Sourced analysis · Documented facts are linked to their primary references. Interpretation is identified in the text.
Captions are not decorative type
W3C describes captions as synchronized text for speech and non-speech audio needed to understand content, while subtitles generally translate spoken audio into another language. Its WCAG explanation adds that captions should not obscure relevant video information. [W3C Web Accessibility Initiative] [W3C Web Accessibility Initiative] In portrait drama, that is a picture decision as much as a language decision.
A subtitle pass must therefore ask two questions at once: is the wording accurate, and can the viewer still see the clue, face, text message or hand action that carries the scene? Passing the transcript against audio alone does not answer the second question.
Build a three-pass QC
Pass one is meaning: check names, negations, speaker changes, deliberate pauses and meaningful sounds. Pass two is timing: scrub at normal speed, then at the chapter cut, to ensure a line does not finish after the reaction it explains. Pass three is frame collision: test default placement against lower-third interfaces, on-screen messages, hands and any plot-critical object.
W3C says automatic captions are not sufficient unless confirmed fully accurate and notes that they usually need significant editing. [W3C Web Accessibility Initiative] Use machine transcription as an input if appropriate, never as evidence that the audience has received the scene correctly. Platform-specific requirements and player behaviour must be verified separately for the intended release.
Worked fictional exercise: the lower-third evidence
Original exercise: In a 9:16 confrontation, detective Arin says, “I never saw the receipt.” At the same moment, the lower third shows a receipt timestamp that proves she did. A default two-line subtitle covers the timestamp. The QC record marks the line as plot-critical and moves it upward only for this cue, keeping enough contrast and a safe reading interval.
A later cue reads “[phone vibrates]” before Arin reacts. It stays because the sound motivates her glance; a caption of generic room tone does not. The team exports a review file, watches it on a phone-sized preview with interface overlays where available, and logs the decision by episode and timecode. This exercise is fictional.
Make accessibility a delivery gate
Keep a caption master, subtitle-language versions and a change log alongside picture lock. When dialogue changes, identify every affected cue rather than making a silent replacement. Check an opening, a high-density argument, a phone screen, an action beat and the final cliffhanger in each episode; these are likely collision points, not a guarantee of complete coverage.
This workflow does not certify legal compliance or any specific platform delivery. It establishes a responsible editorial question: can a viewer read what matters without losing what they need to see? That question belongs before release, when a fix is still cheap.
Sources & evidence
W3C Web Accessibility Initiative · Accessibility guidance
Source date: 17 Sept 2024 · Checked: 19 Sept 2026
- W3C distinguishes same-language captions from translated subtitles.
- It says automatically generated captions need confirmation for full accuracy and usually need significant editing.
- It identifies WebVTT as a common web caption format.
W3C Web Accessibility Initiative · Accessibility guidance
Source date: Not stated by source · Checked: 19 Sept 2026
- The WCAG understanding document explains that captions cover dialogue and meaningful non-speech audio.
- It notes captions should not obscure relevant video information.