Nordlicht Reisen GmbH has customers in three languages (R-01), but the editorial team's destination information exists only in German and English. The travel assistant is supposed to be able to answer in Danish (R-05). This lesson decides how the destination information gets into Danish: with Azure Translator in Foundry Tools or with a language model.
Prior knowledge. The row "destination information into Danish" of the decision between model and Foundry Tool is in lesson 1.1 (decision 2). Fencing in foreign text as data: "AI track, lesson T6.2". A counter at every model call: "AI track, lesson T2.4" (R-11). None of that is repeated.
New in this lesson. What is translated are texts of the editorial team, not customer texts. The company name "Nordlicht Reisen" and the place names in the destination information stay untranslated. The terms of travel (R-03) apply in Danish only after approval by a person (R-02).
Paper, account, costs. Paper: decision, exercises. An Azure account is needed only for the optional run. Cost drivers: characters of the source (Translator with NMT), tokens (Translator with LLM, language model); pricing pages instead of amounts.
The requirement
All numbers are assumptions of the task.
| No. | Requirement | Type | Measure | Weight | Source |
|---|---|---|---|---|---|
| 4.2-01 | Destination information from German and English into Danish. | F | 2 language pairs | Must | R-01, R-05 |
| 4.2-02 | The editorial team rarely needs to intervene on a sample. | Q | at least 90% of 40 paragraphs without change | Should | R-10 |
| 4.2-03 | Company and place names stay unchanged. | R | 0 deviations in the test set | Must | Editorial team |
| 4.2-04 | Call without a key. | R | 0 keys in code and configuration | Must | R-07 |
| 4.2-05 | Data stay in the EU. | R | processing location evidenced | Must | R-06 |
| 4.2-06 | The cost per quarterly update can be estimated in advance. | Q | characters or tokens per run | Should | R-09 |