多語言提示詞技巧 — 用 Claude 處理全球語言

# 直接方式
You are a helpful assistant. Always respond in [TARGET_LANGUAGE].

# 更具體的方式
You are a helpful assistant operating in a [LANGUAGE] context.
- Use [LANGUAGE] for all responses
- Adapt to [LANGUAGE] cultural norms
- When translating, prioritize meaning over literal word choice
- When technical terms don't exist in [LANGUAGE], use established conventions

User says: "請幫我翻譯這段英文"

System should:
1. 檢測用戶說話的語言（中文）
2. 推斷目標語言（英文，基於「翻譯」前的語言）
3. 設定輸出語言為目標語言
4. 在 system prompt 中明確「respond in English」

Content: "The report from 市場調查 shows 很有希望的結果..."

Handling:
1. 保留原始語言分佈
2. 不強行統一
3. 在輸出時保持相同的混合比例
4. 明確標記每段落的語言

翻譯（逐字對應）：
English: "This feature is amazing!"
Literal translation to Japanese: この機能は素晴らしい！
✗ 自然度低，有點不自然

本地化（文化和上下文適應）：
English: "This feature is amazing!"
Localized for Japanese: この機能は本当に優れています
✓ 更自然，符合日文說話習慣

提示詞區別：
❌「Translate this to Japanese」
✓「Localize this to Japanese, adapting for cultural context」

術語詞彙表：

Technical | English | Simplified Chinese | Traditional Chinese | Japanese
-----------|---------|-------------------|-------------------|----------
API | API | 應用程序接口 | 應用程式介面 | API（アプリケーション・プログラミング・インターフェース）
Database | Database | 數據庫 | 資料庫 | データベース
Cache | Cache | 緩存 | 快取 | キャッシュ
Authentication | Authentication | 身份驗證 | 身份認證 | 認証

在提示詞中使用：
「When translating technical terms, use this terminology table:
[TABLE]

This ensures consistency across languages.」

System Prompt:
You are a customer support assistant. Based on the user's message:
1. Detect their language
2. Respond in that language
3. Maintain professionalism appropriate to that culture

Special considerations:
- For Chinese users: More formal than English, avoid directness
- For Japanese users: Use appropriate politeness level
- For European languages: Vary tone based on culture
- For Spanish: Regional differences (Spain vs Latin America)

User message (in Chinese): "你們的產品有問題，我很不滿意"
Response should:
- Acknowledge frustration respectfully
- Apologize genuinely (important in Chinese business culture)
- Offer concrete next steps
- Use appropriate formal tone

Task: Generate product description for [PRODUCT]
Target markets: US (English), Japan (日本語), Korea (한국어)

For each language:
{
  "language": "target_language",
  "description": "product description tailored to market",
  "keywords": "localized keywords",
  "tone": "culturally appropriate tone",
  "cultural_adaptation": "market-specific considerations"
}

Example for Japan:
{
  "language": "Japanese",
  "description": "このプロダクトは高品質で、信頼できるブランドです。",
  "keywords": ["高品質", "信頼性", "日本製"],
  "tone": "Polite, emphasizing quality and reliability",
  "cultural_adaptation": "Japanese customers value craftsmanship (職人技) and long-term reliability"
}

Task: Classify text and provide language-appropriate response

Schema:
{
  "original_language": "detected language",
  "category": "sentiment | topic | intent",
  "classification": "result",
  "confidence": "0-100",
  "language_specific_notes": "any language-specific context"
}

Example for Chinese text:
Input: "產品質量不錯，但價格太高了"
Output: {
  "original_language": "Simplified Chinese",
  "category": "sentiment",
  "classification": "mixed (positive product, negative price)",
  "confidence": 92,
  "language_specific_notes": "Chinese directly contrasts quality vs price (但 = but/however)"
}

Important: Don't normalize the meaning across languages.
Each language conveys nuance differently.

Task: Translate document to [TARGET_LANGUAGE], not just text

Configuration:
{
  "source_language": "English",
  "target_language": "Traditional Chinese (Taiwan)",
  "context": "business contract",
  "formality_level": "high",
  "regional_preferences": "Taiwan usage"
}

Special considerations:
- Legal terms must match Taiwan legal system
- Use Taiwan-specific formatting (date format, address format)
- Honorifics appropriate to business culture
- Currency: NTD (新台幣) not other Asian currencies

Output includes:
1. Translated content
2. Notes on culturally-specific adaptations
3. Terms that may need local legal review

Task: Validate data with language awareness

Example - Name validation:
{
  "name": "王小明",
  "detected_language": "Chinese",
  "validations": {
    "is_valid_format": true,
    "common_order": "Family name first (王) then given name (小明)",
    "cultural_context": "Chinese naming convention respected"
  }
}

Example - Date validation:
Input (Chinese): "2026年4月11日"
Output: {
  "date": "2026-04-11",
  "format": "Chinese format (year-month-day)",
  "converted_to_ISO": "2026-04-11",
  "note": "Chinese uses same year-first order as ISO, unlike English"
}

Task: Help user learn [TARGET_LANGUAGE]

Configuration:
{
  "user_native_language": "English",
  "learning_language": "Japanese",
  "level": "beginner",
  "focus": "grammar"
}

Response format:
{
  "explanation": "in English (user's native language)",
  "grammar_rule": "in simple language",
  "examples": [
    {
      "japanese": "私は学生です",
      "romaji": "watashi wa gakusei desu",
      "english": "I am a student",
      "grammar_note": "Topic marker は (wa) marks the subject"
    }
  ],
  "cultural_context": "Why this grammar matters in Japanese culture/business"
}

Task: Check content for cultural sensitivity across languages

Input: Marketing message in English
Target markets: China, Japan, Middle East, Latin America

Output:
{
  "message": "original English message",
  "cultural_analysis": {
    "china": {
      "issues": "none detected",
      "adaptations_needed": []
    },
    "japan": {
      "issues": "message too direct, not polite enough",
      "adaptations_needed": ["Add honorific language", "Soften assertions"]
    },
    "middle_east": {
      "issues": "potential religious sensitivity",
      "adaptations_needed": ["Review religious references"]
    },
    "latin_america": {
      "issues": "none detected",
      "adaptations_needed": []
    }
  },
  "overall_recommendation": "Adapt for Japan; localize Spanish version"
}

□ 明確指定目標語言（不要假設）
□ 區分方言和地區變種
□ 建立技術術語詞彙表
□ 確定文化適應需求
□ 提供語言特定範例
□ 測試輸出是否自然
□ 避免中間翻譯
□ 本地化（而非僅翻譯）內容
□ 考慮格式差異（日期、時間、地址）
□ 審查文化敏感性
□ 監控 Claude 輸出質量（按語言）