Skip to content

Knowledge that lights the world

Translating With AI: How to Use Google Translate, DeepL and Chat Assistants Well

Machine translation has taken a big leap in recent years and can produce fluent text in seconds. The strengths and weaknesses of translation tools, tips for better results and situations where you must use a human translator.

Translating With AI: How to Use Google Translate, DeepL and Chat Assistants Well

Machine translation improved markedly after the switch to neural network-based systems from 2016, and large language models now produce translations that better preserve style and context. Today you can understand a web page, email or menu in seconds. But there is still an important gap between “good enough to understand” and “ready to publish”.

Which tool for what?

  • Classic translation tools: Google Translate, DeepL and Microsoft Translator are fast, support many languages and translate documents.
  • Chat assistants: You can tell assistants such as ChatGPT, Claude or Gemini about style, audience and preferred terms, and ask them to explain the translation or offer alternatives.
  • Camera and voice translation: Reading signs and translating conversation are handy when travelling.

Tips for better results

  1. Give context: for example “This is a formal email to a client”.
  2. Add a glossary: List your brand’s or field’s special terms and their preferred translations.
  3. Use short, clear sentences: Long, ambiguous sentences increase errors.
  4. Back-translate: Translate the result back into the source language to check meaning is preserved.
  5. Have a native speaker read it: Especially for texts that will be published.

Common errors

  • Idioms and proverbs translated word for word
  • The wrong level of formality
  • Mixed-up number, date and unit formats
  • Information added that is not in the source, or a sentence left out

When is a human translator essential?

The cost of errors is high in contracts, court documents, medical reports, medicine leaflets, official applications and books or advertising copy for publication. For these, use a professional and, where required, sworn translator. Using machine translation as a draft corrected by an expert (post-editing) is also a common method.

How did machine translation develop?

Attempts at computer translation go back to the 1950s, but the methods have passed through several major stages:

  • Rule-based systems: Linguists wrote dictionaries and grammar rules one by one; the results were often crude and word-for-word.
  • Statistical translation: In the 2000s, systems began to learn, through probabilities, which phrase corresponds to which from millions of sentence pairs translated by people.
  • Neural machine translation: In 2016, Google announced its move to a neural network system that processes the sentence as a whole, and fluency improved markedly.
  • Transformers and large language models: The Transformer architecture introduced in 2017 forms the basis of today's translation tools and chat assistants.

Why are less widely spoken languages harder?

Translation systems need large amounts of bilingual text to learn. Quality is high for languages such as English, Spanish or French because such data is plentiful; for languages with little text on the internet, errors increase markedly. These are called “low-resource languages”. Research such as Meta's NLLB-200 project, announced in 2022, is trying to narrow this gap with models covering 200 languages. Even so, if you are translating an important text into a low-resource language, having the result checked by a native speaker is more important than ever.

AI tools’ features, plans and prices change often. Always check the tool’s official page for current information.

Sources and further reading

  1. Yonghui Wu et al., “Google’s Neural Machine Translation System”, arXiv, 2016.
  2. ISO 18587:2017, “Translation services — Post-editing of machine translation output”.
  3. European Commission, eTranslation service and machine translation guidance.
  4. Wu, Y. et al. (2016), “Google’s Neural Machine Translation System”, arXiv:1609.08144
  5. Vaswani, A. et al. (2017), “Attention Is All You Need”, NeurIPS
  6. NLLB Team (2022), “No Language Left Behind: Scaling Human-Centered Machine Translation”, arXiv:2207.04672
Share
Follow YulaHubGet the day's most important stories first.
RSS

Leave a comment

Your email address will not be published. Required fields are marked *