Batch Translating with Gemini AI

Batch translation is more efficient than individual API calls for multiple texts.

Approach

Combine multiple texts into a single prompt to reduce API overhead.

Prompt Structure

const prompt = `Translate the following Japanese news headlines to English.
Return ONLY a JSON array of objects with "original" and "translated" fields.
Do not include any explanations or formatting outside the JSON.

Headlines to translate:
${headlines.map((h, i) => `${i + 1}. ${h}`).join('\n')}

Expected output format:
[
  {"original": "Japanese headline 1", "translated": "English translation 1"},
  {"original": "Japanese headline 2", "translated": "English translation 2"}
]
`;

Complete Implementation

async function batchTranslate(headlines: string[]): Promise<Array<{original: string, translated: string}>> {
  const prompt = `Translate the following Japanese news headlines to English.
Return ONLY a JSON array with "original" and "translated" fields.

${headlines.map((h, i) => `${i + 1}. ${h}`).join('\n')}
`;

  const response = await model.generateContent(prompt);
  const jsonText = response.response.text();

  // Parse JSON (handles markdown code blocks)
  const cleaned = jsonText.replace(/```(?:json)?\s*([\s\S]*?)\s*```/, '$1');
  return JSON.parse(cleaned);
}

Token Optimization

• Group related translations together
• Use numbered lists for clear mapping
• Request JSON output for easy parsing
• Limit batch size to ~50 items per request

Error Handling

try {
  const result = await batchTranslate(headlines);
  return result;
} catch (error) {
  // Fallback: translate individually
  return Promise.all(headlines.map(h => translateOne(h)));
}