How People Translate Quickly | IMSC Text Profile 1.3 Caption QA: Translate Dialogue Without Losing Timing, Regions or Accessibility

To translate IMSC 1.3 subtitles quickly, replace only the intended human language inside the timed-text document, assign the right language, and check every cue against the same video clock and display regions. A convincing translation that appears after a speaker has finished, clips against a screen edge or disappears behind a graphic is not ready for delivery. Keep a source-to-target cue ledger, validate XML, test the profile, render the file, and review it with sound and picture together.

IMSC Text Profile 1.3 is a W3C Recommendation dated 21 May 2026, based on Timed Text Markup Language 2. It is a defined text profile, not simply a different extension for ordinary subtitles. The translation project must preserve the identities of timing constructs, language declarations, style and layout references, and any delivery constraints while letting target-language sentences take a natural length and order. The most important distinction is between readable words and machine instructions that place those words on the screen.

This guide follows a fictional bilingual science video used in a Singapore classroom. Its publisher has approved the English script, a timed IMSC file and a specific display size. The target language is Malay, with a separate lesson in adapting captions for Japanese. We will inspect the source, prepare a safe translation surface, diagnose failures, propose repairs and perform a changed-context exercise. The examples teach method rather than claiming certification by a particular streaming service, device or IMSC validator.

Start with the viewer’s actual task

A learner watches a scientist fill two measuring cylinders while the teacher explains that one liquid has a greater volume. The caption must make the comparison intelligible at precisely the moment when the correct cylinder is visible. An elegant target sentence that reverses which cylinder is larger destroys the learning task. A correct sentence that arrives during the next experiment may be equally misleading because the learner connects it to the wrong image. Language accuracy and timed evidence have to be tested together.

Write a release criterion before translating: the learner should identify which cylinder has the larger volume without hearing the narration, while a hearing viewer should not experience a conspicuous disagreement between the spoken explanation and captions. Note where visual evidence appears, when the relevant words are spoken, and whether a speaker change needs identification. This criterion becomes more useful than a vague instruction to make the subtitles short, because it shows which detail may not be deleted during compression.

Identify the profile instead of guessing from the filename

Files named .ttml, .xml or .itt are not automatically interchangeable. TTML2 is a broader timed-text language, while IMSC defines a subset and additional constraints intended to improve delivery interoperability. The W3C published IMSC Text Profile 1.3 on 21 May 2026; the corresponding profile designator can appear in a content-profile declaration. A vendor’s ingest pipeline may accept a particular older profile or a restricted implementation, even when a document follows a newer W3C Recommendation. Record both the normative specification and the destination platform’s contract.

If the incoming file does not state a profile, request the delivery specification rather than setting the newest version by assumption. Translators often receive a file already exported by editing software that silently alters namespace declarations or introduces unsupported styling. The absence of visible trouble in the editor is not proof of profile validity. Keep an untouched original and a checksum, note the software and export settings, and validate the exact final bytes that will be delivered, not an intermediate preview.

Separate words, semantics and the timed-text machinery

The main human-language surface usually lives in paragraph and span content. The surrounding XML may contain a document language, namespace declarations, a timing model, styles, regions, IDs and references. Translate words only after classifying each component. An element name, a namespace URI or a reference such as region="lower" is not English prose even if it looks readable. Changing it can disconnect captions from their intended placement or cause the processor to reject the file entirely.

Create three columns: translatable utterances, reviewable human-facing metadata, and protected structure. Each cue should retain a source ID or stable editorial row number, an incoming start and end time, speaker and intent, target sentence, and a reviewer note. Protect numbers spoken as measurements by marking their role: a quantity may be translated in words, but its physical value and unit relationship must survive. This is especially important when an onscreen diagram provides labels that the learner must match to the subtitle.

Inspect timing before choosing target wording

A cue with begin and end attributes occupies an interval on the media timeline. Record the timeline model used by the asset and check whether cue intervals overlap intentionally. Simply counting characters in English does not establish that a Malay, Arabic or Japanese translation can be read in the same window. Some target phrases expand; others are shorter but require different syntactic order. The real question is whether the audience can understand the intended proposition while the corresponding shot remains visible.

Use a cue ledger with start, end, duration, source characters, target characters, essential information and measured render outcome. Identify hard constraints: a reaction shot may allow no timing drift, a title card may contain a fixed name, and a scene cut may make a late subtitle misleading. Translation can reorganise wording within the available interval, but moving a caption across a visual evidence boundary requires review by an editor who can inspect the video, not only a language specialist reading a text export.

Keep the W3C language contract meaningful

The xml:lang attribute identifies natural language and is not a magic translation instruction. In a translated document, declare the intended language where it applies and inspect inherited language context. If a caption contains an untranslated quotation or a classroom term from another language, its language metadata may need more careful handling than a document-wide replacement. Keep the chosen language tag in a delivery manifest so a player or distribution system can select the correct asset for viewers.

Do not confuse language with script, locale and audience. A target file in Malay may use the Latin script, but that does not justify copying English text into its translated track. A regional expression may be understandable to one group and unfamiliar to another. Define terminology, register and intended ages before translation. For scientific language, have an educator approve terms for volume, mass, displacement and scale; the correct everyday word can still be wrong for a concept students are supposed to distinguish.

Understand the role of styles and regions

A TTML document may define styling in its head and reference named regions from cues. A region describes where content can be laid out; styles affect properties such as size, alignment and contrast. Target text that is wider than the source can wrap into an extra line, approach the bottom of the picture or collide with an important diagram. Translating the line without rendering it is like changing a poster’s headline while refusing to inspect the finished poster.

Track all referenced regions and styles in a preflight table. A region ID that exists in the source must still resolve after editing. Do not change every position to accommodate one long sentence; that adjustment may obscure another shot. First try a shorter faithful sentence, then line breaking, then an approved layout adaptation. Preserve the relationship between content and imagery, and compare a frame capture before and after changes. Layout repair is an editorial decision constrained by technical rules, not an automatic permission to reduce font size indefinitely.

Work a small XML example without overstating validation

This illustrative TTML fragment shows the kind of boundary to preserve. The paragraph’s human words may be translated, while timestamps, profile signaling, namespaces and referenced region names belong to the technical contract. An XML parser can establish that markup is well formed, but that alone cannot prove that the document satisfies every IMSC 1.3 profile requirement or renders correctly in the intended player. The example is therefore a learning specimen, not a guaranteed production delivery file.

<?xml version="1.0" encoding="UTF-8"?>
<tt xmlns="http://www.w3.org/ns/ttml"
    xmlns:ttp="http://www.w3.org/ns/ttml#parameter"
    xmlns:tts="http://www.w3.org/ns/ttml#styling"
    xml:lang="ms"
    ttp:timeBase="media"
    ttp:contentProfiles="http://www.w3.org/ns/ttml/profile/imsc1.3/text">
  <head><layout><region xml:id="lower"
      tts:origin="10% 75%" tts:extent="80% 20%"/></layout></head>
  <body><div><p region="lower" begin="00:00:08.000"
      end="00:00:11.000">Silinder kedua mempunyai isi padu lebih besar.</p></div></body>
</tt>

A safe reviewer checks that the Malay sentence says the second cylinder has greater volume, that the second cylinder is visible from eight to eleven seconds, and that text remains in the intended lower region. They also test that the profile identifier is supported by the delivery system. If an engineer changes the region ID from lower to bawah, the visual result is not an improved translation; it is a broken reference. If a translator copies English xml:lang into a Malay file, the media system receives misleading language metadata.

Design a source-to-target cue ledger

A cue ledger is more than a spreadsheet of source and target strings. It should preserve cue ID, start, end, speaker, on-screen evidence, character count, critical terms, and any caption-for-accessibility marker such as a sound description. Keep a traceable revision field and the person who approved the semantic change. This prevents a later editor from making a shorter caption while unknowingly deleting the very distinction that a teacher or examiner had asked students to notice.

Give each cue an explicit decision status: untouched, translated, needs timing review, needs layout review, scientific terminology dispute, or approved. Avoid colouring cells without a textual status because colour may not survive a format conversion or assistive technology. Pair every approved target with the source revision used to create it. When the narration changes, you can then identify which translations need revalidation rather than comparing whole videos by memory and assuming that unchanged timestamps mean unchanged meaning.

Translate meaning rather than matching source word order

A good subtitle conveys a proposition in time, not each grammatical item in the same sequence. When an English speaker says, “The volume of the second cylinder is greater, even though the levels look similar,” target-language syntax may require the comparison earlier or later. The caption’s essential claim is the relationship between two cylinders, supported by an observation about the levels. A shorter target sentence can retain the relationship while removing a redundant conversational phrase, but it must not imply both volumes are equal.

Mark three components before editing: claim, evidence and conversational padding. Keep the claim and evidence that determine the educational answer. Padding can sometimes be removed if it does not affect tone, cause, qualification or speaker intention. Beware of hedges such as “approximately,” “may” and “under these conditions.” They often look expendable in a length-restricted cue, yet removing one can turn a tentative scientific observation into an absolute rule. Compression should not make a sentence easier to read by making it more false.

Respect line breaks as a reading decision

Line breaks affect how viewers group clauses. A break between a quantity and its unit may slow recognition; a break after a negation can lead to a momentary false interpretation. In languages whose word order differs from English, the original line break may divide a grammatical unit. The translator should propose a line boundary that suits the target language, then confirm that the renderer respects it and that the resulting caption remains within the allocated region.

Prefer breaks at meaningful phrase boundaries, but do not assume a line count or character-per-line limit that the receiving platform has not specified. IMSC profile conformance and a broadcaster’s style guide are related but separate layers. A broadcaster might impose readability or safe-area requirements stricter than the formal profile. Test both. Keep screenshot evidence for the longest and most complex cues because their layout defects will not always be predicted by a spreadsheet’s character counts.

Make numbers, units and notation unambiguous

An experiment subtitle may mix numerals, spoken quantities and graphical units. When a teacher says “two point five litres,” the target caption should retain the numeric relationship and the correct unit. A locale may use a different decimal separator in ordinary text, but the visual display may continue showing a decimal point. The learner must recognise that they refer to the same value. Always cross-check against what is actually visible rather than applying a global punctuation substitution.

A scientific video also uses symbols that look like ordinary letters: m for metre and M for molarity are not interchangeable. Preserve superscripts and subscripts where they change meaning. IMSC Text Profile 1.3 specifically adds support for superscript and subscript text compared with the earlier text profile, but that does not guarantee every distribution endpoint implements the feature identically. Prepare a compatibility note and test the target renderer with the relevant scientific expression before treating a mathematically correct text file as a successful caption.

Carry speaker identity and sound information deliberately

A translated subtitle track and a caption track for deaf or hard-of-hearing viewers may have different communicative duties. A dialogue subtitle may focus on speech, while a caption can also indicate a siren, laughter or a speaker entering off-screen. Do not delete meaningful sound cues merely because they are not spoken words. If the educational meaning depends on an alarm or an audible countdown, the caption must communicate its function and timing for viewers who cannot hear it.

When multiple people speak, determine whether speaker identification is necessary from visual context. The symbol used to signal speaker change may differ by house style; do not invent a style rule and declare it mandatory under IMSC. Keep an editorial note for overlapping speech and decide how to prioritise essential information. A classroom video can still be accessible when a brief phrase is omitted, but it should not misattribute a safety instruction to the student rather than the teacher.

Test right-to-left and mixed-direction captions

A later Arabic version may combine right-to-left text with Latin chemical symbols, numerals or a displayed website address. The raw order of Unicode characters is not enough to predict the final result. Verify shaping, punctuation placement, bounding boxes, reading order and any embedded Latin unit in the actual rendering environment. A technical name that appears in an unexpected position can make a caption confusing even when the Arabic translation itself is correct.

Set a separate bidirectional-text test pack covering parentheses, negative numbers, ratios, abbreviations and speaker identifiers. Do not solve a rendering defect by adding random directional control characters without documenting them; such controls can make searching, copy-paste and later conversion difficult. Compare at least two representative target devices or renderers where available and document what each one does. Translation QA here includes typography, layout and software behaviour, not only spelling.

Use Japanese typography tests for scientific lessons

Japanese captions may render vertically or horizontally depending on the programme’s design, but editors should not infer that the profile automatically chooses the appropriate treatment. The 2026 IMSC Text Profile 1.3 Recommendation adds Japanese-language authoring guidance. For a horizontal school video, check punctuation, brackets, character width, line wrap, ruby or pronunciation support where the approved format permits it, and whether the audience can read the result before the cue ends.

A Japanese target might be shorter in visible glyph count while still demanding reading time for unfamiliar terminology. The audio and the onscreen experiment determine when the sentence is useful. Have a Japanese-language editor review technical terminology and a separate renderer test confirm the actual appearance. This separation matters because a fluent reviewer can approve a sentence in an editor that renders correctly as plain text but fails when a player substitutes a different font or positions annotations outside the safe region.

Never treat profile validation as semantic validation

XML parsing catches malformed tags or undeclared prefixes; a profile-aware validator checks a more demanding set of structural and feature constraints. Neither test decides whether “less” was translated as “more,” whether a chemical formula was mistranscribed or whether a subtitle explains the correct shot. That distinction matters for procurement, where a vendor might present a green validation report as evidence of full translation accuracy. The report proves the scope of the checks it performed, not every possible quality requirement.

Use a three-layer gate. First, parse XML and compare protected structural fields to the baseline. Second, validate against the specified profile and receiving-system rules, retaining the tool version and any warnings. Third, render the target with the video and conduct human semantic and accessibility review. A defect in any layer blocks release until repaired or explicitly accepted by the owner of that requirement. This prevents one successful automated check from concealing a serious failure somewhere else.

Construct a practical invariant diff

A translator does not need to understand every XML namespace to benefit from a focused structural comparison. Extract the cue identifiers, start and end times, region references, style references, profile declaration and selected metadata from the source and target. Define which fields are permitted to change, and compare every other field exactly or according to a documented conversion rule. When a tool recreates IDs, require an explicit mapping to demonstrate that information was not lost.

The diff should report concrete deviations rather than a blanket assertion that the files differ. For example, changing xml:lang from English to Malay is expected, while changing the second cue’s end time from eleven seconds to thirteen seconds requires an editorial timing decision. An unplanned change to the profile URI is a technical defect until justified. Keep this report alongside target screenshots and the glossary; together they explain why the translated asset is both meaningfully equivalent and technically fit for delivery.

Check timing against cuts, gestures and delayed captions

In a demonstration, the scientist points to a measurement just before a camera cut. A subtitle that survives into the next shot can cause the learner to attach its comparison to a different object. Conversely, entering too early may reveal an answer before the evidence appears. Record cue timings alongside shot boundaries, not only the voice waveform. For a classroom explainer, a two-second displacement can be educationally material even when a viewer still understands the general topic.

Review the shortest cues, cues near scene changes, and cues whose target language becomes longer. If the content cannot fit, choose among authorized repairs: tighten nonessential wording, redistribute a sentence across adjacent cues that refer to the same scene, or ask the editor to adjust timing within the approved video. Do not shift an instruction beyond the point where the student can act on it. Any change should be replayed against picture and sound after the final export.

Inspect regions against background information

A subtitle near the bottom of the frame may cover a measurement label, safety instruction or speaker’s hand. A translation that wraps into an extra line can hide evidence that the caption is supposed to describe. Compare the source and target overlay against the actual frame at the cue’s most crowded instant. A static black screen used by a translation vendor is insufficient because it ignores conflicts with objects and captions elsewhere in the picture.

For a low-contrast background, a named style may need an approved adjustment, but avoid changing an entire programme’s typography to repair one cue. Identify whether the problem is the string, line break, selected font, region extent or unexpected fallback behaviour. Log the final cause and test the repair on related cues. A systematic font metric change may move every caption, so a local correction should not be released without sampling the full video again.

Build a failure-mode and repair record

Common failures include an untranslated cue, the wrong language tag, a renamed region, clipped text, a stale timestamp, corrupt markup, incorrect speaker identity and an inaccurate scientific comparison. Each failure has a different owner. A glossary issue belongs with a language or subject expert; a region reference may belong with the exporter; a media offset may belong with post-production. Assigning every issue to “translation” delays resolution because the person best equipped to repair it may not be the translator.

Write each defect as an observable statement: “Cue 14 says the first cylinder has the greater volume, whereas the narration and visible scale show the second.” Include the source cue, target cue, video timestamp, environment and a screenshot. Then record the required repair and a retest result. Avoid vague comments such as “looks odd” without a reproducible frame. A precise defect makes review faster and leaves an audit trail for the next language or updated source video.

Create a regression set rather than replaying at random

Build a compact set of edge cases that exercises every class of risk. Include a long sentence, a scientific number, an off-screen speaker, a shot change, a sound cue, a region near important graphics, a mixed-direction string if relevant, and an example of superscript or subscript. Each case should have an expected semantic result and an expected visual result. When the file is regenerated, rerun this set before trusting the export.

Regression testing is especially valuable when an editor changes caption software halfway through a project. The XML can remain well formed while default styling, whitespace handling or line-breaking changes. Compare rendered frames and timing records, not just source files. Keep the test asset, the source version and the output profile in the release packet so that future re-export work can reproduce the checks rather than inventing new criteria under deadline pressure.

Manage updates to the narration without overwriting approvals

A revised science lesson might replace “larger” with “approximately twice as large.” That is not a cosmetic source change: the target must now preserve an approximate ratio. Link each translated cue to the source revision that was reviewed. If one cue changes, mark its approval invalid until the target has been checked again, even when the previous translation still sounds plausible to the language editor.

Treat updates as constrained changes rather than starting from a blank file. Retain IDs and timing where the media remains unchanged, but do not assume continuity if an editor inserts a new camera shot or re-records the narration. A stable cue ID can survive a substantial revision and mask a semantic difference. Versioned hashes, source wording and notes about affected shots give the team evidence for deciding which previous approvals can be carried forward and which need new review.

Draw a firm line between translation and editorial adaptation

Some captions can be translated faithfully within the existing timing; others need adaptation to fit the viewer’s reading window. Decide in advance who is permitted to shorten, reorder or paraphrase. An editor may approve removal of an aside but refuse deletion of a safety condition or scientific hedge. If a caption says “Do not add water until the flask cools,” shortening it to “Add water when cool” subtly changes the action and could be unsafe in a demonstration.

Document every adaptation that changes emphasis, timing or on-screen placement. The point is not bureaucracy for its own sake. When a teacher reviews an error months later, they need to know whether the translation introduced it, the timing forced it, or the source itself was ambiguous. A light approval record allows faster corrections. It also shows why “shorter” is not a universal measure of translation quality: the task is to preserve the right decision for the viewer.

Handle fonts and unsupported characters explicitly

A font may contain Latin letters but lack the glyphs or metrics needed for another script. A fallback font can alter line length and cause clipping even when a developer has not changed the region dimensions. Confirm which fonts are available on target playback devices, whether any external font references are permitted, and how the distributor handles unavailable glyphs. IMSC support must be considered together with the player’s capabilities and the distribution specification.

Do not silently replace scientific symbols with visually similar letters to fit the font. That can change a quantity, formula or pronunciation cue. If a symbol cannot be rendered, escalate the issue and propose a documented alternative that the subject-matter expert approves. Test diacritics, combining marks, punctuation and line height. The appearance of a word on a desktop preview does not guarantee that it will fit inside the same region on a low-power television or a classroom projector.

Validate the exact exported asset rather than a working copy

Translation tools often round-trip XML through an internal representation. That process can alter whitespace, ordering, numeric precision or unsupported attributes. A file that was valid before import may emerge different after export. Keep an untouched input and compare the exported target with the agreed structural invariants. Validate the output after compression, packaging or conversion where a delivery workflow applies further transformations.

A useful final gate checks the file hash, XML well-formedness, expected profile, target language, list of IDs, cue timing intervals, region references, font availability and visible playback. Make a small report that an educator and a technical reviewer can both interpret. If the ingest service refuses the file, preserve the error and determine whether it is a real profile violation or a narrower vendor restriction. Do not weaken the source standard in a misleading way just to silence an opaque error.

Prepare the changed-context transfer exercise

Imagine the same lesson is now a museum kiosk with no sound, a smaller screen and a visitor who can approach midway through the demonstration. The original captions were built for a classroom video, and a simple language swap would not make them adequate kiosk instructions. The reader must decide which elements can be reused, what must be re-reviewed and how the test evidence should change. A kiosk has different attention patterns, display constraints and accessibility expectations.

The exercise is to propose a safe release plan for a Japanese caption track, retaining the experiment’s volume comparison and one audible alarm that now needs a visual equivalent. Assume the kiosk player supports an identified IMSC text profile but has not been tested with Japanese fonts. Decide whether source timestamps, xml:lang, region sizes, accessibility cues and spoken quantity units can be copied unchanged. Then explain what constitutes a successful test when visitors cannot rewind and may miss the opening explanation.

Transfer exercise answer and rationale

Preserve machine identifiers and the evidence-bearing relationship between the two cylinders, then translate the wording and language metadata for Japanese. Treat timestamps as a starting point, not immutable truth: compare them with the kiosk footage, shot boundaries and new reading conditions. Review region extents on the smaller screen, select and test Japanese font support, and include an appropriate caption for the alarm because there is no sound. The quantities and their units must still agree with the display.

Test the longest captions, a visitor entering midway, the alarm cue and the volume comparison at the kiosk’s actual resolution. If a learner could see the comparison but miss the words because the region obscures the scale, the deliverable fails. If font fallback makes a chemical symbol ambiguous, it also fails. Record visual screenshots and educator approval. The right transfer is not simply an XML that parses; it is a caption track that allows the same scientific decision under a changed viewing environment.

Make the final review checklist a release decision

Before release, confirm that every intended cue has a target, every scientific claim matches the narration and image, and all target-language terminology has been reviewed for the audience. Confirm that the document declares the intended language, protected structure remains linked, all start and end times are sensible, and long lines have been rendered with the actual picture. Include accessibility sound cues and speaker identities where the format and brief require them.

Next, retain the XML parse result, profile-aware validation record, renderer environment, screenshots for boundary cases, defect log and final file checksum. Record any unresolved vendor-specific constraint as an explicit limitation. A sign-off should name the source video revision and the exact translated file; otherwise another exporter may distribute a different build under the same filename. The checklist’s purpose is to make a documented yes-or-no decision, not to convert incomplete testing into a decorative green badge.

Limits and closing principle

IMSC Text Profile 1.3 is a published technical Recommendation, but conformance does not prove translation accuracy, age appropriateness, broadcaster acceptance or universal font rendering. A valid caption can still teach the wrong scientific relationship. Conversely, a correct translation may fail in a particular delivery pipeline because its profile, styling or metadata does not meet that pipeline’s contract. Keep those questions separate so the responsible reviewer can address each one with the right evidence.

People translate quickly when they stop treating timed captions as a sequence of isolated sentences. They prepare a protected machine layer, define the educational propositions, maintain a cue ledger, translate with timing and layout in view, and test the rendered experience. That workflow produces speed through fewer preventable rework cycles. The governing question remains straightforward: can the target-language viewer understand the right thing, at the right moment, on the screen they will actually use?

Compare the audio-description boundary with ordinary caption text

Captions and audio description perform different accessibility jobs. A caption records dialogue and relevant sound information on the visual display, while an audio-description track describes visual facts for listeners who cannot see them. When a learner asks for both, do not squeeze an entire visual narrative into timed caption regions and assume the accessibility requirement is met. Maintain separate requirements and approval channels even if the same translator works on both. A small scientific video may have an equation on screen while a voice explains a different step; treating those two streams as interchangeable can create a subtle but consequential conflict for a learner.

A useful joint review asks which facts come from speech, which are visible, and which require an explicit description. Inspect overlaps with graphics and identify who owns any added explanatory wording. If a caption includes a bracketed non-speech description, verify that it corresponds to actual sound, not simply a visual impression invented by the editor. Check the hearing audience’s and deaf audience’s routes separately and record unresolved access gaps. One well-tested timed-text track can be necessary without being sufficient for complete media accessibility.

Frequently asked questions

Is IMSC 1.3 the same as a regular SRT file?

No. IMSC 1.3 is a defined TTML2 text profile with XML structure, styling and profile requirements; SRT is a different, simpler convention. Conversion requires separate validation.

Was IMSC Text Profile 1.3 already a Recommendation in 2025?

No. The W3C Recommendation date is 21 May 2026. Earlier drafts and candidate stages should not be represented as the final publication.

Can I translate timing attributes into another language?

No. Timing expressions are structural values. Any timing change needs media review and must remain syntactically correct for the chosen profile.

Does an XML parser guarantee a valid IMSC delivery?

No. Parsing only establishes XML well-formedness. Profile-aware conformance and actual player acceptance are separate checks.

Should a document’s `xml:lang` match its caption language?

Yes for the applicable language scope. Mixed-language content can need more detailed language annotation than one document-level tag.

What happens when a subtitle line becomes longer after translation?

Measure readability, then edit wording faithfully, adjust breaks or seek approved layout or timing changes. Never hide a meaning change behind a character-count target.

Can captions omit background sounds if the dialogue is correct?

It depends on the deliverable. Captions intended for viewers who cannot hear may need significant sound information beyond spoken dialogue.

Why do scientific symbols require special tests?

Fonts, superscripts, subscripts and script direction can alter the interpretation or visibility of formulas and units. Test the target renderer rather than trusting plain text.

Can a valid IMSC 1.2 text file be reused under 1.3?

The W3C describes compatibility for conforming earlier text-profile documents, but a particular distributor may still impose its own supported profile and ingestion rules.

Who signs off on a shortened subtitle?

The language reviewer and relevant subject or editorial owner should approve any compression that might affect meaning, while technical reviewers validate timing, XML and rendering.