Leonid Bershidsky: Don't use Google Translate to annoy trolls

Humans can easily ruin one of the best things about the internet. Photo / 123RF

Algorithmic translation has taken some giant steps forward in recent years, with services like Google Translate or Bing Translator now actually resembling texts written by humans -- albeit extraordinarily awkward, unintentionally funny ones.

But that impressive run of robotic progress can be quickly reversed if humans insist on intervening in the wrong way. Louise Mensch, the British parliamentarian who transformed into a leading anti-Trump personality on Twitter, has just stumbled into one of those ways.

Use Google Translate to tweet back at Russian trolls in Russian, she suggests, and they'll go away. But I'd be careful with that: the strategy could easily backfire, and, if widely employed, really mess up an invaluable service that took years to build.

The first generation of online translation systems relied on the flawed approach of trying to teach computers how language was structured. Unfortunately, living languages tend to disregard rules. Instead, modern translation engines use statistical analysis, parsing parallel text in different languages to figure out how a word, a turn of phrase or a sentence is most frequently translated.

Coupled with artificial intelligence -- neural translation, Google calls it -- the results are even more impressive because the algorithm looks at whole sentences rather than smaller bits and pieces. Google Translate now uses this approach in multiple languages.

As a result of the steady improvement, it's producing far fewer of those hilarious "Google Translate Fails."

But since it relies on previous translations, the statistical method is vulnerable to abuse -- especially since Google allows users to suggest alternate translations.

If you click on the incorrect French translation of "go-to person," you'll see a correct alternate version someone has suggested -- "une personne-ressource" -- and you'll be able to submit your own version. That, very likely, is how Google Translate came to interpret the Korean characters for "Supreme Leader" -- Kim Jong Un's title -- as "Mr. Squidward," a character from the Spongebob Squarepants cartoon.

In a similar situation last year, when Google Translate repeatedly translated "Rossiyskaya Federatsiya" (Russia's official name in Russian) into Ukrainian as "Mordor" and "Lavrov" (the Russian foreign minister's last name) as "sad little horse," Google said it was just a glitch. That's highly unlikely. Though the company said in its statement that its translations were based on "patterns in hundreds of millions of documents," there are far fewer occurrences of any particular word, especially a proper name, in bilingual texts.

A group of dedicated people out for some fun can easily turn the North Korean dictator into Squidward, Russia into Mordor and the "lorem ipsum" placeholder text into code for political statements. In that bizarre 2014 case, playing with the garbled Latin placeholder produced various references to China, NATO and the Internet. For example, this:

"Lorem ipsum dolor sitlorem ipsum ipslorem ipsum lorlorem ipsum lolorem ipsum loremlorem ipsum amatLorem Ipsum"

... translated into this:

"Our goal is tovehicle dimensionsFree of painChina, elsewhereFree InternetChina lovesNATO"

Cybersecurity researchers even suspected this was some kind of secret communication channel, but apparently someone merely messed with Google Translate to create a clue for an elaborate quest played at the Defcon hacker convention. It no longer quite works, although playing with the "lorem ipsum" text on Google Translate still produces some quaint results in multiple languages.

This all sounds like harmless fun until one considers a future society that comes to trust artificial intelligence so much that human translators are no longer used as too expensive and inefficient (grounds on which Silicon Valley proposes to get rid of drivers). There would be limitless possibilities for maliciously creating misunderstanding.

Even now that humans are still very much involved, translation errors can lead to political consequences. For decades, Soviet leader Nikita Khrushchev's words at a Moscow reception, addressed to the capitalist West -- "My vas pokhoronim" -- was translated literally as "We will bury you," and interpreted as a threat. Khrushchev meant the Communist system would outlive the capitalist one.

More recently, President Donald Trump decided that Russian President Vladimir Putin had called him "brilliant" or "a genius," though the original Russian -- "yarkiy" -- merely means "colorful."

Imagine if machine translations flew back and forth in the propaganda wars of the future. Weaponized mistranslation would quickly stop being a prank. Machines would learn from bodies of deliberately distorted bilingual text produced by other machines, and there wouldn't be enough people available to parse the results because professional translators would be as outmoded as chimney-sweeps.

I'm discussing this only half in jest. Mensch, after all, appears to be serious. But she doesn't realize how quickly the trolls could convince Google Translate to redirect their Twitter fight directly into the heart of Mordor.

Bershidsky is a Bloomberg View columnist. He was the founding editor of the Russian business daily Vedomosti and founded the opinion website Slon.ru.