A video contains several possible text sources: speech, visible lettering and supplied subtitles. Translation can use the wrong one for your goal even when the result is fluent. Alibaba IMS documents ASR, OCR and supplied-subtitle routes, including combined modes. Choose the source deliberately instead of treating every readable sentence as dialogue.
Inventory what the audience can hear and see
Keep a short sample containing the conflict. Transcribe the actual spoken sentence, copy the visible text and identify any approved subtitle file. Note whether the picture is a sign, a slide, a quotation or an existing burned-in caption. Visible lettering can be relevant information without being words spoken by the person.
For a hypothetical launch clip, a poster says “Opening next month” while the host says “Registration closes Friday.” Translating the poster as the host’s dialogue changes the message. Translating only speech may also omit a required date shown elsewhere. The solution is an editorial source decision, not merely a better font or a longer translation prompt.
Choose an authority for each content role
Read the selected service’s current source-mode definitions. Do not assume combined OCR and ASR means a human has reconciled conflicting evidence. If an approved subtitle file is the authority, confirm how the product accepts it and whether its content and timing match the supplied video.
| Content | Intended role | Source decision |
|---|---|---|
| Host’s sentence | Spoken dialogue | Verify from actual audio |
| Poster date | On-screen information | Translate separately if required |
| Existing caption | Possible transcription | Compare with audio before trusting |
| Approved external subtitles | Reviewed language source | Confirm version and timing |
This original decision table does not claim all modes support every language or that mixed-source recognition resolves contradictions correctly.
Review the intermediate text before rendering
Where the workflow provides recognized or translated text, compare it with the selected authority. Mark the exact point where poster wording entered the dialogue. Correct the source route or reviewed text through supported controls; do not patch only the final visible line while leaving a wrong dubbed sentence in the soundtrack.
If no intermediate evidence is available, inspect the actual result with a qualified language reviewer and preserve the uncertainty about source selection. Avoid claiming that OCR or ASR definitely caused an error without task evidence. A wrong translation can also originate after correct recognition.
Deliver both information layers clearly
Review dialogue audio, dialogue captions and translated on-screen information separately in the final file. Ensure a replaced poster does not contradict the spoken announcement. Check that decorative words are not accidentally promoted into factual speech and that required visual information has not disappeared.
Keep the source decision and approved text with the deliverable for future language versions. In an Imagild film these may be externally prepared assets; this article does not claim that Imagild exposes IMS source modes. Its pass condition is a translation whose spoken and visible information each follow an explicit, reviewed authority.