Advertisement

Personals
Enter our FREE personals site!
Advertisement



Advertisement


About Shanghaiist

Shanghaiist is a website about Shanghai, China. More

Managing Editor: Dan Washburn
Editor: Kenneth Tan
Publisher: Gothamist

tips@shanghaiist.com

info@shanghaiist.com

advertising@shanghaiist.com

RSS (FB) | About | Advertising | Archives | Facebook | Mobile | Staff | Twitter | Write For Us

Recent Comments
Public Calendar
Contribute

Latest tip:

There was a story a week or so ago about a new Mumbai Shanghai route, well now the other carrier [more]

 

Latest link:

 

Latest Photo:

 

Favorites
Newsletter
Too busy to check the site? Receive a daily email with links to all Shanghaiist posts from the previous 24 hours.

Enter your email


Powered by FeedBlitz
Subscribe
Use an RSS reader to stay up to date with the latest news and posts from Shanghaiist.
Advertisement


Advertisement

March 21, 2007

Still lost in translation, but with Google we lose less

googletranslate.pngAfter an enthusiastic e-mail circulated its way through some expatriate distribution lists last week touting the precision of an ostensibly upgraded Google Translate tool, the verdict from at least one blog is in: meh. According to China Herald:

Of the English to Chinese translations, my Chinese colleagues say about 60 percent is correct. The software picks the most commonly used characters, and that is not always the right one. It might be an interesting experiment. The system allows you to suggest corrections and when people start doing so, it might become better over time.

China Herald posted a follow-up yesterday, concluding after 14 tests of translations in at least four different languages that, while the translation to English from Chinese is readable, the translation to Chinese is "a disaster."

What makes Google Translate better than other alternatives such as AltaVista's Babelfish, but still not good enough to send professional Chinese-English translators scurrying to update their C.V.s? According to their own research scientists, Google shuns the traditional rules-based translation for a statistical machine translation:

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. We have achieved very good results in research evaluations.

Shanghaiist male bimbo translation: They do magical smart people stuff.

Google Translate does have some genuine innovations to it, including an option for you to suggest a better translation if you're not satisfied with what the tool spits out. Google will then incorporate your feedback and throw it into their mysterious statistical soup for continued maturation of the platform. When mousing over the text of a translated page, a little Google window will pop up showing the original text, and also give you an opportunity to make a better suggestion. It also appears to be the only free service out there that translates between Simplified and Traditional Chinese.

Does Shanghaiist recommend using it? We tend to agree with the opinions of the always insightful readership at China Law Blog, who suggested in this comments thread that statistical translation may be effective with words and phrases that are defined within a very narrow scope (legal contracts, scientific research, public signage), but perhaps not for mainstream language where one word can have multiple meanings. Always use discretion when doing automatic translation, and conduct your best double-check before using it, even when translating small snippets of text. The last thing you want to do is publish something that ends up hurting the feelings of the Chinese people.


Email This Entry







Advertisement: Shanghaiist Continues Below!


Comments (2)

I believe the correct term is 'himbo'.

 

Great posting! There's definitely still ways to go with statistical machine translation. If you'd like another example of SMT in operation, please feel free to check out Kontrib.com which quietly launched several weeks ago.

Kontrib is a multilingual social bookmarking website which leverages Language Weaver's baseline statistical machine translation servers. Kontrib launched with support for English, French, and Spanish as well as limited support for Arabic (Arabic into English). We're looking to add in Chinese support soon. Our goal for Kontrib is simple - foster communications on the Internet by lowering the language barrier.

While Machine Translation is still certainly in it's infancy, the great potential for this technology is clearly apparent. Consider - content deemed popular in English could now be read by French speaking users of the Internet. The same would hold in the other direction too. Users can comment and chat with one another with much more clarity. Are we there yet? Not quite.. but we're quickly moving to a point where MT can already providing excellent gisting quality output if not better.

Feel free to use Kontrib and if you have any questions, do not hesitate to ask! We're hoping as many people utilize the technology as possible. We'll also be adding in a user feedback loop which will allow users to help improve our systems too!

 
Post a comment (Comment Policy)

2003-2008 Gothamist LLC. All rights reserved. Terms of Use & Privacy Policy. We use MovableType.

Site Meter