ChatGPT in different languages

Is OpenAI’s ChatGPT good for translating among all languages?

The ChatGPT user interface supports 50+ languages. But for everyday use and localization tasks, it can handle many more, as it was trained on large multilingual datasets. The more widely used a language and the more content exists in it, the more training data is available to ChatGPT or other LLMs — and the more reliable and natural the output quality.

Based on our research and experience working with multilingual content, here’s a rough breakdown of localization quality by language group:

• Highest and most predictable translation quality: West European (Spanish, French, German, Italian, Portuguese), East Asian (Mandarin Chinese, Japanese, Korean), and Middle Eastern (Arabic, Hebrew)
• Generally good results, but less consistent: Central European (Polish, Czech, Slovak, Hungarian, Romanian, Ukrainian), Balkan (Serbian, Croatian, Bulgarian), Baltic (Lithuanian, Latvian, Estonian), and Caucasian (Georgian, Armenian, Azerbaijani)
• More variability in output, particularly for specialized content: Less common and rare languages with limited training data

You can test ChatGPT’s translation ability across languages directly on your bilingual files in our Hera AI software. For less common language pairs, using English as a pivot language can improve consistency.