Google computers hurt the feelings of the Chinese people

You’re playing with the Google English-Chinese translation beta, and like any curious individual, you start trying out some dirty language. So you enter “I thought this was fucking shit” and out comes “我认为这是中国运动员拉屎” (I think this is Chinese athlete shit.) Hmmm.

Then there’s “i thought this was shame” which ends up “我认为这是中国的耻辱” (I think this is China’s shame), and “i thought this was fucking” which becomes “我认为这是中国运动员”, (I think this is Chinese athlete).

Tianya forum poster “Fat cat who walks by himself” discovered this and wrote it up in a post calling for a boycott of Googles machine translation. In the thread, posters concentrated on the “fucking” problem. “Seoii” summarizes:

It’s only when you have “thought this was” + “fucking” that “fucking” is translated into “Chinese athlete”

Otherwise it’s just “athlete”. My guess is that it’s a programmer messing around.

Shanghai Morning Post picked up the story, running it under the title “Google E-C translation tool loses its mind.” The reporters noted that it was not only derogatory phrases that got associated with China; the sentence “I thought this was glory” became “我认为这是中国的荣耀” (I think this is China’s glory). In another example, the Chinese “坏学生” (bad student) became “good students.”

Google’s Ogilvy PR representative explained that the errors were due to the way Google’s statistical machine translation operates: it analyzes corresponding words and phrases in a huge pool of bilingual documents to determine the most likely translation. Documents related to international affairs get translated correctly, he said (the inference being that since “fucking” is unlikely to be found in many of the documents in Google’s translation database, the translator did not have much data to work with).

The rep also ruled out programmer mischief, saying that human interference “did not exist and could not possibly occur.” There appears to have been some human interference in eliminating the problem, however – the offending examples were gone this morning, leaving the original Tianya forum thread, which had no screenshots, to devolve into accusations of rumor-mongering.

The speed at which this all was resolved is actually pretty impressive. The Tianya post was made yesterday, it was picked up on Google’s translation forums last night, and it’s already in Wikipedia as an example of criticism of Google’s Language Tools.