Google Fixed "Nip It in the Butt." Microsoft Translated It Literally.

We gave two translation engines the same mistake and sent it one hop through Japanese and back:

In: We need to nip it in the butt.
Microsoft: I have to bite my butt.
Google: It needs to be stopped in the early stages.

Google's Chinese went one better and returned "We must nip it in the bud," the idiom exactly as it should be.

In our test of its, their and your mistakes, both engines fixed nearly everything, because those errors leave only one sensible way to read the sentence. Eggcorns, malapropisms and spoonerisms are different. Every one of them is a valid English sentence that happens to be wrong, and that is where Google and Microsoft stopped behaving alike.

All results below are real output from Translation Mixer. We ran each sentence through Spanish, French, German, Japanese, Russian and Chinese twice, once on Microsoft's Azure Translator and once on Google Translate, plus a set of smaller languages that only Google supports. Each test is one hop out to a single language and straight back.

What an eggcorn is, and what each one should say

An eggcorn is a misheard phrase that still sounds logical. The name comes from someone who wrote "egg corn" for acorn, which is wrong but makes a certain sense (an acorn is a seed, roughly egg-shaped), and the linguists at Language Log adopted it in 2003. Grammar Girl has a clear guide to how eggcorns differ from malapropisms, spoonerisms and mondegreens, if you want the full taxonomy. Here are the four we tested, with the phrase each one is reaching for:

Each eggcorn survives because it almost works. Intensive purposes sound like serious ones. A wolf could plausibly wear a cheap disguise. "Doggy dog" sounds like a playground version of the real thing. And a nip in the butt does, in fairness, tend to stop a dog from doing something.

That is also what makes them a hard test for a translator. None of them is ungrammatical, so nothing in the sentence says it needs fixing.

The results, engine by engine

Here are the same six languages on each engine:

Microsoft translated every word it was given. "Intensive purposes" came back as "intensive uses" from French and "focused purposes" from Japanese. The wolf kept its cheap clothes, and French turned him into "a cheap wolf in disguise." The doggy dog world produced "The world of dogs is a world of dogs" in Spanish and "The world of dog dogs is a real world" in Russian. "Nip it in the butt" went the worst, because "butt" is a real word with a clear meaning:

German: We have to smother it in the butt.
Japanese: I have to bite my butt.
Russian: We need to detain him in.

Russian, at least, had the decency to leave the butt out.

Google, on the same six languages, produced no butts at all. Spanish wrote Tenemos que atajar esto de raíz, "We need to tackle this at its root," which is how Spanish says nip it in the bud. Japanese said it needed to be "stopped in the early stages." Chinese used 扼杀在萌芽状态, its own equivalent of the idiom, and came back as "We must nip it in the bud." French and German lost the "early" part and settled for "put a stop to it," which still gets the point. Only Russian slipped, and it slipped in a strange direction: "We need to stop this at the very end."

"Intensive purposes" was understood everywhere. Spanish wrote A efectos prácticos, and the English came back as "For all practical purposes." The others gave "in practice," "essentially," "in effect," "in essence" and "from all perspectives," all of which mean roughly what the real phrase means. The wolf got his sheep's clothing back in Spanish (piel de cordero) and German (Schafspelz), both of which have the same idiom from the same verse.

Some of Google's repairs happened on the way home

Those were repairs on the way out: the middle language already had the right idiom. Others happened on the way back. Google's French wrote un loup déguisé en minable, "a wolf disguised as a loser," and its Chinese wrote 披着廉价外衣的狼, "a wolf wearing a cheap coat." Both middle sentences still had the eggcorn in them, and both came back to English as "a wolf in sheep's clothing." French did the same with the doggy dog world. The middle said un monde de chiens, a world of dogs, and the English said "It's a dog-eat-dog world out there."

So Google is fixing these at both ends. When it reads your English, it notices that "cheap clothing" next to a wolf is probably a famous phrase gone wrong. When it writes English, a sentence about a wolf in some kind of disguise gets pulled toward the version English speakers say all the time.

Google's smaller languages did much less of this. Across Hawaiian, Esperanto, Latin, Scottish Gaelic, Cebuano, Javanese and Yiddish, the wolf was restored 0 times out of 7 and "intensive purposes" twice. Those two, in Esperanto and Scottish Gaelic, were also return-leg repairs. Esperanto's middle sentence was Por ĉiuj intensaj celoj, "for all intensive aims," and the English that came back said "for all intents and purposes." Asked to write an English phrase that starts with "for all" and ends with "purposes," the English side fell into the most familiar groove available.

Google fixed the pineapple

A malapropism swaps in a real word that sounds similar to the right one. The term comes from Mrs. Malaprop in Sheridan's 1775 play The Rivals, who calls a man "the very pineapple of politeness." She means pinnacle, the highest point, and the joke is that she reaches for impressive words and keeps grabbing the wrong ones.

On Microsoft, Spanish, French, German, Japanese, Russian and Chinese all kept the pineapple, and so did Google's Esperanto. Chinese went with "He was practically the pineapple of politeness," which is almost a compliment.

On Google, the same six languages fixed it every time: "the epitome of politeness," "the very embodiment of politeness." German is the clearest example of how. Its middle sentence was Er ist die Ananas in Person, wenn es um Höflichkeit geht, "he is the pineapple in person when it comes to politeness." The pineapple was still there. The English that came back said "He is the epitome of politeness."

That is a fix, but think about what it fixed. Mrs. Malaprop's mistakes are the whole joke. A translator that silently corrects her would hand you a version of The Rivals in which nobody can work out why she is funny.

Nobody fixed the spoonerism

A spoonerism swaps the opening sounds of two words. They are named after William Archibald Spooner, warden of New College, Oxford, who supposedly made a habit of them, though most of the famous examples were probably invented by his students. The best known is "You have hissed all my mystery lectures," meaning "You have missed all my history lectures."

It went through 14 language runs across both engines, and none got there:

Microsoft French: You whistled all my mystery lectures.
Microsoft Japanese: You're silent about all my mystery lectures.
Google Spanish: You've booed all my mysterious lectures.
Google Chinese: You scoff at all my esoteric lectures.

Google did no better here, and the reason is the same thing that made it good at eggcorns. "Nip it in the bud" is a fixed phrase that turns up constantly, so a near miss has something to snap back to. "You have missed all my history lectures" is just an ordinary sentence. Nothing about "hissed all my mystery lectures" points toward it, and hissing at lectures is a perfectly sensible thing to accuse a student of.

Which engine got it right?

It depends on what you wanted. If you typed "nip it in the butt" by accident, Google read your mind and Microsoft embarrassed you. If you were translating a character who talks in malapropisms, or quoting someone's mistake on purpose, Google's version quietly destroys the thing you were trying to keep, and Microsoft's literal habit is exactly right.

We can't see inside either engine, but the outputs tell a consistent story. Both read your sentence and translate what it says. Google, in the languages where it has the most data, also checks whether what you wrote is a near miss of something people say all the time, and if it is, it gives you the thing people say. That works for idioms, because idioms are repeated millions of times. It does nothing for a spoonerism, because there is no famous phrase on the other side of it. And it does much less for Hawaiian or Latin, where there simply isn't enough text for any phrase to become famous.

Try it

"Nip it in the butt" is loaded, going out through Japanese. Which engine handles it depends on how the site is routing that language when you run it, so you may get the bud or you may get the butt:

→ Run "nip it in the butt" through the mixer

Then try Mrs. Malaprop's pineapple, and see whether it comes back as fruit.

---

Related: Translators Fix "It's Paws" Every Time. "Over Their" Is Another Story. covers the homophone mistakes both engines fix. Famous Mistranslations in History, Recreated shows the opposite problem, where the input was right and the translation wasn't. Words That Survive Translation Perfectly covers what makes it home intact.