Facebook Offers Translation Tool to Other Sites

Social Networking

Back in January 2008, Facebook was available in one language: English. That’s when the company introduced a new translation tool that allowed its members to do the hard work of translating the site into their native languages. As a result, Facebook is now available in 65 languages, and a majority of its users are overseas.

Translations for Facebook Connect screen shot.

On Thursday, Facebook will make that translation tool available to other sites. The new program is called Translations for Facebook Connect, and it is being offered to the 15,000 sites and applications that use the Connect service, which allows visitors to log in using their Facebook ID and password and broadcast some information back to their friends on the social network.

The translation tool works by asking users to submit possible translations of phrases, and then soliciting their votes on which is the most accurate. So now a country’s tourism Web site, for example, can use the tool to solicit help with a translation, and then present the site to users in their native language when they log in using their Facebook ID. It is free for developers, but Facebook hopes it will increase the use of the Connect Service.

Facebook’s human-powered approach juxtaposes quite sharply with Google’s service, which uses technology to automatically translate Web sites and text — with occasional unintentionally comical results. (The Facebook system, of course, has had to handle a relatively tiny number of phrases.)

“Other businesses try to accomplish the same thing using technology to solve these problems, and it’s not always 100 percent accurate,” said Ethan Beard, head of platform at Facebook, in a veiled reference to Google. “But technology doesn’t take into account cultural values, idioms that are hard to translate. In the same way we think reviews are better when they come from friends, translation done by people is significantly better than what you would get otherwise.”

Comments are no longer being accepted.

It would be better if people who help translating could be paid for their work.

Of course, this approach assumes that anyone who speaks two languages is capable of translating anything from one into the other, effectively rendering valueless the hard work of trained, experienced, trained translators. A company wouldn’t go on line to solicit unpaid help from its users with its accounting, or legal issues, or human resources strategy–why let amateurs handle the task of translating the most public part of a company’s online presence–its Web site?

The human-powered approach is a must when it comes to translations. I hope that following Facebook’ step, more and more website owners will discover the benefits of multi-lingual websites. But you should learn a lesson from Facebook too: if too many people translate your website you face a serious problem of consistency in the translations. You can see it easily in any local version of Facebook. I recommend – whenever it is possible – to use few translators, and also to make a thorough checkup of the translations by an integrator. I know that the idea in crowd-sourcing is to do all those things free of charge, but believe me – a good local version worth the money. Personally, I use a service called OneHourTranslation.com to translate my website content to other languages. Their pricing is affordable and you get a high-quality, consistent translations – good value for your money. I don’t think the approach Facebook shows juxtaposes Google’s approach. They are simply intended for different purposes. Google’s machine translation approach is good mainly for general understanding of texts meaning. It mustn’t be used for website content translation, and may harm the website image and reputation.

Funny we should have this article today being the International Translation Day. It is my opinion, and that of the many translators and translators to be, that the ability to speak in a bunch of different tongues doesn’t enable one to translate. Consider, for example, we’re all taught Calculus at school, but that doesn’t mean we’re all capable of becoming the CFO of a company, or even, many of us attended Mass when we were young, but that doesn’t mean we can be priests. Consider that next time you use a web-based translation system and try remembering the amount of professionals that could do a better job for a surprisingly affordable pay.

This is such a good idea. Because it worked soo well on Facebook’s site.

This is sarcasm, since the Spanish version on Facebook looks like a rap sheet for the number one 101 intro to spanish class errors. I’m sure other languages are just as broken.

How is “Other businesses try to accomplish the same thing using technology to solve these problems, and it’s not always 100 percent accurate,” a “veiled reference to Google”? This seems lazy to me. There are dozens, if not hundreds, of machine-learning players in this game. Google may now be the biggest, but they have not been for long, and they’re probably not even the best.

While the issues of translation consistency are quite real as stated well in these comments, I note that this is a pretty technologically interesting way of showing the potential of capturing web delivered content which is probably stored in a database and then voting on the translation. The workflow enabled shows real promise, as compared to traditional processes of working on lists of words, and then iterating on localization testing when you see the application complete. I like the tech, but maybe the crowd sourcing approach has some potential quality issues. Even so, perhaps Facebook wouldn’t be feasibly presentable in a wide variety of languages if they didn’t do it this way. I wonder if they had zero actual ROI, when taking into account coordinating all the volunteer efforts. They clearly also did a good job with software internationalization to enable this to work.

The automated approach to web page translation is fine if you are prepared to take the reputation and business risk associated with it.

The collaborative human approach as noted is also not without its issues.

The possibly more costly but most effective way is using experts – a must for any business serious about global marketing.

I guess I can sympathize with those wishing for ‘professional translators’ to do this work for pay; but that’s just not the reality (not least of all because FB doesn’t generate a profit). This method seems like a logical extension of the method by which people interpret ‘captchas’ generated from scanned material that the OCR software can’t translate accurately. Thus two jobs (one, a secure log in procedure; and two, a way to ‘crowdsource’ translation of scanned text) can be done with one process.

(typos corrected):
This is just another in the recent spate of gigo translators
(garbage in, garbage out) Google translation is the same,
just build a big hash table of ’sure’ translations of fixed
sentences, those will ”translate” (i.e. map exactly),
As for phrases with no matching hash key:
GIGO! There will be no useful machine translation until
there is AI.

Semantics = meaning. When amateurs approach translation tasks with the same gusto as text messaging, sooner or later ‘R U ready’ for the results? I am one worried Professor — try grading papers of the under 21-generation…..

responding to Steven:
There is a vast difference between being an expert in law, or accounting vs. being fluent in a given language. The whole point of this technology is to tap into the vast resource of multi-lingual internet users that can, for fun and with minimal effort translate web content into other languages. This is something that can be one by anyone with language fluency and doesnt necessarily require advanced degrees. Certainly there is a risk of human error, but agurably this human error will be MUCH lower than machine error due to voting capabilities built into the software.

Instead of a binary approach to translation, Facebook’s approach may be more useful in modifying machine translated text. Machine translations are extremely accurate at the phrase level. At 12 word sentences they begin to lose their way especially in Languages with different grammatic structures such as Japanese. Couple that with the Japanese predilection to writing 50 word sentences and you have randomized meanings. I have developed a Japanese English translator that maps the grammar reasonably well so that it maintains its meaning. However the translation does not result in smooth flowing English text. I think Facebook’s approach can add the finishing touch to the translation.

To see Google’s translation efforts you need to dig down to the Translator Toolbox. It is much more in depth than the Facebook UI tweeks you are showing and was quietly released in August
//translate.google.com/toolkit

As usual Google will use the work of the cream of contributors to build its in house capacity…. crowd-sourcing top translations. It includes:

– A full *professional* parallel text translation system with Glossaries and Memories
– Submit alternative translations to improve Google’s current version
– Full GoogleDocs-style collaboration and sharing

R3

No matter how many translations are performed, the best translation is going to be a measure from the source text to the translated text. I very much doubt that the best translation quality will be based on a comparison of both the original (source) and translation itself. It’s more likely that users will assess the final translation’s flow and readability rather than the art form of doing as little as possible to dilute the original intended meaning, once translated. It’s why I believe that professional translators will never be replaceable.