March 24, 2008

Austria == Ireland?

In response to yesterday's post about odd transnational substitutions in Google's translations ("Made in USA == Made in Austria|France|Italy|... ?"), Martin Marks writes:

I'm afraid I don't have an answer for your crazy Google mistranslation question, but I do have some even crazier data for you to deal with. On a whim, I "translated" your entry from German to English in its entirety. Most of the entry remains unchanged, with a few weird exceptions. ("German-to-English" is unchanged, for example, but "German-to-French" becomes "English-to-French".) However, one sentence really jumped out at me. You wrote "Of course Austria is not a German word...", but Google translated that as "Of course Ireland is not a Spanish word..."

Madness! Madness! I don't know if I can deal with this.

Of course, the translation program's main fault in this case is not to recognize that it's being asked to translate from German to English a sentence that's already in English to start with:

That said, we're certainly well beyond the simplistic notion that certain place-names are being translated into and out of a sort of interlingual <here>. Given the previously-documented translations between "Austria" and "Australia", I suspected some sort of string-matching in comparable text as the culprit, but I don't think this can help explain "Austria" and "Ireland".

Pekka Karjalainen found another way to get from Austria to Ireland:

German to English translation can also give this strange result. It appears to only result in Ireland when I have three !'s.

Vienna is in Austria!!! ==> Vienna is in Ireland!

I couldn't find any similar phrase that changes so when going from English to another language. This also is the only time I've gotten a result with Ireland yet.

But J.J. Emerson learned by translating the (English) wikipedia entry on Austria from German to English that "The high mountainous alps in the west of Ireland flatten somewhat into low lands and plains in the east of the country", and "The parliament of Canada is located in Vienna, the nation's largest city and capital."

I was just doing a thorough examination of the Google Translation phenomenon when I discovered among other things that Ireland has alps and that the Parliament of Canada meets in Vienna. It looks like I was scooped only by minutes!

I obtained this by feeding the Wikipedia Entry for Austria through the Google translate engine (link).

I've attached a PDF of the whole page in case Google fixes it soon and some choice pngs that are amusing.

I don't understand it either, but it is fun.


[Ron Lee supplies the following:

I think I found another punctuation effect: it seems that "!?!?" often (but not always) evokes a...sense of wonder ("and why?") on the part Google's Machine Translation program (besides Austria sometimes translating to Australia & Indiana to Indianapolis):

1. Is Australia in Austria? ==> Is Australia in Australia?
2. Is Austria in Australia? ==> Is Austria in Australia?
3. Is Austria in Australia!?!? ==> Is Austria in Australia and why?
4. Is Australia in Austria!?!? ==> Is Australia in Australia!?
5. Australia is in Austria!?!? ==> Australia is in Austria and why?
6. Is India in Indiana? ==> Is India in Indianapolis?
7. Is India in Indiana!?!? ==> Is India in Indiana and why?
8. Is India bigger than Indiana!?!? ==> Is India bigger than Indiana and why?
9. Is India not smaller than Indiana!?!? ==> India is not smaller than Indiana and why?
10. Isn't Indiana smaller than India!?!? ==> Is not Indiana smaller than India!?
11. Isn't Indiana bigger than India!?!? ==> Indianapolis Is not bigger than India!?


[Update 3/25/2008: Most (maybe all?) of these oddities have now been fixed. Quick work!]

Posted by Mark Liberman at March 24, 2008 06:20 AM