DeepL, for example, fails hilariously with certain inputs. I once got it to mistranslate things like these:
- "孔明" (Kongming, the genius strategist/statesman from China's Three Kingdoms period) got translated as "Confucius"
- "chage and aska" (a Japanese pop music duo) got translated as "曹操" (Cao Cao, founder of the Wei state during China's Three Kingdoms period) or even "ちんちん" (literally "penis" in Japanese)
AI finally reached the level of a lazy high school history student who didn't study enough.
IIRC at that point it would translate e.g. "23 万" correctly, but 23万 would go to 23 million.
I tried reporting the error, no idea if they actually pay attention to those!
> Machine translation has a loooong way to go.
Since large numbers on e.g. streaming sites are often reported in multiples of 万 in Chinese, but a million in English, I did wonder if the machine "learned" that they were equivalent.