Machine Translation and the case of the Offensive Couch

Note: This post contains a couple of racial slurs in the context of the story – there’s no real way to tell it without using them so if you’re easily offended you may wish to skip this one.

A few weeks ago up here in Toronto there was an unfortunate case where a family purchased a brand new leather couch and shortly after delivery their seven-year old daughter found a label on it. Not an unusual story unless you also take into account the family was black and the label in question used the term “Nigger-brown” to describe the colour of the couch.

Needless to say they weren’t happy and this story got into the news cycles locally, and within a week I caught the story on CNN. An interesting update on the story came out today though which makes it a candidate for this blog: The label is being blamed on an error by Chinese-English translation software.

“We got the definition from a Chinese-English dictionary. We’ve been using the dictionary for 10 years. Maybe the dictionary was updated, but we probably didn’t follow suit,” said Huang Luoyi, a product manager for the Beijing-based Kingsoft Corp., which manufactures the translation software that produced the offensive description.

“I know this is a very bad word.”

He explained that when the Chinese characters for “dark brown” are typed into an older version of the company’s Chinese-English translation software, the word “nigger” comes up

Yikes. It’s an unfortunate case of “Garbage In/Garbage Out” but I still can’t believe that along the way someone didn’t catch it – if there was ever a justification for human, in-country review of any translation this is certainly one of them.

The company is assuring everyone that the slur has been removed from new versions of the software.