Good subtitles do more than put a sentence beneath a video. They need to convey the right meaning, appear at a comprehensible time, belong to the correct edit and remain readable without hiding something the viewer needs to see. The language can be accurate while the timing is late, or the timing correct while a missing negative reverses the message.
This guide explains how to prepare and review a short text track using an original fictional scene and dialogue. The example is not from a real film, and we claim neither user testing nor accessibility certification. The generated image shows a fictional screen with empty caption bars, not a finished usable track. This AI-assisted content has not received independent human editorial review.
Define the track's purpose before writing the first line
The W3C WAI guidance on captions and subtitles distinguishes translating dialogue from representing the audio information needed for understanding. Terminology differs across regions and products, so describe what the track provides rather than relying on its label.
Ask whether the task is to support someone who does not understand the dialogue's language, or to provide speech and important sound information to someone who cannot hear the audio. Those purposes can overlap, but a dialogue-only track does not automatically meet every need for watching without sound. This guide does not state legal compliance requirements or award a quality certificate.
Record the original language, target language, content type and intended audience. Decide whether speaker identification or meaningful sounds need inclusion. Do not fill the scene with irrelevant descriptions, but do not omit necessary information merely because it is not spoken dialogue.
Fix the video version and timing reference
Before starting, name the video version you are reviewing and record its duration. If an introduction is added or a passage later removed, the track may need resynchronising. “The same video” is insufficient when one version is shorter and another contains an introduction or additional scenes.
Keep an editable source file, a review version and a change record. Do not combine an old track with a new edit and randomly alter individual timings. Compare a clear point near the beginning, middle and end to establish whether the issue is a constant offset or a change inside the edit.
In a hypothetical exercise, dialogue begins at two seconds in the original version. Adding a one-second introduction before all existing content, with no other change, moves it to three seconds. This arithmetic follows a specific assumption; it does not automatically apply when parts of the edit or its speed change.
Make an understandable transcript before shortening it
Listen to the phrase in context and write what you can actually establish. Mark an unclear word for review instead of inventing it. In the working file, distinguish verified text from a suggestion needing confirmation, especially for names, numbers, negatives and directions.
In our fictional scene, someone says, “Leave the box here, not near the door.” Removing “not” or misplacing it changes the instruction. Translating “here” requires understanding the speaker's reference, not merely selecting a word that sounds natural outside the scene. This is independent dialogue, not a quotation from a copyrighted film.
An automated transcript may be a starting point when available and permitted, but a complete-looking screen of text does not establish that every word is correct. Review meaning against the sound and picture. Do not turn an ambiguous recording into artificial certainty in the final track.
Divide meaning before concentrating on line shape
Splitting “Leave the box here” from “not near the door” may work with one timing arrangement, but can separate the negative from its context confusingly if the second part arrives too late. Do not split merely to make line lengths equal. Consider what the viewer will understand at each moment.
Write the full sentence first and then test its division in the actual player. Avoid separating a noun from its qualifier or a negative from the phrase it negates when doing so changes interpretation. This is an editorial decision tied to language and context, not one universal character-count rule for every language.
When translating, do not match the original word count mechanically. Preserve the meaning and appropriate tone, and leave language assessment to someone able to evaluate it. An unreviewed translation does not become trustworthy because its timestamps match another version.
Calculate durations and gaps explicitly
The following table is an educational timing plan, not an existing recording. Times use minutes, seconds and thousandths of a second, not frame numbers. Durations and gaps illustrate arithmetic only; they are not general recommendations for reading speed or display length.
| Fictional cue | Start | End | Duration |
|---|---|---|---|
| Opening phrase | 00:02.000 | 00:04.500 | 2.5 seconds |
| Short reply | 00:05.000 | 00:07.000 | 2 seconds |
| Contextually important sound | 00:08.250 | 00:09.750 | 1.5 seconds |
The first gap is 5 − 4.5 = 0.5 seconds; the second is 8.25 − 7 = 1.25 seconds. Cue durations add to 2.5 + 2 + 1.5 = 6 seconds. The span from the first start to the last end is 9.75 − 2 = 7.75 seconds because it also includes the gaps. Do not confuse total display time with the entire timeline span.
If a cue begins at 6.5 while the previous one ends at 7, they overlap by 0.5 seconds. An overlap might be deliberate in some contexts, but its presentation needs checking in the player. Do not automatically classify every overlap as wrong, and do not ignore one without establishing what viewers will see.
Separate the track's text from its formatting and file
The MDN WebVTT documentation explains a timed-text format. Choose a format supported by the publishing destination and then check its requirements. Renaming a file extension does not change the internal structure.
Keep text and timing in a clear source, then export a version for the intended destination. Test it in the player your audience will use, not only in the editor. Arabic characters, punctuation or reading direction can appear differently when display settings or player support are unsuitable.
This article does not supply a complete subtitle file or ready-made implementation settings. Its purpose is to identify what needs reviewing before export is considered successful. Opening a file without an error does not prove correct meaning, timing or readability.
Review meaning, timing and viewing in separate passes
- Check words against audio, especially negatives, names, numbers and speakers.
- Check the start and end of each cue against the speech or action it represents.
- Watch continuously to notice text transitions, gaps and overcrowding.
- Check small and large displays, reading direction and whether text covers important imagery.
- Try the track without audio to find necessary information that has not been conveyed.
- Record what you could not verify instead of marking it approved.
Watching without sound can reveal problems, but it does not replace specialist review or experience with the intended audience. State the scope of what you did, such as “checked line display in this player,” instead of using a general phrase that implies a complete guarantee.
Keep language versions independent but clearly connected
Each language has text, review and timing that may need adjustment. Do not automatically copy line breaks if they make the new language unnatural. Preserve a cue identifier linking translations to the original meaning while allowing division to change when language requires it.
If a name or number changes in the source, record which versions are affected and review them again. Do not assume a correction in one file has reached every version. At handover, specify video version, language, review status and unresolved notes so an old file is not accidentally published.
Questions before approving a track
Is it enough that all cues end within the video duration?
That is only one technical check. Cues can lie within the duration yet appear with the wrong dialogue, obscure important imagery or convey an incorrect meaning. Combine timing validation with editorial review and actual viewing.
Can a long sentence be shortened by removing details?
More concise phrasing may be appropriate, but do not remove a negative, condition or attribution that changes meaning. Check the alternative in context and seek language review when you cannot establish accuracy. Quickly filling a file is not a measure of translation quality.
References and limits of the example
The linked W3C WAI and MDN pages were read on 2 November 2026. The dialogue, scene and timing table were independently created for this guide, with no reused film footage or user-study results. The examples provide neither accessibility certification nor a legal requirement or mandatory reading-speed standard.
