An unofficial blog that watches Google's attempts to move your operating system online since 2005. Not affiliated with Google.

Send your tips to gostips@gmail.com.

October 22, 2007

Google Switches to Its Own Translation System

Google switched the translation system from Systran to its own machine translation system for all the 25 language pairs available on the site. Until now, Google used its own system only for Arabic, Chinese, and Russian.

"Most state-of-the-art commercial machine translation systems in use today have been developed using a rules-based approach and require a lot of work by linguists to define vocabularies and grammars. Several research systems, including ours, take a different approach: we feed the computer with billions of words of text, both monolingual text in the target language, and aligned text consisting of examples of human translations between the languages. We then apply statistical learning techniques to build a translation model," explains Franz Och.

You can compare the new Google Translate with Babel Fish, a site that uses Systran to provide translations. The switch is a sign that Google's system has improved a lot and could soon be ready for expanding its coverage.



{ Thanks, Steve Rubel. }

45 comments:

  1. I tried to translate an english website into french and it's still not a great translation.

    ReplyDelete
  2. I typed in some basic phrases from Eng>French and was surprised at how poorly they were translated. Google translate has a long way to go before matching babelfish.

    ReplyDelete
  3. You shouldn't expect perfection from an automatic translation. Humans are always better for this job, but that requires time and money.

    The translation should be good enough to understand a text, even if the output doesn't necessarily contain grammatically correct sentences.

    Google's approach works better for some languages and worse for others, but at least Google can expand to other languages without having to know them and manually create models for each one.

    ReplyDelete
  4. Oh, and there's an option to suggest a better translation. You can edit the wording and help Google improve its system.

    ReplyDelete
  5. I tried to transalte this http://mytourbook.sf.net/ webpage from German to English with both Google and Babelfisk and it seemed to me, that Google produced more enjoyable results (maybe it's because there is lots of common phrases).

    There is nice joke in the transalation - the sentence "This Website is currently only in **German** available" (the only English sentence on the page) got translated into "This website is currently only available in **English**" ;-)

    ReplyDelete
  6. I think it is still just unusable. I just tried to translate from Portuguese to English a text in which I was saying something like "I heard from my wife that the actor Paulo Autran died." And the resulting translation claimed that my wife was dead. It is not just grammatically unacceptable, but also the meaning is comppletely lost.

    The text I translated is the first paragraph of the following post in my blog, if you want to try.

    http://leonardofernandes.blogspot.com/2007/10/paulo-autran.html

    I challenge you to make any sense of the resulting translation. And no, I don't write that badly in Portuguese ;-).

    ReplyDelete
  7. If I translate "Estou aprendendo português" from Portuguese to English I get this translation "I am learning English"...!

    ReplyDelete
  8. The translation "English to French" fails.
    "I'm watching TV" is translated into... "I"m watching TV"...

    ReplyDelete
  9. Check this:

    Translate
    "The weather is fine."
    from English to French

    Google Translate:
    "Il fait beau."

    Babelfish:
    "Le survivre à est très bien."

    ReplyDelete
  10. Plucking a random few examples out of the air really tells us nothing.

    Someone with a little time on their hands needs to prepare a bunch of trial texts, and use it on several languages, in both directions.

    Then we'll see how they get along.

    And then they need to repeat it after six months - Babelfish is 'mature', while the Google model should still be learning in six months time.

    ReplyDelete
  11. Now and then I need to translate a phrase. I paste it in Google and see how many exact matches there are. This is far more reliable than using a translation service like Systran. I 'm sure Google's translation concept is much better than what Systran has to offer. It's based on the wisdom of the crowds.

    ReplyDelete
  12. I translated this blog post into German, and I have to say: Awesome.

    It proves what I already thought for a long time: Google's statistical approach is not much different from how our brain works. It's based on statistics, too. That's why listening to adults when we are small lets us learn the basic language.

    I corrected a few grammatical mistakes, but it's basically awesome. The system works WAY better on more complicated sentences than the other solutions available.

    ReplyDelete
  13. It is not "a few random examples". I simply have never seen a Portuguese to English automatic translation that works (except for translating stupid sentences like "the book is green". I am not criticizing google system in particular. I am talking about automatic translators in general. If someone knows Portuguese to English examples that work, please let me know.

    ReplyDelete
  14. > Babelfish is 'mature'

    I'm glad you put that in parentheses... I'd guess Babelfish/ Systran is around 1 years old in human age :)

    ReplyDelete
  15. Have you tried http://translator.live.com/

    ReplyDelete
  16. http://translator.live.com/ == Babelfish.

    ReplyDelete
  17. I really must say that it works very good. I tried to translate German into English and I was pretty impressed. It is fast and I like the option to give some better suggestions. It would be nice that it would also support the Dutch language... ;)

    ReplyDelete
  18. Wow. It's a lot better than the old Systran system.

    However, I would like to see more languages supported, and to have two selection boxes for selecting to/from languages instead, so we can translate from any language they support to any other language.

    ReplyDelete
  19. jp: 日本語は本当に難しいです。(means: japanese is really difficult).

    translates to: english is really difficult.

    the same happens to other languages I tried (portuguese, spanish).

    Imagine this with a report comparing japanese and american stats :-)

    ReplyDelete
  20. Has the Spanish to English been switched over to the in-house Google system?

    It's still not paying attention to accents in the Spanish text. This leads to problems with, for example, distinguishing between a verb form ending in o and another form ending in ó. The former is a first person singular present tense, while the latter is a third person singular past tense. That's a big difference. I can't seem to get Google to see the difference.

    Take as an example the following Spanish text...

    Entonces les pidió a todos que se sentaran.

    (In case you can't see it, this "pidió" has an accent on the last letter)

    The above is translated by Google as...

    "So I asked them all to sit."

    If the accent had been taken into account, we would have gotten something like this:

    "Then he asked them all to sit."

    It seems like an elementary error for a system to fail in preforming a translation because of not taking into account an accent.

    ReplyDelete
  21. chlori said...

    If I translate "Estou aprendendo português" from Portuguese to English I get this translation "I am learning English"...!

    I think I know what's happening here. The translator took the word "português" to mean "my native language." Since the native language of the original text was Portuguese, the machine translated that into "English" -- the native language of the translated text.

    Their method would make some sense if I were, say, yelling at someone, "I don't understand you! Speak English!"

    From one (odd) point of view, that could be translated into Portuguese as "I don't understand you! Speak Portuguese!" (ie., into the speaker's native language.)

    But I tried translating "I don't understand you! Speak English!" into Spanish, and got back the following:

    "Yo no entiendo! Por favor, hable en inglés."

    That is correct. In fact, I had Google translate the result back into English, and got back

    "I do not understand! Please speak in English."

    And that is pretty good.

    ReplyDelete
  22. I have just tried a translation from English to Spanish and I feel it's much better than before. It's not just raw translations, it seems it can 'get' the context of the sentence.

    ReplyDelete
  23. This comment has been removed by the author.

    ReplyDelete
  24. Never translate poetry using automated translation systems.

    ReplyDelete
  25. Yeah, after fiddling around, I see how completely pointless that is. Its too bad I don't read much in the way of legal documents.

    ReplyDelete
  26. Yes, machine can help humans achieve a faster translation. But one should not rely on it, as words have subtle meanings that only humans can detect. And the reason is, that they can feel the words, not only read and understand them. They can adjust their meaning to a culturally appropriate translation. If you are English<>French bilingual, see what I mean by reading the above text translated with Google Translation System:

    Yes, machine can help humans achieve a faster translation. But one should not rely on it, as words have subtle meanings that only humans can detect. And the reason is, that they can feel the words, not only read and understand them. They can adjust their meaning to a culturally appropriate translation.
    Oui, une machine peut aider à atteindre plus rapidement l'homme traduction. Mais il ne faut pas compter sur elle, comme les mots ont des significations subtiles que seuls les êtres humains peuvent détecter. Et la raison en est, qu'ils peuvent ressentir les mots, et non seulement de lire et de comprendre. Ils peuvent ajuster leur signification culturellement appropriés d'une traduction.

    ReplyDelete
  27. It looks so very good on that example, except when you actually read the French "source", its horrible grammar and sentence structure, that you can see that something is amiss. This is terrible self-aggrandizement! A little sneaky, if you ask me.

    ReplyDelete
  28. I am doing an English-French translation for a technical document. I compared the translation for an 8 lines paragraph, between the new Google translator and Babel. Nothing is perfect in both case, but Babel had a lot better translation. However I like the option to suggest a better translation. I will like to know what will be the use of this feature in practice.

    ReplyDelete
  29. It just goes to show how fragile a language is, and how its integrity can be compromised in the wrong 'hands'. This test suggests that the world-wide aspect of the Web seems only open website authors who figure out that translation is more than clicking on a link.

    ReplyDelete
  30. On my weblog on internet searching, I did some comparisons of Spanish sayings into English, and compared them to Babelfish. There's not the big improvement, but I'd still say, that Google Translate wins by an inch over Babel Fish. The blogpost is in Danish, but check out
    1) The Spanish original
    2) Google Translate into English
    3) Babelfish into English.

    The test can be found on:
    http://erikhoy.blogspot.com/2007/10/google-translate-igen-iflge-google-egne.html

    ReplyDelete
  31. I can't find out if we are allowed to use this service for a commercial application. How can I find out?

    ReplyDelete
  32. I hate it with a passion. It is of no help when reading translated pages and just gets in my way and really ticks me off.

    Good old Google, always trying out what they think is good. I have news for them, it sucks!!!!!

    ReplyDelete
  33. I found an awesome translation tool – LEC’s translation software. Please see the example below.

    Example in Spanish:

    Luego de la primera comunión del niño, el padrino brindará con los padres.

    Google: After the child's first communion, the godfather provide with their parents.

    LEC: After the boy's first communion, the godfather will toast with the parents.

    I also tested with a few other examples in other languages and LEC’s translation is better than Google’s.

    You can get a trial at www.lec.com

    ReplyDelete
  34. Urgh. Can anyone help me? I used a version of google translation a couple of years ago (I bookmarked the translated version and it had google's IP). Today I went to the bookmark, and the translation came out as in the "google translate" version, and it is TERRIBLE. I am translating from Chinese to English, and Google Translate is about as gibberish as babel.

    I don't know how to get back the original translated page I used for a couple of years.

    ReplyDelete
  35. frustrated,

    to get back your previously translated page you would have to use the older version of google translations. meaning: google would have to change back to the older version. and that is highly unlikely. unfortunatly you cannot get that previously translated page unless you saved an html document when you bookmarked the page.

    ReplyDelete
  36. Actually I don't see any difference between thos two.

    ReplyDelete
  37. I have tried translating Hindi text into English, but its not even at a begginer's level. It is not doing a good translation. It only translates words at their place, but it cannot arrange the output sentence, so the meaning or the sence of the sentence remains of the input sentence.
    It only provides a raw conversion.
    They need to work a lot hard on the translater.

    ReplyDelete
  38. why there isn't persian translator??

    ReplyDelete
  39. Google translation from Hindi to English are a kind of joke which makes people to laugh. This not even near to perfection.

    ReplyDelete
  40. Because google's translation is based on statistics, you can occasionally run into translations that change the meaning significantly, but keeps a credible sentence structure. For instance, if I translate "Jeg spiller for Vålerenga" (Vålerenga is a Norwegian football team) to English I get "I play for Liverpool FC". The translator doesn't know it is a proper name, and "Liverpool FC" seems to it to be a word used very similarly to "Vålerenga" in Norwegian.

    ReplyDelete
  41. need russian -> english !

    current ones sounds very funny

    ReplyDelete
  42. In google translation Tamil language can be also added is most welcome by tamil peoples

    by
    arasiyalkalam

    ReplyDelete
  43. it improved in my solution.I tried to translate an english website into arabic and it's still not a great translation of MSN.

    ReplyDelete
  44. It seems these translation services work best when you are translating from another language back to English. I'm working on a project to see how to make these translations more grammatically friendly.

    ReplyDelete
  45. BabelFish is now redirected to Microsoft Translator. However, become a Spelling and Grammar Nazi to translate the BF3 Mission 8 Comrades introduction after the cinematic scene from French to Polish.

    ReplyDelete

Note: Only a member of this blog may post a comment.