Not logged in
Log in now
Create an account
Subscribe to LWN
LWN.net Weekly Edition for May 16, 2013
A look at the PyPy 2.0 release
PostgreSQL 9.3 beta: Federated databases and more
LWN.net Weekly Edition for May 9, 2013
(Nearly) full tickless operation in 3.10
Linux at the end of the world (our 2012 predictions)
Posted Jan 4, 2012 14:13 UTC (Wed) by corbet (editor, #1)
Posted Jan 4, 2012 17:29 UTC (Wed) by wookey (subscriber, #5501)
Posted Jan 4, 2012 21:13 UTC (Wed) by job (guest, #670)
In order to get a feeling what Translate produces, try using it on closed loop of languages (ie. going back to your native) and see. That's roughly what you intend to send. (Yes, I've received auto translated texts more than one time. The first time I could at least have a good laugh about it.)
Unfortunately it's not so simple...
Posted Jan 4, 2012 21:38 UTC (Wed) by khim (subscriber, #9252)
In order to get a feeling what Translate produces, try using it on closed loop of languages (ie. going back to your native) and see.
This is deeply flawed idea.
That's roughly what you intend to send.
Not really. Every translation loses piece of original. No matter who does it. After few translations you can get entirely different text. If you are doing machine translation then it'll lose some of the original meaning and get some random noise instead. If it's human translation then it'll keep about the same amount of original but the rest will not be a random nose - it'll include personality of people who did the translation.
The great example is the tale of "Ein Gleiches" which was translated in 1902 to Japanese, then in 1911 to French and then shortly after that back to German.
Über allen Gipfeln
In allen Wipfeln
Kaum einen Hauch;
Die Vögelein schweigen im Walde
Warte nur, balde
Ruhest Du auch.
Tranlation of translation of translation:
Stille ist im Pavillon aus Jade
Krähen fliegen stumm
Zu beschneiten Kirschbäumen im Mondlicht.
The very fact that the last German translator had no idea that he's back-translating quite famous work of Goethe should say you something...
Posted Jan 5, 2012 12:11 UTC (Thu) by job (guest, #670)
A reasonably intelligent human with a dictionary fares much better. Your German example, while being a clear example of the difficulties involved, is still parseble in the end.
Posted Jan 8, 2012 20:52 UTC (Sun) by csawtell (subscriber, #986)
Posted Jan 9, 2012 20:29 UTC (Mon) by rgmoore (✭ supporter ✭, #75)
I get the impression that machine translation is still one of those cases where the results are highly variable. It's better when languages are more closely related and where they've put more effort into manually tuning the results. So a pair of reasonably closely related languages that get translated between a lot, like English and Spanish, are likely to produce much better results than two languages from different language groups, like English and Japanese, or a rarely translated pair, like Scots and Faroese.
Posted Jan 18, 2012 1:06 UTC (Wed) by Duncan (guest, #6647)
This has become one of several regularly reoccurring themes on languagelog, which I discovered about a year ago and immediately added a feed subscription to my feed-reader for.
Easy to remember version http://languagelog.com , tho that redirects to languagelog.ldc.upenn.edu (University of Pennsylvania Language department hosts, but a number of linguists regularly contribute including Victor Mair, the specialist in the various forms of Chinese).
I find the language debates there generally pleasant and often fascinating and/or entertaining diversion from the world of freedomware technology that I otherwise spend so much time in. =:^)
Posted Jan 10, 2012 0:16 UTC (Tue) by sumanah (guest, #59891)
Signed, a US-born daughter of Kannadiga parents
Copyright © 2013, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds