Why is Facebook's Chinese translation still so terrible?

« previous post | next post »

[This is a guest post by Jenny Chu]

Has Language Log been following up on the great sorrow that is Facebook's (Chinese) translation feature? The last reference I found was this one

It came up today when I was reading this somewhat viral post on Facebook

I switched on the auto-translate option to help me understand. The results were not just astonishingly bad, but had a surprisingly medical bent.
今天這個主權政府作承諾的時候大辭炎炎,七情上面,結果又是如何?–> "Today, when the private government is working, the weather is colon inflammation, above the sentiment, what is the result?"
我對很多前輩及同輩仁人不斷被威嚇打壓感到憤怒,–> "I guess that many former Egyptians and peers are angry for not being bothered by threats,"
如果要扭曲如蛆蟲,或者把扭曲了的現實荒謬視作正常,於我而言,將會是十分痛苦的事。–> "Cervixing is extremely painful to normalize a skeleton, or to normalize a crooked current situation, for me. "
Former Egyptians? Colon inflammation?? Cervixing???
A few years ago, I would have shaken my head and agreed that Chinese, with its idioms, references, etc., is just too tough for any AI to handle. However, Google Translate has proven that wrong. So – when I'm on Facebook, why do I still feel like I'm reading a translation from Babelfish circa 2007?

Selected readings


  1. Jan said,

    April 27, 2022 @ 6:17 am

    Even for German to English it's still comparatively bad. E.g. "Bauer" as a stand-alone word means "farmer", but when used as a suffix in a compound noun, it gains the non-agricultural meanings of "builder" or "worker" in general.

    Despite that, Facebook still loves literally translating any mentions of e.g. "Straßenbauer" (road builder) or "Metallbauer" (metalworker) as "road farmer" respectively "metal farmer".

  2. Tommi Nieminen said,

    April 27, 2022 @ 8:22 am

    I can think of three reasons for this:
    1. Nobody at Facebook cares much about the machine translation systems in production use (as opposed to machine translation research, where Facebook has a good track record).
    2. Facebook has optimized the throughput of the production machine translation systems so aggressively that quality has been degraded.
    3. They are using some kind of cutting-edge machine translation technology that is very impressive technically but breaks down in actual use (maybe something like the system described here: https://about.fb.com/news/2020/10/first-multilingual-machine-translation-model/)

    The interesting thing here is the apparent contradiction of having a first-class MT research department, and the actual production MT systems being worse than systems that anyone can train with open-source frameworks and data.

  3. Terry K. said,

    April 27, 2022 @ 2:20 pm

    Facebooks translations being relatively poor isn't specific to Chinese. I mostly encounter Norwegian, and I know that if I really want to understand something, I should copy it and use Google Translate.

  4. Dara Connolly said,

    April 27, 2022 @ 3:06 pm

    Facebook, by default, presents a (poorly) translated version of the original post into the user's feed, and does not show the original post. This is very annoying for a number of reasons. First, because the mangled translation often makes the poster look like an idiot, or worse*. Second, because the whole point of the post may have been some wordplay in the original language, which is lost in translation. Third, because it means that when you write a post on Facebook you have no control of (or knowledge, even after the fact) of what your international friends will see and think you have written.

    * A Finnish friend on a work trip wrote on Facebook that she was going to become a tourist. Facebook's automatic translation to English had her saying she was going to become a satanist.

  5. mg said,

    April 27, 2022 @ 4:44 pm

    Some of the Russian translations are unbearably bad.

  6. Jenny Chu said,

    April 27, 2022 @ 9:41 pm

    I was just testing and it there seems to be no rhyme or reason as to whether the translation is automatically displayed or whether I have to click to see it. Just now I went to FB to check, and found:

    – my friends' posts in Chinese and Danish were not auto-translated
    – my other friend's post in Russian was
    – for the posts of another friend, who always writes in French, translation was not offered

    (Does this imply that French is too impossible to even try to translate? Or that it is so similar to English that it is not recognized as foreign? Or that it is so easy for English-speaking people to understand French that there is no need to translate it?)

  7. Jenny Chu said,

    April 27, 2022 @ 9:43 pm

    I mean – for Chinese and Danish, the translation option was offered, but I had to click to see it. For French, the option was not available.

  8. v said,

    April 28, 2022 @ 7:02 am

    I didn't even know facebook had "translation".

  9. Cervantes said,

    April 28, 2022 @ 7:09 am

    For the first example, Google gives "Today, when this sovereign government made a promise, it was rhetoric, and what was the result?" which I presume is correct. So it doesn't seem that Chinese is necessarily hard. I'm guessing that "colon inflammation" comes from an idiomatic term comparable to "bullshit."

  10. Dara Connolly said,

    April 28, 2022 @ 4:35 pm

    Jenny, it is possible in Facebook settings to select languages that you understand and that you don't want Facebook to translate in your feed. It is not unfortunately possible to ask Facebook not to translate your posts on other people's feeds. This means that you have no control or visibility of what nonsense is presented to your friends under your name.

  11. Jenny Chu said,

    April 29, 2022 @ 3:20 am

    Dara – oh, very interesting! To my knowledge I have never tried to tell Facebook what to translate for me (or not). Certainly if I had asked, I would have asked it to translate Danish, since that's a language I don't know at all. But FB has a history of making assumptions about what settings it thinks users want …

    Cervantes – yes, the Google translation of that post was very respectable, despite the somewhat flowery/abstract nature of the writing. So yes: somehow Google gets it right but FB abandons all pretense.

RSS feed for comments on this post