You ever seen those Google Translate music videos? Where singers or other entertainers sing songs that have gone through Google translate or another “machine translation” program? Whelp, it turns out, that’s how Moscow often creates its lower-tier propaganda. It either uses Google Translate or low-rent translators who are not especially proficient in the target language, leading to a problem where anyone who can read at a middle school level or better is largely resistant to it.
Google Translate Sings: “Shape of You” by Ed Sheeran
(Side note: How is Ed Sheeran just as catchy when the lyrics become total nonsense? I’m in love with your system, baby.)
RAND Researcher Joe Cheravitch has a new paper in the Small Wars Journal and on the RAND Blog that discusses the problem in great detail, but it’s not new. In 2017, BloombergOpinion published a piece about how people are intentionally throwing off the machine learning of translators like Bing and Google to get funny results.
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.
Basically, old machine translation was horrible because languages change too often and break their own rules constantly, so it’s impossible to translate living text with the rigid rules that computers follow. So Google and other mass translators switched to neural AI, where machine learning is used to look at entire passages of text in multiple languages.
Over time, the AI gets better and better at translating according to how the language is actually used. But it is always limited by the quality of the text it receives. And pranksters, bad actors, and others can throw off the translation of any rarely used word, such as a proper name, by suggesting a specific alternate translation repeatedly.
But of course, Russia can just drag in a couple of top-tier translators and fix the issue, right? There are native speakers in Russia. That’s where Edward Snowden ran off to and where he can still be found when he needs to promote his new book.
Well, apparently it can’t. Because while the Russian military hacking network “Guccifer 2.0” was legendarily successful at hacking the U.S. political apparatus and leaking data through WikiLeaks, it has also operated in Europe and elsewhere. Its ability to break into computers is impressive; its language skills are laughable. (Also, amusingly enough, its ability to prevent incursions on itself was also bad, according to reports in VICE.)
The obvious question is why Russian military intelligence approves these operations at high levels and recruits high-level hackers to break into the targeted computers but then fails to hire sufficiently skilled translators. There are a few potential explanations for this.
First, talent is expensive, and Russia needs translators that are fluent in foreign languages in a lot of places that are arguably more important than undermining Romanian support for a particular candidate. Russia’s economy is heavily reliant on oil. In 2017, 60 percent of its GDP came directly from oil exports. Since it’s selling across Europe through pipelines and the rest of the world through shipping, translators can make more money in that sector.
But worse, there appears to be a bit of a problem holding on to talent in the military if it becomes sufficiently proficient. Avid military news readers know that the U.S. military is struggling to retain pilots as civilian airlines scoop them up. Well, Russian-English translators can get easy work by joining the military. But the constant experience sometimes makes them better translators, allowing them to break into a new income bracket by leaving a few years later.
Back to Cheravitch’s paper for a moment, this brain drain may give digital forensic teams and U.S. policymakers a chance to catch these Russian influencers and create new programs to limit their effect:
Tipped off partly by linguistic mistakes, researchers with the Atlantic Council’s Digital Forensics Lab were able to piece together a distinct influence effort attributed to Russian military intelligence following the 2016 election-meddling effort. This sort of work could have obvious benefits for policymakers, who can more appropriately respond to this activity with a better understanding of the actors behind it.