuts to another player. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: A delayed translation makes the second player the passer. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The action must remain attributed to the first player. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask who made the pass without replay. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 23: sports clip — scene relationship 2

Source beat: An official raises an arm while commentary says a decision is being reviewed. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translation announces a final ruling. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The visible signal and spoken status must not be interpreted as confirmation of a decision. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask if the matter has been resolved. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 24: sports clip — scene relationship 3

Source beat: A scoreboard gives a team’s total while the voice cites an individual record. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translation assigns the team total to the athlete. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The total and the individual value must keep their distinct subjects. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask which number belongs to whom. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Documentary chart

In documentary chart, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 25: documentary chart — scene relationship 1

Source beat: An animated chart shows an estimated range and a midpoint. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target narration describes the midpoint as an exact observed value. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The range represents uncertainty while the midpoint is a summary statistic. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask viewers to report the range and what it means. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 26: documentary chart — scene relationship 2

Source beat: One map icon denotes observations while a second denotes projected conditions. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: Both icon labels are translated as confirmed observations. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The visual key must distinguish measurements from forecasts. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask viewers which regions are observed and which are forecast. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 27: documentary chart — scene relationship 3

Source beat: A dataset end date changes after the footage is updated. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: An older subtitle retains the former end year. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The displayed period and speech must reflect one authorised dataset version. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Compare spoken and visible dates in the final release. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Live meeting

In live meeting, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 28: live meeting — scene relationship 1

Source beat: A presenter says look at this column while the shared spreadsheet scrolls. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: Delayed captions point toward a different column. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The referenced column must be named or temporally aligned to its appearance. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask attendees which figures were under discussion. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 29: live meeting — scene relationship 2

Source beat: Automatic captions garble a participant’s surname and the archive keeps the error. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The archive attributes a position to someone with a different name. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The corrected published recording must preserve accurate attribution. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Compare transcript against the verified participant list. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 30: live meeting — scene relationship 3

Source beat: A draft slide has an approved-looking template placeholder while the speaker says no decision has been reached. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target narrator accepts the placeholder as final policy. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The provisional status in speech must not be overridden by a misleading source template. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask whether the target audience believes the proposal was adopted. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Short drama

In short drama, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 31: short drama — scene relationship 1

Source beat: A character quietly hides an object as another asks a question. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translation adds explicit wording saying the character is lying. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The picture should carry the dramatic irony without unsupported new dialogue. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask viewers what they infer and how they inferred it. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 32: short drama — scene relationship 2

Source beat: A visual pun relies on a literal key shown as someone uses a figurative expression. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target chooses a phrase with no possible connection to the key. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The language should preserve the scene’s comic or meaningful contrast where achievable. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Test whether viewers notice a meaningful relation. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 33: short drama — scene relationship 3

Source beat: A long hesitation before a cautious response is shortened in dubbing. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The target speaker sounds fully confident and willing. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The underlying uncertainty and interactional beat should survive the target performance. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask if the character seems certain or hesitant. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Scene laboratory: Vertical tutorial

In vertical tutorial, fluent translation is not enough when the viewer must connect words to a visible object or timed event. Review the source as a scene with its own goals, not as a collection of disconnected subtitles. Record what changes when the audience can hear but not see, see but not hear, or read a descriptive transcript instead of watching. A domain specialist should flag factual ambiguity; the translator should preserve the authorised meaning across the target forms.

Worked example 34: vertical tutorial — scene relationship 1

Source beat: A pointing hand is cropped out of the portrait export while the narration says that one. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The translated subtitle retains the unresolvable pronoun. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The target should specify the control or use a crop restoring the visual referent. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Show the portrait export to someone who has not seen landscape. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 35: vertical tutorial — scene relationship 2

Source beat: Two captions cover a small warning label in the lower-left corner. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The user sees only the action and misses the limitation. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The limitation and demonstrated action must both remain visible or otherwise available. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Review on a small-screen device with captions on. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Worked example 36: vertical tutorial — scene relationship 3

Source beat: A fast cut from a measurement display to the final product overlaps a translated unit. The important information is distributed across more than one channel. Determine the speaker, action, referenced object or numerical quantity, and the exact moment at which the relationship becomes understandable. Draw a simple alignment record: source time span; visual event; spoken or displayed phrase; intended audience inference. It is possible that neither the dialogue alone nor the image alone contains enough information to complete that record.

Faulty target: The unit looks like a property of the product rather than the measurement. This may be perfectly grammatical. Its problem is that a target viewer receives a different claim or instruction when the wording is combined with the final video. It may change direction, speaker, evidentiary status, object identity or temporal order. Identify the lost dependency, rather than simply changing the phrase to a vague synonym. If the source itself contains inconsistent cues, escalate the issue to the content owner before choosing a reading.

Required reconstruction: The measurement and its unit belong to the earlier device screen. The target-language author should now choose a sentence, cue placement, speaker marker or display treatment appropriate to the actual delivery medium. Keep the concept stable even if the natural word order changes. An additional noun phrase can solve an otherwise unresolvable pronoun, but it may also exceed the reading window; when that happens, coordinate with the video editor instead of abandoning an essential relation.

Acceptance test: Ask what quantity the unit describes. Have an independent reviewer use the released target video without access to the source script. Their answer should identify the intended actor, action, referent, timing and status. If the participant gives the wrong answer, capture whether the failure arose from translation, visual design, audio mix, reading speed or source contradiction. Save the repair with the cue ID and cut version so it can be regression-tested after any recut.

Transfer practice: rebuilding meaning for new media modes

Transfer workshop 1: Caption-only playback

Task: Turn off sound and confirm that spoken actions, speakers and meaningful audio events remain understandable in the caption deliverable. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Successful output is a testable result, not a polished assertion. Have a second reader or viewer describe what they understood and compare the account with a source-alignment ledger. Note one repair option, one limitation and an explicit release criterion. Transfer is complete when the relation survives a change in medium, not when the same sentence appears verbatim in all media.

Transfer workshop 2: Audio-only playback

Task: Remove the image and find visual information necessary for comprehension that the soundtrack does not supply; identify appropriate description options. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 3: Descriptive transcript

Task: Move temporal visual relationships into readable text without inventing dialogue or converting speculation into observed fact. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 4: Screen-cropped playback

Task: Test a portrait and square crop to find missing fingers, arrows, captions or labels required to interpret the language. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 5: No-pause reading

Task: Play a complex segment without stopping to see whether meaning is genuinely acquired inside the allowed reading window. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 6: Control-label switch

Task: Localise the UI artwork and check that every spoken instruction and caption points to the final displayed control. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 7: Silent demonstration

Task: Find actions that remain visually clear and those needing text or audio descriptions to be accessible in alternative forms. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 8: Different script direction

Task: Test mixed punctuation, numbers, names and interface codes in a right-to-left target using actual caption rendering. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 9: Character-voice comparison

Task: Compare the same speaker in dub, subtitles and transcript to detect unintentional changes in status, certainty or register. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 10: Source revision

Task: Re-edit an old scene and measure which alignment decisions must be renewed before the translation can be released again. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 11: Attention competition

Task: Test a scene with captions, on-screen labels and a key moving object to determine which information is obscured or overloaded. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 12: Contrast and focus

Task: Change which word receives emphasis and verify that the target audience identifies the intended comparison. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 13: Ambiguous gesture

Task: Ask a target-language viewer to explain the gesture without assuming that one culture’s reading must be universal. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 14: Visible contradiction

Task: Show a clip where narration and picture disagree; document a source-author query instead of inventing a final answer. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 15: Speaker overlap

Task: Preserve two overlapping speech acts in a usable target mode without fabricating a single consensus. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

Transfer workshop 16: Timed correction

Task: Replace an inaccurate number in the source cut and verify all captions, descriptions, transcripts and graphics change together. Begin by stating which semantic relation matters most and why the selected output mode might break it. Identify the source channel carrying that information, the target channel available to the audience, and the practical editorial choice that reconstructs the relation. A missing channel should not be patched by inventing unseen facts; a faithful alternative should describe what the source actually establishes.

A practical acceptance checklist for multimedia translation

1. Does every spoken deictic reference point to the correct visible object at the cue time? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

2. Are the localised screen labels and narrated control names consistent with the final UI? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

3. Are speaker identity and quotation boundaries accurate across camera cuts? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

4. Are warnings and conditional instructions perceptible before the relevant action? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

5. Are quantities, units, chart series and dates consistent across audio, captions and graphics? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

6. Can a user distinguish simulations, forecasts, examples and verified observations? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

7. Is meaningful non-speech audio available in the specified caption track? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

8. Have audio-description and transcript deliverables been tested separately rather than assumed complete? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

9. Is each caption readable at playback speed within the relevant delivery specification? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

10. Are captions positioned to avoid important hands, labels, faces and evidence? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

11. Has bidirectional display, typography and language-specific layout been tested in actual target players? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

12. Have creative humour, gestures, contrast and social intent been checked with target-language reviewers? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

13. Is the exact source cut and target asset version recorded? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

14. Are source contradictions escalated instead of silently rewritten? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

15. Has a target-only reviewer successfully reconstructed meaning after the final mix? A pass requires observed evidence from the assembled media, not a verbal assurance. Record the tested language, output mode, media version and relevant timecode so that subsequent revisions can be assessed against the same criterion.

Research and technical sources

These references support terminology and requirements, but they do not replace project-specific specifications, playback testing or qualified language review. Source publication dates matter because technical standards and provider guidance may change. The following links are checked for the November 2026 editorial context.

W3C WAI, Captions/Subtitles explains captioning, translated subtitles, meaningful non-speech audio and the distinction between track purposes. It is a useful starting point for specifying rather than conflating deliverables.

W3C WAI, Making Audio and Video Media Accessible distinguishes captions, transcripts, audio description, sign language and accessible media-player requirements. W3C WAI, Transcripts details descriptive transcripts and ways they can include visual information.

WCAG 2.2 Recommendation provides normative accessibility success criteria for prerecorded and live media. Consult the relevant conformance requirements rather than claiming that the presence of any translated subtitle track alone establishes full accessibility.

W3C WebVTT specification defines cues, cue timing and presentation structures used in web text tracks. A syntactically valid WebVTT file is not automatically an accurate or accessible translation.

W3C IMSC Text Profile 1.3 Recommendation (21 May 2026) defines a text-only TTML2-based profile for international subtitles and captions, including display and authoring considerations. The publication is a technical profile, not a universal linguistic quality metric.

Netflix English (USA) Timed Text Style Guide illustrates provider- and language-specific line and reading-speed expectations. Apply the actual current target-language client specification instead of treating the English limits as universal.

W3C Internationalization: Unicode controls versus markup for bidirectional support explains why mixed-direction text needs careful rendering and why structural and inline contexts may call for different approaches.

Next routes and editorial ownership

This system chapter takes responsibility for a distinct translation problem: the connection between language and simultaneously presented media. Readers who need language-specific vocabulary should use the Vocabulary Learning Hub. Readers studying English dialogue, reference or discourse should use How English Works. Readers implementing TTML or IMSC should follow specialist format standards and the existing technical eduKate guide. This chapter does not replace any of those owners or reuse their narrower tutorial titles.

Return to Master Art of Translation | The Complete System for Moving Meaning Between Languages for the architecture and related study routes. The preceding chapter, The Procedural Action System, explains how to keep instructions executable when steps and conditions change across languages.

Use the protected Vocabulary Learning Hub for word and concept learning, and How English Works | Translation and Meaning Across Languages for English-specific language explanation. Technical readers may also consult the existing W3C TTML2 Timed-Text Translation System for format-level validation rather than mistaking this cross-modal chapter for a markup specification.

Series architecture: EDKSG-TRANS-MASTER-WORLD-1350. World-facing editorial reference; no geography-specific assumptions.

The final test is not whether a subtitle is grammatical. It is whether a target reader or viewer receives the same actionable relationship among speaker, event, object, sound, image and time. When that relationship survives captions, voice, graphics, transcripts and accessibility modes, translation has preserved the meaning of the scene rather than simply moving its words.

Master Art of Translation | The Multimodal Meaning Alignment System — How Speech, Captions, On-Screen Text, Gesture and Timecodes Stay Connected