Implement text analysis solutions 4.1 Apply language model text analysis
A bank uses Azure Translator to translate customer letters into German and must produce a…
Study guide skill: Build solutions that translate text by using Azure Translator in Foundry Tools or LLM-powered translation flows
Question B2-025
A bank uses Azure Translator to translate customer letters into German and must produce a formal tone. Which approach supports this?
Select one answer.
Answer: D
The 2026-06-06 text translation API (GA) lets you choose standard neural machine translation or an LLM deployment, and LLM-based translation supports tone variants (formal, informal, neutral) and gender-specific output. LLM-based translation requires a Microsoft Foundry resource. The v3.0 NMT path doesn't offer tone controls, document translation offers general and custom models and glossaries rather than tone variants, and a phrase dictionary only forces specific terms.
Deep dive
How it works
Tone control is new in the 2026-06-06 text translation API, and it's tied to the model you pick. That GA release adds model choice: standard neural machine translation (NMT) or an LLM deployment, selected per scenario for quality, cost and requirements. It also adds tone and gender controls, but specifically for LLM-based translation, which supports tone variants (formal, informal, neutral) and gender-specific output.
For the bank's letters:
- Use a Microsoft Foundry resource. Using LLM-based translation requires one.
- Call the 2026-06-06 translate operation with an LLM deployment selected and German as the target.
- Request the formal tone variant.
Plan the upgrade carefully if you're coming from v3.0. The 2026-06-06 release includes breaking changes, including a revised request and response schema (inputs in the request, value in the response), so validate payloads and parsing before production. You can also steer LLM output style and terminology with up to five reference translations or an adaptive dataset index.
The other options, and when you'd use them
- Custom Translator with a phrase dictionary only: Custom Translator builds customized NMT systems that know your business terminology, and you can even train with only dictionary data. Terminology control isn't a tone setting.
- v3.0 with the standard NMT model and a profanity filter: Tone variants belong to LLM-based translation in 2026-06-06. A profanity filter handles profanity, not register.
- Document translation with the default model: Document translation is for translating whole documents, with features such as custom glossaries. The formal tone variant is described for LLM-based text translation.
Exam tip
"Formal or informal tone" or "gender-specific output" in Translator = 2026-06-06 API + LLM deployment (needs a Foundry resource). Terminology = dictionaries, glossaries or Custom Translator.
References
Last reviewed: