When I was composing my content page regarding my Master’s degree study regarding the neural effect of reading awkward machine translations, I tried to find some examples showing that even the neural-network-powered Google translate still makes funny mistakes.
But alas, that was hard!
Neural networks are just too good at finding statistical relationships between words either within or across languages. And, with the help of contributors around the world to improve the translations, funny mistakes are much rarer than prior to 2016.
To my surprise, I found one just two days later.
My girlfriend (henceforth “JK” for short) asked me how to translate this common Chinese phrase “對事不對人(duiˋ shiˋ buˊ duiˋ renˊ)”. This phrase is often used in the context of a heated debate. People say this phrase to express that they are discussing the issue in a matter-of-fact fashion, concerning themselves with the facts and not the individuals.
A word-by-word translation of this phrase “對事不對人” would be “concerning the thing, not concerning the person“. But Google Translate seems not to have learned the context in which this phrase is often uttered, thus unable to figure out “對” is a preposition rather than an adjective. Instead, it fetched the most common translation of “對”, which is the adjective “correct” or “right”, leading to this hilarious example.
Agnostic of the context, Google Translate doesn’t know why people say this phrase. Not knowing why people say this phrase, it goes with the statistically most prominent association of individual words. There’s still quite some way to go towards a satisfactory and really intelligent machine translation service.
