Update 6/1/2017: The issues discussed in this post are more relevant to Statistical Machine Translation (SMT), which is how Google Translate used to work. The advent of Neural Machine Translation (NMT) makes my theory a moot point. But you should still be wary of machine translation for essential business publications.
Last week, when translating from Russian into Ukrainian, automatic translation tool Google Translate converted “Russian Federation” into “Mordor,” the evil realm from Tolkien’s Lord of the Rings. “Russians” became “occupiers” and foreign minister Sergey Lavrov was called “sad little horse.” The political relationship between Russian and Ukraine has been especially strained, to say the least, by their ongoing territorial dispute. But Google, a multinational corporation, didn’t side against the Russians. The politically charged mistranslations were dismissed as a bug or glitch. In a public statement, a Google spokesperson emphasized that the online translation tool works “without the intervention of human translators.”
That’s not exactly true. No individuals directly intervene in an instance of automatic translation, but humans determine the outcome at the level of the population. Google Translate works by scanning for patterns in the hundreds of millions of sentences already on the Web. The algorithm decides the best possible match for a term by tracking how it’s already been translated and how it is commonly used. This can be manipulated. In 2003 a popular columnist masterminded a public smear campaign against an American politician. By establishing web domains named after the politician and encouraging readers and allies to pair the politician’s name with a raunchy concept in online searches, a profane definition of that politician’s name was established that still tops Google results today. In the evolution of language, if a term is used enough, it becomes common usage. Google just accelerates the process.
It is true that Google Translate does not employ human translators. It works through a form of crowd sourcing because it uses what people have already written on the Web. If enough Ukrainians were angry enough at the Russians to vent their feelings online, and if certain disparaging trends and terms became ubiquitous in the Ukrainian-language web archive (which includes websites, comments, social media – everything) those terms became common usage. To the Google Translate algorithm, “Mordor” is the “correct” translation, according to the wisdom of a people hostile to the Russian Federation.
While it may still be revealed that the incident was the result of a deliberate hack (which would also make Google look bad), the basic principle remains. When you use Google Translate for business translation, the results may reflect population-level language trends of which you are not aware. To maintain control of your brand and your reputation, don’t rely on a free online tool. You will get what you paid for. If your message matters, use professional human translators.