Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Lorem Ipsum translated by Google

translate.google.ca

Post reply on HN
Print view View original

1–10 of 77 posts

Re: Lorem Ipsum translated by Google

#2

Post by makmanalp »

I was going to jump in and suggest that google's statistical translation methods were being thrown off by lipsum being used in so many strange contexts, but it turns out "Lorem Ipsum" is a mangled nonsensical version of the original text:

http://en.wikipedia.org/wiki/Lorem_ipsum#History_and_discove...

Re: Lorem Ipsum translated by Google

#4

Post by agnokapathetic »

Google Translate uses statistical machine translation [1] seeded from a gigantic automatically curated parallel corpus of similar documents.

As"lorem ipsum" is a typographic placeholder, the filled in version appears appears to have the same document structure (HTML) and would therefore be statistically likely candidates as translatable pairs.

[1] http://www.youtube.com/watch?v=y_PzPDRPwlA

Re: Lorem Ipsum translated by Google

#5

Post by bbx »

Oddly, "Vestibulum ante ipsum primis" translates to "Cisco Security", although taking each word separately translates to "Manufacturing before football first". I don't remember much about my Latin year but if each word has a different meaning depending upon what other word it's combined to, the possibilities are endless.

Re: Lorem Ipsum translated by Google

#7

Post by pepijndevos »

makmanalp wrote: post #2

I was going to jump in and suggest that google's statistical translation methods were being thrown off by lipsum being used in so many strange contexts, but it turns out "Lorem Ipsum" is a mangled nonsensical version of the original text: http://en.wikipedia.org/wiki/Lorem_ipsum#History_and_discove...

The original is translated much better.

Re: Lorem Ipsum translated by Google

#8

Post by mjn »

bbx wrote: post #5

Oddly, "Vestibulum ante ipsum primis" translates to "Cisco Security", although taking each word separately translates to "Manufacturing before football first". I don't remember much about my Latin year but if each word has a different meaning depending upon what other word it's combined to, the possibilities are endless.

Because of Google's approach based strictly on statistical regularities, words can completely change translations based on context even in languages where that wouldn't normally happen, because the contexts can swing the estimates.

One funny one comes with city names, where Google sometimes mistakenly "translates" a city to a different city that happens to have frequent usage in the target language, in contexts that it must find analogous.

For example, here are some translations involving the Danish city Billund (location of Lego), which change even based on punctuation:

   Billund -> Billund
   Jeg er i Billund -> I am in Billund
   Jeg er i Billund. -> I'm in London.
For whatever reason, intriguing place-name translations are particularly common in the Danish->English case. Brøndby is often Red Sox, Odense is Kentucky, and Hillerød is sometimes Whatfield.

Re: Lorem Ipsum translated by Google

#9

Post by jdmitch »

Surely there has been some manipulating of the search results (this non-standard version of) Lorem Ipsum to get those results form google translate. A more standard Lorem Ipsum text comes up with a pretty standard translation (it actually seems to revert to the original Ciceronian text which it is based on).

Re: Lorem Ipsum translated by Google

#10

Post by nknighthb »

mjn wrote: post #8
bbx wrote: post #5

Oddly, "Vestibulum ante ipsum primis" translates to "Cisco Security", although taking each word separately translates to "Manufacturing before football first". I don't remember much about my Latin year but if each word has a different meaning depending upon what other word it's combined to, the possibilities are endless.

Because of Google's approach based strictly on statistical regularities, words can completely change translations based on context even in languages where that wouldn't normally happen, because the contexts can swing the estimates. One funny one comes with city names, where Google sometimes mistakenly "translates" a city to a different city that happens to have frequent usage in the target language, in contexts that…

Now I'm imagining an automated translation of a story from, say, the Star Trek universe to the Star Wars universe, substituting place and character names based on frequency of use in each universe.
Post reply on HN