Liaisons and reduced conversational forms make names and negation worth a focused replay. In the target track, Choose concise written phrasing, keep names consistent, and do not infer a spoken regional variety from Han text alone. Both writing systems render left to right, but player layout still needs a visual pass.
Decision signals
- Source language: French (fr), written with Latin.
- Target language: Simplified Chinese (zh-Hans), written with Han.
- The workflow preserves cue IDs and timing first, then makes visible text and line-break changes under human review.
Prepare the French cue track
Avoid leaving articles, elisions, or short prepositions at the end of a subtitle line. Check French spacing around paired punctuation and keep apostrophes with the following word. Correct uncertain names, numbers, and speaker changes against the recording before asking a translation model to transform the text.
Shape readable Simplified Chinese subtitles
Choose concise written phrasing, keep names consistent, and do not infer a spoken regional variety from Han text alone. Chinese lines do not use spaces as word boundaries, so break by clause and meaning rather than character count alone. The source review policy was different: Maintain tu/vous choice, gender agreement, and the level of formality established by the speakers. Keep a stable mapping back to the source cue so reviewers can compare meaning without losing the original timing context.
Run bilingual timing and rendering QA
Use full-width punctuation consistently and inspect Latin abbreviations, numbers, and units at script boundaries. Both writing systems render left to right, but player layout still needs a visual pass. On the source side, Liaisons and reduced conversational forms make names and negation worth a focused replay. Watch the result with audio at normal speed, inspect every speaker change, and export only the reviewed track rather than an unexamined model response.
Boundaries to keep visible
- The page documents a French-to-Simplified Chinese review workflow; it is not an accuracy score or a promise that every configured model, account, or deployment accepts the pair.
- Keep source cue timestamps as the initial anchor, but permit a human subtitle editor to revise segmentation when target-language readability would otherwise fail.
Reviewed sources
- W3C WebVTT specification checked 2026-07-19
- Unicode bidirectional algorithm checked 2026-07-19
- French language and orthography editorial reference checked 2026-07-20
- Mandarin Chinese language and orthography editorial reference checked 2026-07-20