The sliced raw fish shoes it wishes

« previous post | next post »

The crash-blossom-y headline that Geoff Pullum just posted about, "Google's Computer Might Betters Translation Tool," has been changed in the online edition of The New York Times to something more sensible: "Google’s Computing Power Refines Translation Tool." The headline in the print edition, says LexisNexis, is "Google Can Now Say No to 'Raw Fish Shoes,' in 52 Languages." This is a typical example of the gap between oblique print headlines and their more straightforward online equivalents designed with search engines in mind. (See the April 2006 Times article, "This Boring Headline Is Written for Google.")

But enough about the headline: the article itself is worth reading (and has a quote from Language Log's own Philip Resnik). The headline in the print edition refers to the article's anecdotal lead:

In a meeting at Google in 2004, the discussion turned to an e-mail message the company had received from a fan in South Korea. Sergey Brin, a Google founder, ran the message through an automatic translation service that the company had licensed.

The message said Google was a favorite search engine, but the result read: “The sliced raw fish shoes it wishes. Google green onion thing!”

Mr. Brin said Google ought to be able to do better. Six years later, its free Google Translate service handles 52 languages, more than any similar system, and people use it hundreds of millions of times a week to translate Web pages and other text.

I don't know what sort of source text in Korean might have generated the failed translation Brin mentions, but the "sliced raw fish shoes" problem has apparently been a longstanding issue for Korean-English MT. In 2007 the blogger Karl Heinz Kremer described running a Korean-language message from Microsoft through Babelfish (apologies if the Korean version looks like mojibake in your browser):

본 메일은 2007년 12월 20일 기준으로 당사의 메일을 수신 동의하신 고객 분들에게만 발송되는 메일입니다.메일 수신을 원치 않으시면 제목란에 "UNSUBSCRIBE"라고 적으신 후 회신하여 주십시오.또한 프로필 센터를 통하여 뉴스레터에 대한 모든 구독 관리를 하실 수 있습니다.주소: 서울특별시 강남구 대치동 892번지 포스코센터 서관 5층 (우편번호 135-777)

Here is the translation: “The mail which it sees in 2007 December 20th standard the mail of theheadquarters of a party the reception is the mail which is sent out atonly the customer minutes which agree. Unit is not and subject is “asUNSUBSCRIBE” after writing, the sliced raw fish shoes to do the mailreception. Also pro there is a possibility of doing all subscriptioncivil official the news letter the center where it will bloom leadsand against. Address: Seoul Kangnam Ku confrontation eastern 892 housenumber guns su from nose center tube 5 layer (postal code 135-777)”

Another blogger wrote of getting the Babelfish result, "The hour is busy with relationship of pressure one but shear mail sliced raw fish shoes entrusting under confirming it gives rightly. Thanks it gives in cooperation." And an automatically translated love letter on Yahoo! Answers includes the ineffable "…like the like that thing the branch doing against the route which is not the after sliced raw fish it does not want."

Anyone proficient in Korean want to get to the bottom of the sashimi-shoe conundrum?



36 Comments

  1. JS Bangs said,

    March 9, 2010 @ 1:25 pm

    Why does that Korean text look like mojibake? I've got an up-to-date version of Firefox, which should be able to handle Unicode just fine.

    [(bgz) I've replaced the mojibake with the properly encoded characters provided by codeman38 downthread.]

  2. Eugene van der Pijll said,

    March 9, 2010 @ 1:38 pm

    Some experimentation with Babelfish (which still has the "raw fish" problem) and Google (which hasn't), shows that the single character "회" means "sliced raw fish", and "회신해" (which starts with that character) means "reply".

    Apparently, older translation programs think it's a compound word.

  3. db48x said,

    March 9, 2010 @ 1:39 pm

    It looks like mojibake because the page as a whole is encoded is UTF-8, but the Korean quote is not. It's been pasted in from some non-unicode source and not properly translated.

  4. Ginger Yellow said,

    March 9, 2010 @ 1:42 pm

    Can't help with the sashimi shoes, but what impessed me about the crash blossom was how avoidable it was. As a web headline, there was no (or less, anyway) space pressure. And yet the writer chose not one but two words – "might" and "betters" – which are both prone to ambiguity and arguably stylistically suspect for a news story. How did this person think "betters" was, well, better than "improves"?

  5. Eugene van der Pijll said,

    March 9, 2010 @ 1:45 pm

    Addition to my previous post: I got the word "reply" from a similar case, which I found here. I don't know if this is the same word which is in Karl Heinz Kremer's message.

    And the raw fish dish is called Hoe (회).

  6. Craig said,

    March 9, 2010 @ 2:16 pm

    I'm not sure I believe the original text makes sense in any encoding. However, that being said, I switched the encoding to EUC-JP and put the results through Google Translate and got:

    Badger Badger breath Valley Village Valley 2007 Valley continue existing goods 12 Badger Party groove measuring 20 serous Valley Village Valley Village shelf reinforced existing genus Badger Badger Valley Village groove cod Badger Badger Badger Badger Badger Valley ∨ breath forming 其谷 Badger Badger Valley genus groove shelves shelves Continued Tani Tani rolling breath 臓谷 Badger Badger Badger genus village valley valley family law groove Amitani γ shelf. Tanuki Tanuki breath Valley Village translocation 捉息 Valley Village Valley Badger Badger Badger Badger graduated grandchildren breathe hard valley seat Badger Badger Badger Badger groove Tanimura UNSUBSCRIBE Badger House Foundation cod cod cod groove groove groove ∨ shelves continue Badger Badger Badger Badger Village Village reinforced meters gifts Badger Badger Badger hardness measurement. pollack Valley ≒ cod cod cod sleeves 促村 臓鱈 Valley Valley Village genus cod cod genus ∨ Badger Badger measure Tani Yasushi 潅 造谷Badger Valley № graduated shelves shelves sleeves measure Valley Valley Valley 捉即 ∨ β Dji cod rolling groove Badger Badger Badger Badger Valley Amitani sleeve groove. Tanuki Tanuki gift from the village: valley measuring sleeve shelves shelves immediately species existing Hard House Badger Badger Badger continue 孫谷 cod 892 Amitani Lu graduated shelves Badger Badger Badger Village genus cod cod steel measuring rule Badger Badger Badger tribe Valley 5 existing sleeves (Badger House existing embedded cod cod Valley Tribe Genus existing 135-777)

    I was just waiting for "a snake, a snake" to appear in that mess. :)

  7. Will said,

    March 9, 2010 @ 2:21 pm

    Following up on Eugene van der Pijll's test, I decided to play around with google translate a little more (not knowing anything about Korean), and it appears that only the first two characters in that sequence (회신) are required for "reply".

    And the second character in that "reply" sequence (신) has about two dozen meanings, but the most common one according to Google is "footgear".

  8. Daniel von Brighoff said,

    March 9, 2010 @ 3:09 pm

    Those of you better versed in another East Asian language might have an easier time recognising the components of 회신 in the hanja spelling, 回信.

    Now that we've resolved that, what about this "green onion thing"? Korean for "green onion" is 파, but I'm stumped as to which of the dozen or so Korean words which can be translated as "thing" forms the second element.

  9. Kutsuwamushi said,

    March 9, 2010 @ 3:17 pm

    English-Korean-English has always been my path to the most hilariously Babelfished text. For some reason, the Korean translation is simply worse than any of the others, even the languages that are just as foreign to English. (English-Japanese-English, for example, can be lol-worthy, but you can often understand the gist of it. With Korean-English-Korean, that's rare.)

    Anyway, to add to what others have said:

    해 is a form of the verb 하다, which often appears in compound verbs. It's one of the most common ways to form a verb from a noun. So it's likely that in "raw fish shoes," 해 isn't contributing to the error.

    My dictionary doesn't have 회신하다, but it does have 회서 meaning "a reply." It says that 회신 means "return postage," or ashes.

    I've checked more than one dictionary and haven't found 회신하다 yet.

  10. Daniel von Brighoff said,

    March 9, 2010 @ 3:24 pm

    Fancy seeing you here, Kutsuwamushi!

    Do you know the Nate (네이트) dictionaries? Their definition for 회신 is "편지(便紙), 전신(電信), 전화(電話) 따위의 회답(回答)", which I would say corresponds well to that of English "reply" in this context.

  11. Michael said,

    March 9, 2010 @ 3:36 pm

    Here's a link to a Google translate page:

    http://is.gd/a3QPe

    If you copy/paste the Korean into this page:

    http://babelfish.yahoo.com/translate_txt

    you can see an example of the "sliced raw fish" issue. I found the Korean text in the middle of a larger block of Korean (followed by a bad English translation) here:

    http://goo.gl/zZoc

  12. Esther Chung said,

    March 9, 2010 @ 3:52 pm

    If you would allow me:

    1. 회신 (回信) can indeed be used to mean "reply", as in "회신해주시길 바랍니다", which translates loosely to something like "A response would be appreciated" or "We would appreciate a response". With that said, "회신으로 화하다" means "to be reduced to ashes or burnt to the ground".

    2. The original letter to Google seems to have read something like "회신해주시길 바랍니다 (or some variant). 구글 파팅!" "파이팅" (which returns "pie thing"), from English "fighting" (via Japanese, I believe), is a word of encouragement. It seems to have been abbreviated.

    3. The relevant word in the "love letter" (it actually isn't a letter of any kind, and certainly not a love letter) is 후회 (後悔), "regret". You will notice that this word also includes the culpable character 회, which translates to "sliced raw fish".

  13. Jennifer said,

    March 9, 2010 @ 3:53 pm

    "To fish" can also mean "to search," and there is a connection between "shoe" and "invention" (footwear + name = "new invention" according to bluedic.com). Also, consider the path of "raw" to "fresh" to "new."

    Although I haven't been able to replicate this on Google translate, I can see how a few passes back and forth between Korean and English in a translator could get you from "search technology" (a perfectly reasonable description of Google) to "sliced raw fish shoes" (a fashion monstrosity).

  14. Bokai said,

    March 9, 2010 @ 4:05 pm

    Babelfish is a source of infinite fun, whether you're into linguistics or not. Not sure if you're familiar with this site, but it translates a phrase into Japanese, then back to English, then back again, until the translations hit an equilibrium. Sometimes they hit an infinite loop instead.

    http://www.translationparty.com/

  15. Jerry Friedman said,

    March 9, 2010 @ 4:47 pm

    Heh.

    Numbers appear to lead to disaster. I tried "Two roads diverged in a wood, and I took the one less traveled by." This ended up at

    The two modes of time: I have one one of 1,222,211,111,111,111 in one or two trees, two branches have one is 11231.

    It is doubtful that this phrase will ever reach equilibrium.

  16. Richard Howland-Bolton said,

    March 9, 2010 @ 4:52 pm

    Several years ago for one of my silly radio things I did the (presumably apocryphal) "Out of sight out of mind" into Russian and then back to English as "Invisible imbecile" machine translation repeatedly for a variety of languages. They mostly went to an 'end point' with the translation not changing (apart from the French series which, of course, ended with “hopeless, nil, non-existent” which I think was less of a perceptive translation than the engine giving up in typical French disgust).
    The jolliest result was Chinese, where the English side went:

    Out of sight out of mind

    Stemming from sight outside brains

    Source to sight outside brain

    The origin sees exterior brain
    The essay is here.

  17. Will said,

    March 9, 2010 @ 6:47 pm

    I tried translationparty on Donald Rumsfeld's famous epistemological quote, and the equilibrium came out surprisingly faithful.

    Original:

    "There are known knowns. These are things we know that we know. There are known unknowns. That is to say, there are things that we now know we don’t know. But there are also unknown unknowns. These are things we do not know we don’t know."

    Translationparty Equilibrium:

    Known knowns. These are known to us. Known unknowns. In other words, some I do not know what it is. However, it is also unknown unknowns. Some of these I do not know I do not know.

    I think if Rumsfeld had stated it that way, the Plain English people might have liked it more.

  18. Eli Morris-Heft said,

    March 9, 2010 @ 6:50 pm

    I have always found translation from Japanese and Korean into English by Babelfish and Google Translate to be fraught with peril. Though things are getting better, it's hard to read when every noun ends up with 'the' in front of it and every sentence or clause begins with 'it is the'. The weird thing is, there's nothing in Japanese or Korean to promote these translations, at least that I have encountered in my 7+ years of study.

  19. Will said,

    March 9, 2010 @ 6:55 pm

    And to test Jerry Friedman's theory of numbers leading to disaster, I tried "A bird in the hand is worth two in the bush", and it actually reached equilibrium!

    One 12:59 1 2 3 1 hour, 1 block, the first two minutes, the hand is worth two birds

    But yes, definitely a disaster!

  20. Nanani said,

    March 9, 2010 @ 9:03 pm

    @Will

    I suspect "it is the" comes from the は particle in Japanese. I don't know about Korean, though.

  21. Joseph Dart said,

    March 9, 2010 @ 10:01 pm

    I tend to find I get marginally better translations by

    1. Running Korean text through a third-party Korean -> Japanese translator first
    2. Then feeding it to Google's Japanese -> English translator

    Try Excite.co.jp's service for step 1. Google, irritatingly enough, uses English as its pivot language for translation between Japanese and Korean, resulting in double the normal amount of mangling.

  22. Dan Bloom said,

    March 9, 2010 @ 10:40 pm

    I don't know about this Korean thing, but the Chinese character for "crisis" — can it also mean "opportunity" as many Western pundits and oped writers and reporters often say? NO. This is an Western/Asian myth: there are two different words and characters for Crisis and Opportunity, and they are not the same. The West needs to be disabused of this falsehood. However, there is an old saying in Chinese to the effect that "Every crisis also has in its the seeds of an opportunity" — perhaps this is where the confusion comes from….

    [(bgz) See this post for some of the history of the "crisis = danger + opportunity" and "crisis = opportunity" memes.]

  23. Nicki said,

    March 9, 2010 @ 11:17 pm

    More number disasters:

    ninety-nine bottles of beer on the wall, take one down, pass it around

    becomes

    99 1 1 1 1 1 1 wall, this issue, two, three beers, two of three instructions, the two paths, the first 2 1 2 Example 1,2,1,1,1,1 , one of the first one or two – to avoid the first two days.

    It is doubtful that this phrase will ever reach equilibrium.

  24. Clarissa at Talk to the Clouds said,

    March 9, 2010 @ 11:43 pm

    Craig: I actually did laugh out loud.

    Esther Chung: Pie thing! Good work.

  25. Chas Belov said,

    March 10, 2010 @ 3:24 am

    It might help if the Korean paragraph had a p tag with the attribute lang="ko"

    In Firefox, in the View menu, I chose Character Encoding then More Encodings then East Asian, then tried the four Korean encodings in turn. 3 of the 4 produced some, but not all, Korean characters, with (JOHAB) encoding producing different results from (EUC-KR) or (UHC). All three gave their fair share of question marks, and (EUC-KR) or (UHC) disconcertingly had a few Japanese kana, which I believe is not a feature of Korean. But I don't know Korean beyond how to pronounce some of the Hangul characters, so I don't know which encoding might be correct, if any.

  26. Eric Vinyl said,

    March 10, 2010 @ 3:51 am

    Well, it’s not gonna work if you look at this page in different encodings—as stated, it’s already mis-encoded in UTF-8. However, if you save those characters as ISO-8859-1, and then view that as Unicode, you get:

    본 메일은 2007년 12월 20일 기준으로 당사의 메일을 수� 동의하� � 객 분들에게만 발송되는 메일입니다.메일 수� 을 원치 않으시면 � �목란에 �UNSUBSCRIBE�라� � �으� 후 회� 하여 주십시오.또한 프로필 센터를 통하여 뉴스� �터에 대한 모� 구독 관리를 하실 수 있습니다.주소: 서울특별시 강남구 대치동 892번지 포스코센터 서관 5층 (우편번호 135-777)

    which, unfortunately, I can’t view on this computer as there are no CJK fonts installed, but Google Translate autodetects it as Korean.

  27. A-gu said,

    March 10, 2010 @ 4:00 am

    If I may say so, Esther Chung's comment has been most helpful to me, especially considering the use of hanja, which I as a speaker of Chinese can actually read. Perhaps we can increase the hangul/hanja combo usage on this thread and get some better results as various expertises converge.

  28. Joseph Dart said,

    March 10, 2010 @ 5:30 am

    @a-gu: here's the message with all the Sinitic and English words clarified. What remains are mostly just particles and verbs, especially light-verb (hada) constructions.

    本 mail은 2007년12月20日 基準으로 當事의 mail을 受信 同意하신 顧客 분들에게만 發送되는 mail입니다. mail 受信을 願치 않으시면 題目欄에 "UNSUBSCRIBE"라고 적으신 後 回信하여 주십시오. 또한 profile centre를 通하여 newsletter에 對한 모든 購讀 管理를 하실 수 있습니다.
    住所: 서울特別市 江南區 大峙洞892番地 POSCO Center 西館5層 (郵便番號 135-777)

    The Korean-Japanese translator link I posted above is also helpful when you want to do this, since the translation tends to preserve the word order pretty well (so you can match the kanji in the Japanese translation to the hangul in the original)

  29. Mary Kuhner said,

    March 10, 2010 @ 1:36 pm

    The result of running Translation Party on the last lines of Eliot's _The Love Song of J. Alfred Prufrock_ is remarkably appealing. It never settles down, the meaning changes drastically, but many of the results are highly evocative and poetic.

    (If you want to try it:

    We have lingered in the caverns of the sea/ By sea-girls wreathed with seaweeds red and brown/ Till human voices wake us and we drown.)

  30. codeman38 said,

    March 10, 2010 @ 5:06 pm

    @Eric: Here's a better re-encoding, with significantly fewer "unknown" characters, which I derived by copying the text directly from the HTML of KHK's original post:

    본 메일은 2007년 12월 20일 기준으로 당사의 메일을 수신 동의하신 고객 분들에게만 발송되는 메일입니다.메일 수신을 원치 않으시면 제목란에 "UNSUBSCRIBE"라고 적으신 후 회신하여 주십시오.또한 프로필 센터를 통하여 뉴스레터에 대한 모든 구독 관리를 하실 수 있습니다.주소: 서울특별시 강남구 대치동 892번지 포스코센터 서관 5층 (우편번호 135-777)

    [(bgz) Thanks for this — I've replaced the mojibake in the post.]

  31. speedwell said,

    March 10, 2010 @ 5:29 pm

    Wow, translation party… what a ride!

    I used I talk to the trees, but they don't listen to me; I talk to the stars, but they never hear me; the breeze hasn't time to stop and hear what I say; I talk to them all in vain. (From the musical Paint Your Wagon.)

    I got However, if I waste wood, 闻, I have never heard of to stop listening to the wind, all star shoes.

    There's those shoes again.

  32. M. Packman said,

    March 11, 2010 @ 11:55 am

    The Thai translations are pretty hilarious as well. Yet no matter how many times I explain that curry powder is not the same as whore dust and English constructions like 'going to go' would be rendered in ridiculous Thai if the computer did it, the darn kids keep using it for their assignments.

    If I had sliced raw fish shoes, though, I'd be more cold-hearted — I mean, calm. Curse these philistine translations!

  33. M. Packman said,

    March 11, 2010 @ 1:00 pm

    I Translation Partied my grandfather's catchphrase/theme song. I like bananas because they have no bones became I like the bones of the banana. That's surreal enough, but the Japanese is even better:

    私はバナナの骨のように。

    I'm dying to find out what the verb is. What can I do like the bones of a banana, anyway? What can they even do on their own, those lonesome, unappealing bananabones?

    Or, as it is, it could be a creative writing prompt: I am like the bones of a banana because ____________.

  34. CWV said,

    March 11, 2010 @ 10:12 pm

    Another fun crash blossom: Inflation Eroding China Deposits Feeds Asset Pressure.
    (http://www.businessweek.com/news/2010-03-11/inflation-eroding-china-deposits-feeds-asset-pressure-update1-.html).

  35. Graham Asher said,

    March 14, 2010 @ 4:00 pm

    Translation party is the funniest thing since Granny got her tit caught in the mangle, as they say. It manages to reverse the sense of Browning by bringing bad news from Ghent to Aix (or was it Aix to Ghent?): 'I galloped, Dirk galloped, we galloped all three' somehow ends up as: "3 Gyaroppugyaroppudaku 1,1, I need to execute people."

  36. kimchikraut said,

    March 21, 2010 @ 4:05 pm

    This is a possibility:
    "회신을 바랍니다. 구글 파이팅!"
    http://news.chosun.com/site/data/html_dir/2010/03/11/2010031100062.html

    Today, Google translate returns:
    Please reply. Google Fighting!

RSS feed for comments on this post