Fri 13 Jun 2008
Google translate strikes again
Posted by jangari under Google, Korean, Languages, Translation
[10] Comments
I occasionally find myself amused to see in my blog stats that someone has translated my blog into another language. Being so inquisitive, I often follow their lead.
Yesterday morning, I noticed that one of the referring pages was a Google translation of this post into Korean. Naturally, I had a look to see what my blog would look like written in Hangul. As you might expect, it looks really cool, except that I kept noticing a telephone number, the same telephone number, all the way through. Here’s what it looks like:
케빈 Rudd 전화 +852 2907 2112, 자신의 게시물 – 사과 연설에서 거듭 사과를하는 이유는 원상 회복 과정에 필요한는 그들을 처음으로보고 이후에 화해와 일반적으로합니다.
Strangely, each and every time this telephone number appears, it is preceded by the characters 전화, which, according to a Korean-reading friend of mine, means phone, and the whole thing is immediately preceded by Rudd. Looking at the corresponding English of each line (it pops up when you scroll over a line of Hangul), it appears that the phone number is purely being inserted and has no corresponding constituent in the English.
To put this another way, the string of letters Rudd in English, becomes Rudd 전화 +852 2907 2112 in Hangul.
In an attempt to track this a little further to its source, I typed “Rudd” into Google’s translation page, and sure enough, the phone number emerges. This tells me that it’s an artefact of Google translator, and not some mysterious subliminal message that I’ve subconsciously coded into my blog for the sole benefit of Korean readers.
I’m a little discombobulated1 by this, so if you know anything more about this oddity, or could even posit an explanation, I’d love to hear it.
Someone might even like to put their neck on the line and ring the number…
- I’ve always wanted to use that word. ↩

June 13th, 2008 at 7:41 pm
What’s even more intriguing is that it appears to be the phone number for a Hong Kong wine shop that sells ‘fine Niagara wines’ …
June 13th, 2008 at 7:43 pm
What’s even more intriguing is that it appears to be the phone number for a Hong Kong wine shop that sells ‘fine Niagara wines’ …
And what’s more intriguing still is that your blog thinks I’ve already posted this comment when I haven’t pressed submit yet…
June 13th, 2008 at 7:45 pm
Mmmm Niagra wines….*drools*
June 13th, 2008 at 8:00 pm
Ah Claire, you’re too quick!
My housemate just rang it and got the wine shop. I was about to post an update, but you beat me to it.
But we’s getting some fine wine from Hong Kong!
June 13th, 2008 at 8:03 pm
Did I mention that the wine shop in question is actually called Rudd Wine Shop?
Is this a new form of advertising? By stealth on Google Translate?
The plot thickens.
June 14th, 2008 at 12:55 am
I think I figured out some of what’s going on:
In your original article, you have the following quote: “And since the text of Kevin Rudd’s first parliamentary act as Prime Minister…”. I noticed that it contained the word “Rudd”, which you had mentioned was ‘attached’, so to speak, to the phone number, and which also happens to be the name of the wine shop the phone number goes to. So I decided to type it into Google Translate. When you translate just the word “Rudd” (English to Korean), it returns “Rudd 전화 +852 2907 2112″. So I think this is where the translation came from.
As for why it did this, it reminds me of the Google ads in Gmail that try to match what is contained in your current email. Since there is a Rudd wine shop in Hong Kong, and they were translating an entire webpage into Korean, maybe it tried to create a related link and/or advertise for something (?) and they ended up with that number. I also noticed that this does not occur if you try to translate “Rudd” into any other languages, so it seems the translated language and the regions it is often spoken in come into play somehow. Maybe you could try translating other words from a given language into English which would have an associated phone number with them? I tried “Forzieri” from Italian to English, but it didn’t come up with anything interesting.
Anyway, this reminds me of those short murder mysteries… following the lead until you find out “who done it”. =)
June 14th, 2008 at 7:32 pm
An explanation might be that we have encountered a limitation in Google’s implementation of computational translation. My limited understanding of Google Translate is that it looks at two parallel texts each in a different language and then tries to generalise from there when we come along with our translation request. So, to speculate wildly, it may be that they have an English text which refers to the Rudd wine shop without phone number and a parallel Korean text which includes the phone number hence the mistake.
This may be more likely than Google inserting advertisements on the sly as generally they are required and do identify that which is advertisement and that which is not quite clearly.
June 16th, 2008 at 9:23 am
K. Smith and Simon, on reflection I think it’s something else entirely. And unfortunately it’s probably much more mundane than either covert epenthetic advertisements or sophisticated corpus translation technologies gone wrong. But thanks for the suggestions all the same.
Firstly, Korean is not spoken in Hong Kong; not commonly anyway. So I doubt Google Translate would insert this advert for a translation from English to Korean, but not for a translation from English to Mandarin or Cantonese. Furthermore, as Simon points out, Google always identify paid content as advertising, so this should be no different.
Secondly, for the overgeneralised corpus technique of translation to be the culprit, you’d have to say that there is some English corpus text somewhere that has a corresponding Korean text, wherein, if the English version has the string Rudd, then the Korean version has the string Rudd 전화 +852 2907 2112. I think this is unlikely because, first, surely the English text would similarly have the phone number of the wine shop – it’s more of an English speaking region than Korean – and second, Google’s sophisticated algorithms would probably throw up a red flag when a single word in English corresponds in Korean with the very same word, two Korean words, and a phone number complete with international telephony prefix.
What I think is going on is much less exciting. Notice at the bottom of the Google Translate page is a javascript link reading Suggest a better translation. Clicking it brings up a text box where you type in a better translation that the one that Google is giving you. Next to the box is this description:
Would Google sift through probably millions of translation suggestions every day and make a judgement for each? Not likely. I think what is likely is that if the exact same string gets suggested as a translation for some other string, a certain number of times, then Google’s algorithms will accept it.
Ergo, Rudd Wine Shop has diligently and deliberately inserted this ad into Google’s highly fallible system.
June 16th, 2008 at 6:52 pm
Then why only for Korean?
June 17th, 2008 at 9:31 am
Yeah Wamut, good point.
I guess Korean is the only language (found so far) for which Rudd Wine Shop have been successful…
Maybe there’s far too much material in Cantonese and Mandarin that provides better translations for “Rudd” and as such, the system won’t be so amenable to suggestions from the public.