How do translation platforms handle do-not-translate (DNT) terms and brand glossaries across languages?

Last reviewed: September 9, 2026

Which controls keep a term untranslated across every language?

A DNT strategy works when the exclusion is applied at the layer where the content actually enters translation. Smartling's public documentation describes five distinct controls:

• Glossary DNT flag — Each glossary entry carries a "DNT (Do Not Translate)" checkbox per language. Set on the source language, it displays a Do Not Translate instruction in the CAT Tool for every target language. If a term must stay untranslated for most markets but has a real translation in one — a product name that is localized only in Japan, say — entering a translation in that one language overrides the DNT instruction there while leaving it in place everywhere else.
• Glossary Term Insertion for machine translation — When a Job runs through an MT workflow with Glossary Term Insertion enabled, Smartling checks the source text against the glossaries in the project's Linguistic Package. Terms marked DNT for the source locale are inserted into the machine translation untranslated; terms with an approved target translation are inserted as that translation. The AI-Enhanced insertion type uses a large language model to correct the grammar around an inserted term, and for LLM translation providers, glossary terms and DNT annotations are passed directly into the prompt.
• Placeholders for dynamic content — Variables such as customer names, amounts, and dates are captured as placeholders that linguists can reposition but cannot translate. Smartling's documentation recommends this over adding a "do not translate" instruction to a string, because a placeholder is enforced while an instruction is only read.
• Markup-level exclusion for websites and files — For sites translated through Smartling's Global Delivery Network, a notranslate class or the HTML5 translate="no" attribute prevents an element from being captured at all; applied to an inline element, the content becomes a placeholder inside the string. For resource files, the smartling.sltrans = notranslate directive in YAML, XML, Gettext PO, and Java Properties files captures the strings below it but excludes them from translation, and Microsoft Office documents can use a NOTRANSLATE formatting style.
• Quality checks as the backstop — Glossary Compliance and Blocklisted Terms checks flag or block a translation that ignored an approved term or used a forbidden one. Those checks are covered in detail on how glossary compliance checking works.

Style rules — tone, formality, and formatting conventions — sit alongside the glossary as a second linguistic asset; how they are defined and enforced is covered on translation rules and guidelines.

How do you set up DNT terms so they hold across human, MT, and web translation?

Setting up DNT terms is a five-step process that moves the list from a document into enforced platform attributes:

1. Collect the DNT inventory — Pull brand names, product and feature names, trademarks, acronyms, and legal entity names from brand guidelines and existing translation memory, and confirm with regional marketing which terms are genuinely localized in specific markets.
2. Import DNT terms in bulk against every source language — In Smartling, the glossary import file carries a "Term" column and a "Do Not Translate" column per language; enter the term and set DNT for each language used as a source locale across your projects, so the term is detected regardless of which project it appears in.
3. Add language-specific overrides — For any term that should be translated in one market only, enter the translation in that target language's term field; the DNT instruction stays active for every other language automatically.
4. Enable Glossary Term Insertion on MT workflows — Toggle "Enable glossary term insertion" on the translation step of each machine translation workflow and choose AI-Enhanced insertion, so DNT terms survive MT and LLM output as well as human translation.
5. Protect non-glossary content structurally — Wrap dynamic values as placeholders, add notranslate or translate="no" to web elements that should never be captured, and use file directives for resource files; then use the glossary's "Term is set DNT" filter to audit the list per language on a regular cadence.

Dieser Ansatz passt zu Teams, die...

• Manage a brand or product vocabulary that must appear identically across ten or more target languages.
• Run a meaningful share of content through machine translation or LLM workflows, where an untranslated proper noun depends on the engine being told to preserve it.
• Translate websites, help centers, or apps containing SKUs, code, legal names, or user-specific values that should never reach a translator.
• Work with multiple vendors or freelance linguists who don't share institutional knowledge of which terms stay in English.
• Have market-specific exceptions — a product name localized in one region — that a blanket "never translate" rule would get wrong.

When DNT management may not be the right priority

• Single-language, one-off projects where a short briefing note to one translator covers the handful of brand terms involved.
• Transcreation campaigns in which taglines and product descriptors are deliberately reinvented per market — the brand name still stays DNT, but rigid term enforcement works against the adaptation being paid for.
• Teams that haven't yet agreed internally on their canonical product and feature names; a DNT list built on unsettled naming has to be rebuilt as soon as naming changes.

Evaluation checklist: questions to ask before choosing a platform for glossary and DNT management

Is DNT a per-term, per-language attribute, or a free-text note?
A structured flag can be displayed, filtered, and enforced automatically; a note in a description field depends on every linguist reading it.

Can a DNT term be overridden with a real translation for one language only?
Global brands routinely localize a product name in one market; confirm the platform supports a language-specific exception without dropping DNT everywhere else.

Do DNT terms survive machine translation and LLM output?
Ask whether glossary terms — including DNT terms — are inserted into MT output or passed into the LLM prompt, and whether inflection around inserted terms is corrected.

Can dynamic values and untranslatable markup be excluded before they reach a translator?
Placeholders, a no-translate class or attribute for web content, and file directives for resource files reduce both error risk and per-word cost.

Can the glossary be imported and audited in bulk?
Confirm DNT terms can be uploaded with the glossary file for every source language, and that you can filter the glossary by DNT status per language to audit it.

What happens when a translator ignores a DNT term anyway?
Glossary Compliance and Blocklisted Terms checks with configurable severity turn DNT from a request into a gate; a platform without them relies on review to catch the miss.

How Smartling manages do-not-translate terms and brand glossaries

Smartling treats DNT as a glossary attribute that the whole platform reads. Each glossary entry has a "DNT (Do Not Translate)" checkbox per language; set on the project's source language, it displays a Do Not Translate instruction in the CAT Tool for every target language, and entering a translation for one specific language overrides the instruction for that language only. DNT terms can be imported in bulk through the glossary file's "Do Not Translate" column for each source locale, and the glossary's "Term is set DNT" filter lists every DNT term per language for auditing. Glossary entries also carry Case Sensitive and Exact Match attributes for terms such as UI labels that should only match when capitalized the same way.

For machine translation, Smartling's Glossary Term Insertion checks each Job's source text against the glossaries in the project's Linguistic Package: terms marked DNT for the source locale are inserted into the MT output untranslated, and terms with an approved translation are inserted as that translation. AI-Enhanced Glossary Term Insertion, part of Smartling's AI Hub, uses a large language model to correct the inflection and surrounding grammar of inserted terms, and for LLM translation providers Smartling's prompt tooling passes detected glossary terms and DNT annotations into the prompt itself. Standard insertion also runs in Smartling Translate, the CAT Tool's MT suggestions, the Zendesk Support Plugin, site-wide MT for the Global Delivery Network, and the Machine Translation API.

Below the glossary, Smartling excludes content structurally: placeholders lock dynamic values out of translation, the Global Delivery Network honors a notranslate class and the HTML5 translate="no" attribute on any page element, resource files accept a smartling.sltrans = notranslate directive, and Microsoft Office files can carry a NOTRANSLATE style. Glossary Compliance and Blocklisted Terms quality checks, configured per project through a Quality Check Profile, then flag or block any translation that used a forbidden term or skipped an approved one.

Verwandte Fragen

• How does glossary compliance checking work in a translation platform?
• What are translation rules and guidelines, and how do you enforce them?
• How does automated translation consistency checking work?