Over the past year, machine translation has made significant progress. Tech giants like Google, Microsoft, and Facebook are building powerful Neural Machine Translation (NMT) systems modeled after the human brain.
NMT offers improved accuracy compared to older machine translation systems. To hear the headlines tell it, that means all of our translation problems are solved. Who needs a human translator when you’ve got artificial intelligence?
But if that were true, we wouldn’t have these amusing translation mistakes to share with you, would we?
Google Can’t Translate the South African Parliament
Voice-to-text translation sounds amazing, in theory. Just speak into a microphone, wait a few seconds, and “Voila!” The system automatically translates your words, and you don’t even have to lift a finger.
And it is amazing when it works. But therein lies the rub. Regional accents and dialects can throw these systems off.
For example, the video above shows what happens when Google tries to translate a South African Parliament session.