Man arrested because of a Facebook mistranslation

A Palestinian man was arrested by Israeli police last week, after a post he left on Facebook saying “good morning” in Arabic was taken to meaning “attack them” in Hebrew.

A spokeswoman for the Israel Police’s West Bank district confirmed to The Times of Israel on Sunday that the man, a construction worker living in the West Bank, was arrested “on suspicion of incitement”. The man was held in custody briefly, but was released as soon as the mistake had been realised.

There is one letter’s difference between the Arabic phrase for “good morning to you all” and the Hebrew for “attack them”, the report says. Facebook had automatically translated the post from Arabic, and no Arabic-speaking officer read it before making the arrest.

The post also contained a picture of the man holding a cigarette standing next to a JCB. The authorities took this to be threatening, when combined with the mistranslation, because bulldozers have been used in the past in terror attacks. The post has since been deleted. 

Facebook has recently changed the way it translates between languages, but it seems the new method is not error-free. The company announced in August it had shifted towards neural machine translation. This uses convolutional neural networks (CNNs) and recurrent neural networks (RNNs) to automatically translate content, instead of phrase-based statistical translation. This was expected to quicken and improve the translations.

“Our problem is different than that of most of the standard places, mostly because of the type of language we see at Facebook,” Necip Fazil Ayan, engineering manager in Facebook’s language technologies group, said at the time. “We see a lot of informal language and slang acronyms. The style of language is very different.”

But this is not the first time an online translation tool has caused problems. Last year, Google translated the Russian Federation to ‘Mordor’, the fictional land in Lord of the Rings using its automatic translation software. Because language depends so much on context, Google said, mistakes like this were bound to happen occasionally.