Google Translate Sabotage, part 2

« previous post | next post »

This is all over the Chinese internet:


The Chinese translation given is:

Zhōngguó xìnshǒu nuòyán


"China keeps its promise"

That's the exact opposite of the English original:  "China breaks promise"

This is so far off and politically motivated that it seems someone, or some group, must have sabotaged Google Translate in this particular case.

Baidu Fanyi correctly renders "China breaks promise" as " Zhōngguó shīxìn le 中国失信了".  It's an easy sentence.

Update: Google has already corrected it. The translation of "China breaks promise" is now correctly rendered as " Zhōngguó wéi nuò 中国违诺".

Selected readings

[Thanks to Chau Wu]


  1. Jerry Friedman said,

    January 17, 2021 @ 10:14 pm

    The more grammatical "China breaks a promise" is still rendered as
    Zhōngguó xìnshǒu nuòyán

    "China breaks its promise" is rendered as
    Zhōngguó dǎpòle

  2. Jerry Friedman said,

    January 17, 2021 @ 10:19 pm

    Sorry, the one with "dapole" was the result of my accidental carriage return. "China breaks its promise" is also

    Zhōngguó xìnshǒu nuòyán

    So it's the same as with "a promise".

  3. Persecuted by CCP said,

    January 18, 2021 @ 2:28 am

    Nice try. It is still translated as "China keeps is promise" when you type in "breaks its promise"/"breaks a promise" like the other guy has suggested.

    Not only that, Google Translate's Traditional Chinese translation of terms like "video" and "quality" are also translated into communist China's version instead of their real Traditional Chinese counterparts. These translations are even "verified" by Google Translate's community of contributors.

    Chinese people and pro-CCP Google staff are literally intruding our language and culture (HK, HK, Macau) as we speak.

  4. ~flow said,

    January 18, 2021 @ 2:30 am

    Just tried out a few variations on

    China breaks promise -> 中国违反诺言

    China breaks promises -> 中国信守诺言

    China breaks its promise -> 中国信守诺言

    China breaks a promise -> 中国信守诺言

    China breaks another promise -> 中国违背了另一个诺言


    America breaks promise -> 美国违约

    America breaks promises -> 美国信守诺言

    America breaks its promise -> 美国信守诺言

    America breaks a promise -> 美国违背了诺言

    America breaks another promise -> 美国违背了另一个诺言


    Google breaks promise -> Google违约

    Google breaks promises -> Google违背承诺

    Google breaks its promise -> Google违背了诺言

    Google breaks a promise -> Google违背了诺言

    Google breaks another promise -> Google违背了另一个承诺


    Interestingly Google don't wants to translate its own name although when I go and edit the English text I sometimes see 谷歌 flashing up in the translation.

    America gets the benefit of the doubt for 'promises' and 'its promises' (2/5); China also for 'a promise' (3/5), while Google never second-guesses its own intentions (0/5). FWIW Russia also gets 3/5 benefits.

    I wonder if you trained a network with gazillions of "this is so wrong -> 這太錯了“ and a few "this is so right -> 這是對了", then asked the system for "this is so right"—would it be triggered to shout the wrong answer just because it has been so conditioned to anticipate "wrong" after "this is so", totally neglecting the one important hint?

    I friend of mine just told me he has disabled predictive text and auto-correction on his smartphone "for fear of accidentally starting a war". Given how much across opinions can be about what has been said, what has been meant, and what has been implied, we're surely treading on thin ice were we to hand over translations to AI in a rush. At least when I tested "AI breaks promise" it came out the right way though so there's still hope.

  5. John Swindle said,

    January 18, 2021 @ 11:42 am

    Breaking a promise, keeping a promise: It's important to anthropomorphize here. Because Google Translate is driven by neural nets (waves hand, visualizes black box), it can only go by what it sees most often in a given context. Reversals of meaning sometimes occur. Human mischief isn't required. Human correction may be required.

    @Persecuted by CCP: A US organization that deals with the public sends multilingual notices to its clients offering interpreters as needed for many languages including "Chinese (繁體中文)". I've tried to convince them that Traditional Chinese (繁體中文)isn't a language. It's a language now?

  6. David Cowhig said,

    January 18, 2021 @ 11:51 pm

    The problem seems fixed now.

    Two days ago, I got funny results

    Panda breaks promise came out as Panda keeps promise
    Xi Jinping breaks promise came out Xi Jinping keeps promise
    Tsai Ingwen breaks promise came out Tsai Ing-wen keeps promise
    ROC breaks promise came out ROC keeps promise

    Bald Eagle breaks promise came out as Bald Eagle breaks promise.
    Donald Trump breaks promise came out as Donald Trump breaks promise.

    There seemed to be a little bias towards PRC and Taiwan in there!
    However the Panda breaks its promise was translated correctly.

    "PRC breaks its promise" was translated correctly.

    So most likely a technical glitch!

    Along the way I came across Douglas Hofstadter's article on the weaknesses and strengths of Google Translate from The Atlantic
    "The Shallowness of Google Translate" at

  7. B.Ma said,

    January 19, 2021 @ 9:02 am

    @Persecuted by CCP / John Swindle

    Wikipedia gets it right with the ability to display articles in 6 different types of written Chinese (i.e. Mandarin): Mainland Simplified, HK Traditional, Macau Traditional, Malaysia Simplified, Singapore Simplified and Taiwan Standard. Users can contribute to the complex conversion tables.

    I'm not sure how much difference there is between HK and Macau.

    Of course there are separate topolectal Wikipedias too.

  8. Terpomo said,

    January 20, 2021 @ 5:04 am

    @Persecuted by CCP
    The notion that the traditional Chinese versions are somehow more 'real' is just bad linguistics, the same as if someone were to insist that British English is inherently right and American English inherently wrong. That said, that the Traditional Chinese translations still use Mainland terms is problematic; the problem is that there's only one translator, whose output can be converted to traditional, with no further controlling for terminology like Wikipedia's converter (although a terminology converter does seem manageable, if Wikipedia can do it.) And you can tell it works internally in Simplified, because it translates 發 as 'hair'!

  9. Philip Taylor said,

    January 20, 2021 @ 8:08 am

    "The notion that the traditional Chinese versions are somehow more 'real' is just bad linguistics, the same as if someone were to insist that British English is inherently right and American English inherently wrong". Well, it is. But we'd better not go there !

RSS feed for comments on this post