Web 2.0, lost in translation

April 10, 2006 —

It’s the social component that gets me most excited about the Web 2.0 landscape. Take digg for example. Dump the Ajax, and it’s still digg. But, dump the social ranking engine, and it’s something less.

With audience playing such a key role in the Web 2.0 momentum, I’m surprised to see such a gaping hole in the web-based language translation space. It’s not that online translation services don’t exist, it’s that they don’t integrate into web-based applications. In fact, of all of the services I’m aware of, only one (SYSTRAN) offers a dynamic/automated service. Please correct me if I’m missing any.

SYSTRAN effectively offers two options for server side translation, SYSTRANBox and SYSTRANLink. The former serves as the by-request content translation engine for the likes of AOL, Babbelfish, Google, Lycos and others (I believe the Pentagon is on that list as well.) The latter functions as an automated public web site translator.

– first intermission –

We first used SYSTRANLink in 2003 and were pretty excited with the initial results. Integration worked like this: we selected the subject category for the site’s content (e.g. financial), and then requested the appropriate language pairs. Dictionaries could also be tweaked, but we never got that far. Next, we carved up and then implemented some code to add the magical translation flags. And, bang, we had a multilingual web site. Sounds great, right? Keep reading.

First of all, the solution was expensive. SYSTRANLink cost between $6,700 and $27,000 per year depending on the size and popularity of the site.

Second, clicking the flags took you to a mixed up version of the site on the SYSTRAN server. (I’ll assume that I don’t need to explain why this was bad and move on.)

Lastly, the translations were a little sketchy. I’ll explain further down.

For these reasons, our multilinguality experiment was very short-lived.

What’s that you say? 2003 was three years ago? The services have surely improved? Not so, ’tis still the same. I’ve checked in on several occasions (including today.)

– second intermission –

The intent of this post isn’t to flame the integrity of current translation services, but it should be noted that integration isn’t the only barrier to implementation.

I speak a little French, so I use the English to French language pair as my mark for quality. I’ll assume that at least some of you don’t parlez and take it one step further by illustrating what happens when we translate to French and then back again to English.

SYSTRAN

1. Charisma 18 is a blog by a web guy named Aaron.

2. Le charisme 18 est un blog par un type de Web appelé Aaron.

3. Charisma 18 is a blog by a type of Web called Aaron.

AOL, Babblefish, Google Translator, World Lingo, etc.

1. Charisma 18 is a blog by a web guy named Aaron.
2. Le charisme 18 est un blog par un type d’enchaînement appelé Aaron.
3. Charisma 18 is a blog by a type of sequence called Aaron.

InterTran

1. Charisma 18 is a blog by a web guy named Aaron.
2. Carroças 18 c’est une blog près de une toile type nommé Aaron.
3. Chariots 18 is a blog near the a canvas chap named Aaron.

I think the web community could forgive some of the lossy nature of the translation beast if it meant being able to access new services or offer services to new audiences. Look to Riya for validation – their facial recognition software isn’t perfect, but it’s still accepted as exciting innovation.

Perhaps I’ll kickstart the effort by developing dynamic translation on this blog. I could start with the popular language pairs: US to UK, US to Canadian, US to Kiwi, and, finally, US to Aussie. Someone needs to get the ball rolling.