Names, brands, jargon and acronyms: getting specialist terms right in translation

Four kinds of special term, four different rules

Most terminology problems come from treating every unusual word the same way. A quick classification settles most cases:

People's names

Keep as written in Latin-script languages; transliterate into other scripts using the established spelling if one exists.

Places and organizations

Use the target language's exonym where one is established (Munich is München, Lisbon is Lisboa); otherwise keep the original.

Brand and product names

Do not translate. Where the product's own interface is localized, use the localized name the viewer will actually see.

Jargon and acronyms

Follow the target audience's real usage: sometimes a native term, sometimes the English loanword, sometimes a localized acronym.

The rest of this article goes through each, then covers the glossary that ties them together. Mistakes in this area are among the common machine translation errors that reviewers find most often.

People's names and places: keep, transliterate or use the exonym

A person's name is not a phrase to translate. "Rob Baker" stays "Rob Baker" in French or German; translating it as "Herr Bäcker" is an error, though machine translation does it occasionally when a surname is also a common word.

Between scripts, names are transliterated, and many famous names have conventional spellings. Shakespeare is Шекспир in Russian, シェイクスピア in Japanese katakana, शेक्सपियर in Hindi, شكسبير in Arabic and 莎士比亚 in Chinese, where characters are chosen for their sound. For a lesser-known name there may be no convention, and several plausible spellings exist; consistency then matters more than the specific choice.

Places work differently because many have exonyms. Cologne is Köln in German, Venice is Venezia in Italian, Beijing is 北京 in Chinese. Translators use the established exonym and keep the original name when none exists. Literal translation of place names ("Salt Lake City" rendered word for word) is almost always wrong.

Brand and product names: the do-not-translate list

Company names, product names and trademarks stay as they are: Slack, Figma and Notion are not translated, even though each is also an ordinary English word. Engines usually get famous brands right and lesser-known ones wrong, especially when the name is a common noun. A short do-not-translate list is the cheapest protection.

There is one important exception. Where a product's own interface is localized, viewers see the localized names on their screen, and the translation should match them. Google Sheets appears as "Google Tabellen" in the German interface, and the "File" menu in many apps appears as "Datei". For a software tutorial, the right translation of a menu or button name is whatever the localized interface shows, not a fresh translation of the English label. Check the product in the target language before deciding.

Jargon: when the English term is the right translation

In many technical fields the English term is the one practitioners use. German developers talk about a "Pull Request" and a "Deployment"; Japanese engineers write コンテナ (kontena, container) as a loanword. Translating these into native coinages can make content sound odd to the very audience it is for.

Other fields and markets go the opposite way. Medicine and law have established native terms almost everywhere, and France and Quebec have official bodies that recommend French equivalents, such as "courriel" for email. The right answer depends on who is listening: an internal training video for engineers can keep "API gateway", while consumer help content may need the native term. Decide per audience and record the decision.

Acronyms and initialisms

Some acronyms are international and stay unchanged: API, CPU, USB, PDF. Others have established local forms, and using the English one looks like an error:

• GDPR is DSGVO in German and RGPD in French, Spanish, Italian and Portuguese.
• NATO is OTAN in French, Spanish and Portuguese; the UN is ONU in French, Spanish, Italian and Portuguese.
• AIDS is SIDA in French and Spanish.

Company-internal and product acronyms almost never have local forms. If your audience might not know one, spell it out on first use in the source; the translation will then carry the expansion too.

Transliteration between scripts

Moving a name from one script to another involves choices that machine translation makes inconsistently. Russian names have several Latin spellings (Tchaikovsky, Chaikovsky, Tschaikowski in German). Arabic and Hebrew script usually omit short vowels, so a name written in them can be romanized several ways. Japanese renders foreign names in katakana by sound, and there is often more than one accepted form.

For your own product, team or speakers, choose one spelling per target script and add it to the glossary. If the machine translation spells the same name two ways in one video, viewers will assume two different people.

Building a glossary and reviewing terms with the transcripts

A glossary does not have to be elaborate. A spreadsheet with source term, decision (keep, localize, transliterate) and the approved target form is enough. Then use it to review:

1. Before translating, list every name, product, acronym and specialist term in the video. The source transcript is the fastest way to collect them.
2. Decide the treatment and approved target form for each, checking localized product interfaces and audience usage.
3. After translating, search the source-language transcript for each term to confirm the speech recognizer heard it correctly.
4. Search the translated transcript for each approved target form and note any line where the term appears differently or is missing.
5. Correct those lines in the translated transcript and the SRT or VTT file, and keep the glossary for the next video in the series.

Limits: what a glossary can't fix

A glossary makes review fast, but it has limits:

• In languages with rich inflection, such as Finnish, Polish or Russian, a term changes form with its grammatical role, so a search for one exact form misses the others. Search for the stem.
• Some terms are ambiguous even within a field and need a human decision line by line.
• On-screen text in the video is not part of the speech, so a glossary applied to transcripts does not touch labels or slides in the picture.
• If a wrong term was spoken in a dubbed voice track, correcting the transcript fixes the subtitles and documents, not the audio that was already generated.

Terms and names in mydubly

In mydubly, terms pass through two machines. Whisper large-v3-turbo transcribes the speech, and unfamiliar names or jargon may come out as ordinary words that sound similar ("Postgres" as "post grass"). A neural machine translation engine then translates that text, so a recognition error becomes a translation error with no warning. That is why step 3 above checks the source transcript first.

Every job gives you transcripts in both languages, SRT and VTT subtitles, and optionally a translated voice, so the review workflow above uses files you already have. The spoken language is detected automatically; for scripts, Chinese output is Simplified Chinese, Arabic output is Modern Standard Arabic and Norwegian output is Bokmål, which affects how names should be written. Because the voice reads the translated text, check terms before publishing the dubbed version. For recordings with heavy jargon, clear pronunciation helps recognition; see the tips in improving transcription accuracy.

Next step

Take one terminology-heavy video, translate it with the video translator, and build a twelve-to-twenty-term glossary while you review the two transcripts. Reuse it for every later video in the series. If your content is a course or training library, the training video translation page covers that workflow end to end.