For my project, the “Dreck Weg” app, I provide a free and clear waste calendar for regions including Aachen, Nuremberg, and Solingen. Currently, this web app offers its content in 13 languages, significantly improving usability for my users. The app’s texts are organized in structured JSON files – simple, clear, and efficient.

The Problem: As soon as I make even a small text change in the German version, I’m faced with a considerable effort. Ideally, all other language versions need the same update. Manual translations or using traditional translation tools cost time, introduce errors, and unnecessarily complicate my workflow.

The Idea: My goal was to simply pass the changed text blocks in the source language (German) to a small, automated script. This script would automatically translate these changes quickly and reliably into all the required languages.

Technical Implementation with Node.js and ChatGPT

To achieve this, I opted for a combination of Node.js and the OpenAI API. Simplified, the process is as follows:

  • Targeted prompt creation for ChatGPT: The script automatically creates precise translation requests for ChatGPT, clearly specifying context and language.
  • Communication with the OpenAI API: Node.js sends requests to the ChatGPT API and reliably receives translated texts in return.
  • Automatic updating of JSON language files: Translations for each language are returned automatically.

Here’s a simplified snippet of the Node.js code for illustration:

// install using "npm i openai"
import { OpenAI } from "openai";

// this is a list of target languages
const targetLanguages = [
  "ar",
  "el",
  "en",
  "es",
  "fr",
  "it",
  "nl",
  "pl",
  "sq",
  "tr",
  "zh",
];

/**
 * A Promise-version of reading data from STDIN (STandard INput).
 *
 * @returns {Promise<string>} The promise with the input data.
 */
function readInput() {
  return new Promise((resolve, reject) => {
    let input = "";

    process.stdin.once("error", (error) => {
      reject(error);
    });

    process.stdin.on("data", (chunk) => {
      input += String(chunk ?? "");
    });

    process.stdin.once("end", () => {
      resolve(input);
    });
  });
}

async function main() {
  // read the JSON from STDIN
  const inputJSON = await readInput();

  // get API key: https://help.openai.com/en/articles/4936850-where-do-i-find-my-openai-api-key
  const openai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });

  // tell GPT4 what to do
  const prompt = `I have the following JSON with german translations: ${JSON.stringify(
    inputJSON
  )}
You will translate all values to ISO languages ${targetLanguages.join()}.
Your JSON:`;

  const response = await openai.chat.completions.create({
    model: "gpt-4o-mini",  // https://platform.openai.com/docs/models/gpt-4o-mini
    
    // define the conversation we want to send
    messages: [
      {
        role: "user",
        content: prompt,
      },
    ],

    // we want structured JSON data as result, no chat conversation like a human
    response_format: {
      type: "json_schema",
      json_schema: {
        name: "JsonTranslations",
        description: "A JSON object with translations from the input.",

        // optional, but helpful: npm i @types/json-schema
        /** @type {import("json-schema").JSONSchema7} */
        schema: {
          type: "object",
          required: [...targetLanguages],
          properties: targetLanguages.reduce((prevVal, isoCode) => {
            return {
              ...prevVal,

              [isoCode]: {
                type: "object",
                description: `The input JSON in the exact structure but its values are translated to language with ISO code ${isoCode}.`,
              },
            };
          }, {}),
          additionalItems: true,
          additionalProperties: true,
        },
      },
    },

    // https://community.openai.com/t/cheat-sheet-mastering-temperature-and-top-p-in-chatgpt-api/172683
    temperature: 0,
  });

  // ensure having a valid JSON
  const responseJSON = JSON.parse(response.choices[0].message.content.trim());

  // output in pretty format
  console.log(JSON.stringify(responseJSON, null, 2));
}

// start the script
await main();

Real-world example

If, for example, I pass the translation

{
  "pickupReminder": "Denke daran, morgen wird dein Müll abgeholt!"
}

with a console execution like

cat deutsche-texte.json | node translate-json.mjs > uebersetzte-texte-als.json

It returns a result like:

{
  "ar": { "pickupReminder": "تذكر، سيتم جمع قمامتك غدًا!" },
  "el": { "pickupReminder": "Θυμήσου, αύριο θα μαζευτούν τα σκουπίδια σου!" },
  "en": { "pickupReminder": "Remember, your trash will be collected tomorrow!" },
  "es": { "pickupReminder": "Recuerda, ¡mañana se recogerá tu basura!" },
  "fr": { "pickupReminder": "N'oublie pas, tes ordures seront collectées demain !" },
  "it": { "pickupReminder": "Ricorda, la tua spazzatura verrà raccolta domani!" },
  "nl": { "pickupReminder": "Denk eraan, morgen wordt je afval opgehaald!" },
  "pl": { "pickupReminder": "Pamiętaj, jutro zostaną odebrane twoje śmieci!" },
  "sq": { "pickupReminder": "Kujtohu, nesër do të mblidhen mbeturinat e tua!" },
  "tr": { "pickupReminder": "Unutma, çöpün yarın toplanacak!" },
  "uk": { "pickupReminder": "Пам'ятай, завтра заберуть твоє сміття!" },
  "zh": { "pickupReminder": "请记住，明天将收集你的垃圾！" }
}

Benefits for me and my project

  • Significant time savings: Far less copying & pasting
  • Fewer errors: Uniform, consistent translations

What to consider?

Of course, there are limitations. The OpenAI API incurs costs, and translations should be checked randomly. Longer texts and specialized terms can also sometimes be challenging. These points should be considered.

Conclusion & Outlook

This automated translation approach saves me significant time, allowing me to focus more on the actual app development.

PS: Feel free to check out my free “Dreck Weg” app: https://dreck-weg.app