Protocols of the Quizmasters of Zion
« previous post | next post »
I can usually figure out the specific reasons that statistical MT systems come up with peculiar translations. But this one has me stumped:
Google Translate renders "Montar un vínculo con Israel" into English as ""Build a bond with Israel", which seems accurate. But how did Israel slip into the English-to-Spanish mapping?
"A trivia quiz" comes out as "un concurso de preguntas". And "Assemble a quiz" comes out as "Prepare un cuestionario". Even "Assemble the trivia quiz" yields "Monte el concurso de preguntas".
A plain Google web search for "assemble a trivia quiz" yields nothing that seems likely to have led Google Translate astray. Nor was I able to find any other language for which Google Translate renders "assemble a trivia quiz" with a reference to Israel. If I didn't know better, I'd suspect an easter egg slipped in by some mischievous Googlite.
Can anyone solve this mystery?
[Hat tip to Adrian Morgan]
John Cowan said,
August 31, 2010 @ 8:32 am
A few observations:
English->Catalan gives "armar un vincle amb Israel", which may only mean that it's really English->Spanish->Catalan under the covers.
I went through all the other available languages in Latin, Cyrillic, Greek, and Hebrew scripts (the ones I can make out), and none of the translations mention Israel.
Changing the verb doesn't affect the anomaly either: "start a trivia quiz" becomes "iniciar un vínculo con Israel", and even "fnord a trivia quiz" produces "fnord un vínculo con Israel".
"A trivia quiz is very difficult" comes out right, whereas "It is very difficult to do a trivia quiz" tickles the bug, so I don't think it's an Easter egg.
Matthew Baerman said,
August 31, 2010 @ 8:33 am
Putting quotes around the phrase yields the correct(-ish?) translation, leaving them out yields the funny one.
Kevin said,
August 31, 2010 @ 8:47 am
… but putting "a trivia quiz" on its own in quotes also triggers the bug.
Erik said,
August 31, 2010 @ 8:53 am
I think I've found the culprit. Searched Google for +"trivia quiz +"vinculo con Israel".
http://it-it.facebook.com/pages/The-official-Jack-Irving-Plum-Page/388961269803?v=info
In English:
you'll be getting weekly wall postings of stuff such as Xena question Of The week, Beatlemania trivia quiz, my doo-wop moment of the week, my Chicago Gold Moment Of The week my Chicago Food Stop Of The Week, my weekly Knowledge Of The 60's, The Adventures Of Jack & Trixie, starring me & my cartoon sidekick, Trixie The Waitress and what have you.
In Spanish:
obtendrás un muro de anuncios semanales de cosas tales como Xena pregunta de la semana, Beatlemania vínculo con Israel, mi doo-wop momento de la semana, mi Chicago Oro momento de la semana mi Chicago Food Stop de la semana, mi conocimiento de semana Los años 60, las aventuras de Jack & Trixie, protagonizada por mí y mi compañero de la historieta, Trixie la camarera y lo que tienes.
Dick Margulis said,
August 31, 2010 @ 8:55 am
Okay, this is probably way too far-fetched, but I'll take a stab at it anyway.
Google pretty much ignores punctuation on the search side, and I'm not sure it pays a lot of attention to it on the translate side. If "Iz" is taken to be an [English-language] nickname for Israel, and if you stick an apostrophe in there to get qu'Iz, does that get you any closer to what might be going on?
Dick Margulis said,
August 31, 2010 @ 8:57 am
And by "nickname for Israel," I meant the given name Israel, not the country.
Erik said,
August 31, 2010 @ 9:01 am
Or, uh, this could be a [i]result[/i] of using Google Translate. I guess I should've thought of that first.
Daan said,
August 31, 2010 @ 9:04 am
I don't think that's likely to be the culprit, Erik. The Spanish does not make a whole lot of sense, so I think someone translated that English message into Spanish using Google Translate and posted it without noticing the mistake.
But this really is quite baffling.
Ginger Yellow said,
August 31, 2010 @ 9:06 am
"Google pretty much ignores punctuation on the search side, and I'm not sure it pays a lot of attention to it on the translate side."
I think it depends what sort of punctuation you're talking about. It certainly pays attention to full stops and, annoyingly, line breaks. Which makes it a pain in the arse to cut and paste an article into Translate. The quality (certainly for German and Spanish into English) is notably higher when you manually remove the line breaks.
Ginger Yellow said,
August 31, 2010 @ 9:07 am
As to the OP, could this be an artefact of Google's "suggest a better translaiton" feature?
Matt said,
August 31, 2010 @ 9:09 am
Erik, the translation is very similar to the Facebook Spanish; I think it's probably translated.
It seems to be triggered by a verb just prior to "trivia quiz" – "see trivia quiz" gets it, as does "take a trivia quiz," "freeze a trivia quiz" etc. But it doesnt' bite when using "take one trivia quiz" or "take the trivia quiz". It does kick in when there's an adjective in there too, like "freeze a fast trivia quiz", but not for multiple strings of adjectives like "freeze a fast hairy trivia quiz"
I'm totally losing my lunch hour to this.
Matthew Baerman said,
August 31, 2010 @ 9:20 am
John Cowan: "Changing the verb doesn't affect the anomaly either"
Using 'win' does.
Rick S said,
August 31, 2010 @ 9:32 am
In cases like this, I suspect the "Suggest a better translation" feature on Google Translate. It must be that Google uses these suggestions to train its MT, right? If we suppose that somebody, at some time in the past, submitted a very loose "better" translation of a news story containing these exact phrases, and if we further suppose that the suggested "better" translations aren't vetted, then an erroneous correlation will be the result. For short phrases that occur repeatedly throughout the training material, that's not a problem because the predominance of other correlations will outweight it, but for phrases long enough to be unique, that may not happen. It depends on the algorithm for choosing among alternatives, and such algorithms are complex enough that their behavior is seemingly non-stochastic at some nodes. The moral, of course, is: Never trust MT if you know nothing about one of the languages.
Adrian Morgan said,
August 31, 2010 @ 10:57 am
In case anyone's interested in how I stumbled upon the error in the first place, it was by watching the Referrers section of my blog stats. Someone arrived on my blog from a translated version of one of my blog posts (one that mentions trivia quizzes), and when I clicked the link I noticed the reference to Israel.
(If you want you can actually do the quiz that my blog post was about. Two of the questions, out of fifty, are linguistics-related; the rest are about other sciences.)
Adrian Morgan said,
August 31, 2010 @ 11:12 am
Followup to my previous comment: that should read "out of sixty". My excuse is that I'm writing from Australia where it's coming up on 1 am.
groki said,
August 31, 2010 @ 12:53 pm
a google of +"trivia quiz" +"link with Israel" (translating "vínculo" as "link") finds some pages (10 when "including omitted results") that reference trivia quizzes, mostly on sports.
on these pages, the phrase "link with Israel" shows up oddly, where I would expect the phrase "trivia quiz(zes)" instead.
I don't know if this is cause or effect of the google translation issue.
NW said,
August 31, 2010 @ 12:54 pm
This turns it off:
"a trivia quiz" and
This turns it on for the first and off for the second:
"a trivia quiz" and "a trivia quiz"
groki said,
August 31, 2010 @ 12:59 pm
addendum: to be clear, I only checked a few of the pages themselves. but even in the summaries on the google results page, "trivia quiz(zes)" seems to fit better than "link with Israel."
Eric 0ner said,
August 31, 2010 @ 12:59 pm
Rick S is right on; the "suggest a better translation" thing gets you some awesome howlers sometimes, and as noted it's interesting how altering one insignificant word (such as "the" to "a") can completely change the phrase Google Translate provides.
Also, yes, under the hood pretty much everything in Google Translate goes through a layer of English at some point, though I didn't know that if you want to go to Catalan you have to pass through, instead, Spanish.
Update: Google seems to have ELIMINATED "Suggest a better translation!"
Andrew Greene said,
August 31, 2010 @ 1:11 pm
It recognizes the passive: "trivia quiz is had" comes out "vínculo con Israel se tiene".
Peter Taylor said,
August 31, 2010 @ 1:24 pm
As a point of interest, the Spaniards with whom I play in a pub quiz just borrow the word directly from English as "el quiz". This may well be influenced by the fact that the quiz itself is (usually) in English and intended as an opportunity for native English and Spanish speakers to mix and practice the other language.
army1987 said,
August 31, 2010 @ 4:50 pm
It also famously translates Italian "torino" as English "london".
Gordon Campbell said,
August 31, 2010 @ 9:41 pm
Protocols of the Quizmasters of Zion! That heading wins the Internet for today. And (quick scan through the comments…good, no one has said it yet) a great name for a band.
Luenduen Ren said,
September 1, 2010 @ 10:43 am
It correctly detects Torino as Italian but translates it into London/Londres/ロンドン (London)/倫敦 (Lundun) in ALL languages! What a glitch! How did you find this out?
Dave said,
September 1, 2010 @ 12:08 pm
Eric 0ner: The "suggest a better translation" feature still exists — with a browser with Javascript turned on, hover the mouse over the translated text and the suggestion feature pops up.
Will said,
September 1, 2010 @ 6:10 pm
I spent about 20 minutes googling to try to figure this out but made no headway. I would like to point out though that I think the most likely translation of "vínculo con Israel" would be "ties to Israel", rather than link or bond.
Andrew said,
September 1, 2010 @ 8:08 pm
Google does some very interesting things with their data-mining to feed the translation engine, by looking for documents that exist in multiple languages and analyzing them in parallel — for instance, the only way that I can think of that it would equate "Torino" with "London" is from sheer weight of repetition of "Olimpiadi Torino" and "London Olympics" on what it decided were similar pages (not similar enough, as it turns out).
Since "Assemble a trivia quiz" is a much rarer phrase, it probably takes a much smaller weight of evidence to confuse the engine — it could be that there was just *one* multilingual page on the 'net that google crawled, that described the same task as "build a bond with Israel" in Spanish but "assemble a trivia quiz" in English. Weird, but not impossible.
Helmut said,
September 2, 2010 @ 11:53 am
It looks like every language that has a word for word translation of the words "trivia quiz" gets the bug. English to German, Czech, Italian … to Spanish = Israel. French gets jeu-questionnaire, Swedish gets frågesport, and thus no bug. Using "Pris un jeu-quesstionnaire" there's no bug, but "Pris un info quiz" gets it. Could just be using English as an intermediary.
Nijma said,
September 5, 2010 @ 3:45 pm
A long shot, but maybe it has something to do with QIZs (Qualified Industrial Zones).
Ellie said,
September 6, 2010 @ 10:08 am
Is it just me, or is the bug gone?
groki said,
September 6, 2010 @ 2:50 pm
it's gone for me too now. for example:
"fnord a trivia quiz" gives "fnord un concurso de trivia".
"a trivia quiz" still gives "un concurso de preguntas" and
"trivia quiz" by itself returns "concurso de preguntas".
ahh, the flickering evanescence of teh goog: another metaphor of impermanence for the poets.
Interesting stuff: Early September 2010 « The Outer Hoard said,
September 6, 2010 @ 10:06 pm
[…] sent an email to Mark Liberman that led to a Language Log post, concerning a particularly bizzare Google Translate result that I stumbled upon by watching my blog […]
j-g-faustus said,
September 9, 2010 @ 6:00 am
Another curious translation: Malay "Lady Gaga" apparently translates to French "Britney Spears" – http://gawker.com/5630225/this-is-what-happens-when-you-translate-lady-gaga-with-google-translate
The explanation suggested is that the translation is purely (or mostly) statistical, and that this error could stem from matching lists of bands in two different languages, where the lists are similar but not quite the same. (Say one list has "Michael Jackson" and "Lady Gaga" while the other list has "Michael Jackson" and "Britney Spears"; the algorithm matches them up and concludes that "Britney Spears" is the translation.)
Frans said,
September 19, 2010 @ 6:05 am
While slightly less of a switch in category, until not too long ago (half a year?) Google Translate would translate place names like Amsterdam to New York, Utrecht to Montreal, Rotterdam to Sydney, etc., when translating from Dutch to English. I won't vouch for the specifics of the "translations" I mentioned, but you get the point. Suffice it to say that the results could be hilarious.
@army1987:
It seems to have stopped doing that, although now I'm curious whether examples of similar "translations" of place names may still be found.
[(myl) See "Made in USA == Made in Austria|France|Italy", 3/23/2008; "Austria == Ireland?", 3/24/2008; "Why Austria is Ireland", 3/24/2008; "The (probable) truth about Austria and Ireland", 3/24/2008.; etc.]
Frans said,
September 19, 2010 @ 12:53 pm
Thank you. However, I can assure you that I didn't run into this quirk until late 2008 or early 2009, so apparently it hadn't been fixed for Dutch-English and vice versa. "The (probable) truth about Austria and Ireland" says that the problem was already fixed on 3/25/2008, long before I ever encountered the effect in the first place.
Google Translate Mistakes | ChromeBytes said,
March 28, 2013 @ 10:33 am
[…] English to Spanish: quiz trivia -> vínculo con Israel […]