Nov 062021

Machine translation is hard. To accurately translate text from one language to another, context is essential.

Today, I tried a simple example: an attempt to translate two English sentences into may native Hungarian. The English text reads:

An alligator almost clipped his heels. He used an alligator clip to secure his pants.

See what I did here? Alligators and clips in different contexts. So let’s see how Google manages the translation:

Egy aligátor majdnem levágta a sarkát. Aligátorcsipesz segítségével rögzítette a nadrágját.

Translated verbatim back into English, this version says, “An alligator almost cut off his heels. With the help of an ‘alligatorclip’, he secured his pants.

I put ‘alligatorclip‘ into quotes because the word (“aligátorcsipesz“) does not exist in Hungarian. Google translated the phrase literally, and it failed.

How about Microsoft’s famed Bing translator?

Egy aligátor majdnem levágta a sarkát. Aligátor klipet használt, hogy biztosítsa a nadrágját.

The first sentence is the same, but the second is much worse: Bing fails to translate “clip” and uses the wrong translation of “secure” (here the intended meaning is fasten or tighten, as opposed to guarding from danger or making safe, which is what Bing’s Hungarian version means).

But then, I also tried the DeepL translator, advertising itself as the world’s most accurate translator. Their version:

Egy aligátor majdnem elkapta a sarkát. A nadrágját egy krokodilcsipesszel rögzítette.

And that’s. Just. Perfect. For the first sentence, the translator understood the intended meaning instead of literally translating “clip” using the wrong choice of verb. As for the second sentence, the translator was aware that an alligator clip is actually a “crocodile clip” in Hungarian and translated it correctly.

And it does make me seriously wonder. If machines are reaching the level of contextual understanding that allows this level of translation quality, how much time do we, humans, have left before we either launch the Butlerian Jihad to get rid of thinking machines for good, or accept becoming a footnote in the evolutionary history of consciousness and intelligence?

Speaking of footnotes, here’s a footnote of sorts: Google does know that an alligator clip is a pince crocodile in French or Krokodilklemme in German. Bing knows about Krokodilklemme but translates the phrase as clip d’alligator into French.

 Posted by at 5:51 pm