The biggest problem with machine translation is "every word is right, but the sentence isn't human." The problem isn't the model-it's the instruction. Say only "translate" and it gives you word-for-word literal output. This prompt pack goes from literal to elegant to localized, in three levels. The key is giving the model a "translator persona + glossary + target audience."
Beginner: Faithful Literal Translation
Just starting out-first get it accurate.
You are a professional translator. Translate the following text from {{source_language}} to {{target_language}}.
Requirements:
1. Faithful to the original-do not add or drop meaning
2. For technical terms, keep the original in parentheses, e.g. AI (AI)
3. Output pure translation, no explanation
Text: {{text}}Intermediate: Elegance + Style
Demand fluency and idiomatic target-language phrasing.
You are a senior translator fluent in {{source_language}} and {{target_language}}. Translate the following text to "faithful, expressive, elegant" (信达雅):
1. Faithful: true to the original, no distortion
2. Expressive: fluent, following {{target_language}} conventions; no word-for-word rigidity
3. Elegant: wording refined and fitting the genre ({{genre, e.g. tech/literature/marketing}})
Style reference: {{a sample passage in the target language}}
Glossary (must follow): {{term: translation}}
Text: {{text}}Expert: Localization (Beyond Translation)
Localize for a specific market, accounting for culture, audience, and compliance.
You are a localization expert. Rewrite the following content for {{target_market}} (e.g. mainland China / US / Japan) aimed at {{audience}}-not just translate:
1. Cultural adaptation: swap examples, metaphors, units, and date formats to the target market's conventions (e.g. USD -> RMB, feet -> meters)
2. Audience adaptation: use language the {{audience}} understands-de-jargon for beginners, keep terms for experts
3. Compliance adaptation: avoid sensitive content in the target market (e.g. medical/political phrasing in certain markets)
4. Preserve original structure and information density; don't pad
Glossary: {{term: translation}}
Original: {{text}}
Output: the localized text + a brief note (≤100 words) on which cultural/audience points you changedHow to Use
Beginner for accuracy, intermediate for fluency, expert for "landing." Most everyday use is fine at intermediate; for product launches abroad, marketing copy, or doc localization, go expert. The key is providing a glossary-without one, the model translates the same term differently each time, and inconsistent terminology is translation's biggest failure.