START HERE
Hotwords guide recognition toward important names. Translation terminology records approved equivalents. Forced replacements correct confirmed text patterns. Suppression rules remove explicitly designated content. They operate differently and should not become one ever-growing list.
Collect terms from approved material
Use the final agenda, product documentation and approved translations to identify names, organisations, models and technical terms. Record the source spelling, approved target term, context and owner. Explain the intended meaning of ambiguous abbreviations.
For example, AX-17 might be a recognition hotword and remain AX-17 in translation, while a business reviewer approves the wording for ‘delivery lead time’. These are illustrative examples, not preset rules or customer data.
Review the four rule types separately
Recognition hotwords do not guarantee accuracy: poor audio and similar pronunciations can still cause errors. Translation terminology also needs context because commercial, engineering and legal uses of a word may differ.
Forced replacement suits a confirmed fixed error, not unrestricted replacement of a short common word. Suppression needs a clear purpose and a preview of whether a word or an entire cue is affected. It should not hide source material that reviewers do not understand.
Treat access and versions as part of production
The video workflow uses glossary groups accessible to the member. Selecting a version does not authorise silent rewrites whenever somebody updates that group. Explicitly update the selection and review affected captions when a new version is needed.
If access is removed or a version is stale, confirm whether that group still belongs in the project. Do not copy another customer's names or confidential material into a public glossary. Assign an owner and record why rules change.
Test matches and near-matches
Positive examples confirm intended replacement, preservation or suppression. Negative examples test similar words, partial model numbers and ordinary sentences for unintended matches. Validate a short sample before processing a larger range.
Captions and dubbing should use the same approved translation. A silent replacement applied only to speech creates a mismatch between what viewers hear and read. After a rule change, check whether saved audio still matches the current text.
A practical comparison
| Rule | Purpose | Main risk |
|---|---|---|
| Hotword | Names and models during recognition | Unclear audio still causes errors |
| Terminology | Approved translation | Ambiguity without context |
| Forced replacement | Confirmed fixed text correction | Unintended short-word matches |
| Suppression | Explicitly excluded content | Loss of necessary information |
Before you sign off
- Each term has an approved source and language direction.
- Test intended and unintended matches, including numbers and models.
- Record the selected version and review scope before updating it.
- Check caption and speech consistency after changes.
Frequently asked questions
Is a larger glossary always better?
No. Irrelevant, conflicting or ambiguous rules increase maintenance and matching risks. Keep the list relevant to the project.
Will updating a glossary automatically change earlier subtitles?
A version change is not authorisation to rewrite existing results. Select the new version explicitly and reprocess the relevant content when needed.
Scope and related product information
Scope: current Wali desktop clients for Windows and macOS. Video subtitle production requires administrator enablement and configured recognition, dedicated translation and speech resources. Availability depends on the account and version. A mobile browser is not the local video export workspace.
Examples in this guide illustrate a review method, not a customer case or a performance result. Product availability changes with version and account configuration.
Video subtitle production and translated dubbing | Wali desktop →Current clients and installation guidance →