I translate from French to English, and one of the tools I use is https://context.reverso.net. You search for a phrase and it returns 10-20 examples of that phrase in both languages, in context, pulled from the Internet. It is so helpful for idiomatic translations, because it lets you quickly see how others have rendered it in different contexts. For example, I once used it to see how others have translated the French phrase “Et hop!,” which is one of those all-purpose phrases that, in different contexts, can mean “There you go,” “Done,” “Presto,” “Up you go!” etc. However, in the last year the results have gotten noticeably worse: literal translations are crowding out the idiomatic ones, because of the huge amount of auto-translated text online.
A good example is the French phrase “ni une ni deux”. Literally, “neither one nor two,” but what it actually means is “right away.” (I think the idea is something like, don’t even take the time to count “one, two, three, go,” just do it.)
Here are some good translation examples:
But the majority of the examples on this page are literal, non-idiomatic translations like this one:
Alors ni une ni deux Marc fait vrombir le moteur et nous prenons la route de Louxor pour les y retrouver.
Then neither nor two Marc makes hum the engine and we take the road of Luxor for finding there.
It’s almost certainly a bad machine translation, because it doesn’t even make sense. (The French sentence actually translates to: “Without hesitation, Mark revs up the motor and we hit the road for Luxor to meet our friends.”) I can’t imagine an actual, English-speaking human writing “Then neither nor two Marc makes hum the engine and we take the road of Luxor for finding there.”
Last week Vice noticed that the web is full of bad translations. It’s another example of AI mediocrity. AI translation isn’t doing a good job, but it’s doing a sufficiently mediocre job that people who don’t care about the difference, or can’t tell, have decided to use it anyway. Unfortunately, these bad-to-mediocre translations are poisoning the tools human translators use to render good ones.