Chinese translation app with built-in censorship

« previous post | next post »

What good is a translation app that automatically censors politically sensitive terms?  Well, a leading Chinese translation app is now doing exactly that.

"A Chinese translation app is censoring politically sensitive terms, report says", Zoey Chong, CNET (11/27/18)

iFlytek, a voice recognition technology provider in China, has begun censoring politically sensitive terms from its translation app, South China Morning Post reported citing a tweet by Jane Manchun Wong. Wong is a software engineer who tweets frequently about hidden features she uncovers by performing app reverse-engineering.

In the tweet, Wong shows that when she tried to translate certain phrases such as "Taiwan independence," "Tiananmen square" and "Tiananmen square massacre" from English to Chinese, the system failed to churn out results for sensitive terms or names. The same happened when she tried to translate "Taiwan independence" from Chinese to English — results showed up as an asterisk.

According to "Chinese AI champion iFlytek censors politically sensitive terms on its translation app", by Zen Soo and Meng Jing, SCMP (11/28/18):

The Android version of the company’s iFlyTranslate app has removed items such as “Tiananmen”, “independence” and the name of China’s President, “Xi Jinping”, from showing up as results in its system.

All of this is utterly ridiculous.  What if an innocent user wanted to render this statement into Chinese, "Today in Tiananmen, President Xi Jinping denounced Taiwan independence" or some other equally factual but politically uninflammatory sentence that contained one or more of these terms that would automatically be censored?  Such a system would be completely useless and totally unreliable, because it would eviscerate a user's speech of countless terms of which the government disapproves, such as "harmonize", "river crab", "Tibet", "Xinjiang", "Uyghur", "Falun Gong", "explosion", "protest", and so forth.

Once the authorities start down the road of AI translation censorship, there's no end to it.  To be on the safe side, they will make sure that nobody can translate "frog", "jasmine", "Winnie the Pooh", "pollution", "freedom", "democracy", and countless other words that are potentially anti-Party.  O, that way madness lies.



6 Comments »

  1. Michael Watts said,

    November 29, 2018 @ 4:55 am

    What if an innocent user wanted to render this statement into Chinese, "Today in Tiananmen, President Xi Jinping denounced Taiwan independence"

    I assume the Chinese-speaking audience would already be well-informed about this through media that didn't need to translate the event from the original Chinese into English and then back into Chinese.

  2. Victor Mair said,

    November 29, 2018 @ 7:25 am

    "I assume the Chinese-speaking audience would already be well-informed about this through media that didn't need to translate the event from the original Chinese into English and then back into Chinese."

    That is a highly questionable, strained assumption.

  3. Scott P. said,

    November 29, 2018 @ 9:34 am

    I assume there is already some circumlocution that Xi would use rather than make use of the phrase "Taiwan independence," even if he were giving a speech denouncing it? Something like "anti-Chinese separatism" or thereabouts?

  4. Victor Mair said,

    November 29, 2018 @ 11:30 am

    For as long as I can remember, I've always heard people talk about "Taiwan independence" as "Táidú 台独", which is short for Táiwān dúlì 台灣獨立. Even if Xi used some custom-made circumlocution, others talking about his opposition to it would use the normal expression.

  5. mg said,

    November 29, 2018 @ 4:08 pm

    Nice article on the computer science student who reverse-engineered things to find this:
    https://www.scmp.com/tech/apps-social/article/2174875/jane-wong-explains-why-she-uncovers-hidden-app-features-tech-giants?fbclid=IwAR0kxa3h0G9E3Z5Qu0ipzmNOM8VpXT6OPD6uG3r5o3MQxHJ6vzdH6qDL_r0

  6. KB said,

    November 29, 2018 @ 10:25 pm

    > the system failed to churn out results for sensitive terms or names. The same happened when she tried to translate "Taiwan independence" from Chinese to English — results showed up as an asterisk.

    An asterisk? Should have gone with "thoughtcrime".

RSS feed for comments on this post

Leave a Comment