AI Translation
AI can now translate between over 100 languages almost instantly. That is pretty remarkable — but how does it actually work, and does it always get it right?
Learning from millions of examples
Translation AI was trained on enormous amounts of text that has been translated by humans — books, websites, legal documents, subtitles. By studying millions of examples of the same content in different languages, the AI learned the patterns that connect them.
It does not work word by word
Early translation programs tried to swap each word for its translation. This produced funny results because languages work very differently. "I am cold" in French is J ai froid — which literally means "I have cold"!
Modern AI understands whole sentences and even paragraphs at once, which produces much more natural results.
Where it still struggles
- Idioms — phrases like "raining cats and dogs" do not translate literally
- Humour — jokes often rely on wordplay that does not exist in other languages
- Cultural context — some ideas just do not exist in other cultures
- Less common languages — AI works much better for widely spoken languages than rare ones
Did you know?
Te Reo Maori translation AI is still developing. Because there is less text available in te reo than in English, the AI has fewer examples to learn from — which makes it less accurate.