While Egyptians were busy voting for a new president, a glitch in Google Translate’s system accidentally revived the presidency of Hosni Mubarak.
When one typed “I will respect Egypt’s future president” in English, the Arabic translation was, “I will respect President Hosni Mubarak.”
Mubarak, who is reportedly clinging to life on a respirator, will not be Egypt’s future president. But as his military allies assert their power and Egyptian revolutionaries cry “coup,” Mubarak’s online appearance sparked suspicion and disbelief on Twitter and Facebook.
So was it a conspiracy by Mubarak’s old regime? Google says no. It was just a glitch in the algorithms it uses to learn and translate 64 languages, according to Google spokesman Jason Freidenfelds.
The program is totally automated and learns based on the pattern of how often words and phrases are translated into other languages on the Web.
That means when Google Translate scours the Internet for references to Egypt’s president, the most common results in Arabic pair Hosni Mubarak with that phrase.
The quality of the translation doesn’t depend so much on the prominence of the language. It depends on that language’s prominence on the Internet.
“For English and Spanish, there are tons of translations out there, and there have been for years. So you’ll get a high quality translation with the right phrases and words,” said Freidenfelds, adding that the difference between the structure of the two languages also matters.
“In French or Spanish, you might flip the adjectives and nouns. That’s not that hard for machine translation to pick up on. But if the change pushes a word all the way to the end of the sentence, that’s a lot harder for the system to pick up,” he said.
For a language like Arabic, it may take a while for Google Translate to reflect the changes taking place in the real world.
“Humans process language by understanding the meaning of the words. We know who the president is and that it’s somebody’s name. Google Translate doesn’t have the semantics of the language,” said Freidenfelds.
So can you teach it to better understand the nuances of a language? Right now, Google Translate is as smart as the system the company’s team of engineers and researchers can come up with.
Once they identify a glitch, like the Mubarak translation, they won’t change that one phrase. Instead, they’ll try to improve the algorithms overall.
Update: Google Translate has changed the translation of this phrase since the glitch was first discovered.