Why AI still can’t translate South African languages in 2026 We have all been there. You put a perfectly normal sentence into a popular translation tool and what comes out is absolute gibberish in isiXhosa or Sesotho. It is often funny but for those of us working in professional translation but it is also a bit alarming. While the world is obsessed with how "smart" AI has become, South Africa’s 11 official written languages remain a massive blind spot.  1. The data is thin and dusty AI is only as good as its library. For English, the AI has read almost everything ever written. For languages like Tshivenda or Xitsonga, the digital library is tiny. Most of the data the AI "learns" from comes from old religious texts or formal government documents. This means the AI ends up sounding like a 1920s preacher or a bored civil servant rather than a real person living in 2026. 2. Confident lying and "borrowed" words One of the biggest issues our team of first langu...