AI-103 Practice Question 29: Custom Translator Terminology
Microsoft Azure AI Apps and Agents Developer Associate (AI-103) · Test 3 · question 9 of 10
A life-sciences company translates regulatory submissions from English into German. The general Translator model is fluent but renders the company's controlled process and product terminology inconsistently, and every batch comes back from the reviewers with the same corrections. The company holds twenty years of submissions that were translated by humans and retained in both languages, document for document.
What should they do to get their terminology applied consistently to all future translations?
Show answer and explanation
Answer:
Custom Translator exists for this case: build customized neural machine translation systems "using parallel documents that understand the terminologies used in your own business and industry." It even accepts material that is parallel at the *document* level and matches sentences across the pairs automatically, which fits an archive of previously translated submissions. The published system is reached through the same Microsoft Translator Text API v3 the company already calls, and Microsoft notes BLEU gains of 5 to 10 points when the training data is suitable.
- A is wrong because it swaps a deterministic, trained system for a prompt that must be re-sent, re-paid for and re-trusted on every document, with no guarantee the glossary is honoured on the thousandth term. - B is wrong because the dynamic dictionary is a per-request override for one-off items such as a proper noun. It requires marking terms inline in the source of every request and does not scale to a whole controlled vocabulary. - D is wrong because Document Translation solves a different problem, translating whole files while preserving their layout. It still uses whichever model you point it at, so the terminology comes out exactly as inconsistent as before.
Reference: What is Custom Translator?