Facebook likes Microsoft. Chooses Bing API to power translation

The Inside Facebook blog reported last week that Facebook is testing a Translate button for comments on pages. With the introduction of this new feature, comments that are written in languages other than your account’s default language now include a Translate button next to them. When you click on the Translate button, the comment is automatically translated to your default language. The Translate button is then replaced by an Original button which untranslates the comment.

I have learned that this service will be provided through the Microsoft Translation API. Facebook will not be using a self-developed machine translation for this service. At least not at this point.

Microsoft is fast becoming the ‘go-to’ translation API player. Since Google deprecated their free API and turned it into a paid service, developers are abandoning Google and using Bing instead. Microsoft has stressed over the past few months that they will make no changes to their free translation API and have even increased user limits to make it even more attractive to developers.

Another major software company that went with the Bing Translation API is Autodesk, recently releasing TransTips for AutoCAD, Inventor, Revit, and 3ds MaxThis plugin translates the contents of tooltips in various Autodesk products automatically into one of 35+ languages via the Bing translation API.

The cooperation between Facebook and Microsoft should come as no surprise to anyone. Microsoft is a minor shareholder in Facebook, having bought 1.6% of the company shares for $240 Million back in 2007. I tried to get more information from Microsoft about this new cooperation with Facebook but could not get any information. I can assume with great certainty however, that the terms of the cooperation were discussed at a high level and that Microsoft will provide specially trained MT engines to Facebook at some point in time.

I also tried to obtain more information from Ghassan Haddad, Director of Internationalization at Facebook. I’ll report back with any new information I can obtain.

Enhanced by Zemanta

Social media localization? It’s all about listening. Day 2 of #mlwpisa

The Multilingualweb Pisa workshop came to a close today. It was a good event with lots of information. In participation were people from the European Union, from industry and from academia. Many of the big names were there including Google, Facebook, Microsoft, SDL and Lionbridge.

As I said in my blog post yesterday, I won’t discuss every speaker from Day 2 even though they were all excellent. The conference was widely covered; there was a live video stream and all of the presentations can be found of the web. I will just focus on a few items that I found of interest.
Continue reading

The wisdom of the crowd. Does it work in translation?

Crowdsourcing and translation. Translation and crowdsourcing. It has become the holy grail of the industry, a potential cash cow that is enticing to many companies. But how smart is the crowd? And can it be relied upon? My quick answer is that in the field of translation, the crowd is unreliable and can not be trusted. Facebook learned this the hard way last year when some of their websites spewed out profanity in various languages (read more here).

Another case in point. I just came across a new website called Ackuna. There is absolutely nothing creative or innovative about this website, which is merely an exploitation of Google’s free translation system. It combines Google Translate and the wisdom of the crowd by allowing people to vote on a translation and enter their own suggestions. The company bills this site as ‘a more accurate free translator.’ But the crowd can be stupid sometimes and if a few people voted up a translation, it will be passed off as the correct translation. This can lead to highly inaccurate translations or even worse.

In my own opinion, translation crowdsourcing will only prove effective when stringent crowd control is used. This can be done by having language experts or moderators at the end of the line who decide which translation is more accurate; it can also be done by limiting the crowd to people who can be trusted. These methods will slow things down. But without crowd control, things may get out of hand.