Ö÷²¥´óÐã

« Previous | Main | Next »

SuperPower Nation: an experiment in multi-lingual debate

Post categories:

Henrik Pettersson | 17:40 UK time, Friday, 26 March 2010

On 18 March 2010 Ö÷²¥´óÐã World Service hosted SuperPower Nation day in London, a six hour experiment in multi-lingual debate and discussion.

The idea was to see what would happen if you joined together people from as many language backgrounds as possible and let them communicate freely in their own language in a truly global conversation. The event was broadcast on Ö÷²¥´óÐã Arabic and Ö÷²¥´óÐã Persian TV, as well as on radio in 15 of Ö÷²¥´óÐã World Service's languages.

world_600.jpg

Hi Rafiki: an automatically translated message board

As part of the online offer a custom message board system called "Hi Rafiki" ("Hi friend" in Swahili) was built by the Ö÷²¥´óÐã World Service Future Media team. This offered all the features of a standard message board with the added twist of automatically translating all posted messages into seven other languages using the .

This allowed our audience to read, post and respond to messages in their own language. A user in Sao Paolo, Brazil could post a message in Portuguese that could be read and replied to in Chinese by someone in Beijing, China. The translations weren't always perfect of course, but most of the time they were good enough to get a sense of what was meant.

It was all sited on a single page that included a small, low bit-rate, three screen video-wall streaming live from the venue. Users could opt to open any of the streams in high quality to get a sense of what was happening in the room which included activities as diverse as video conferencing from around the world to live musical performances. In addition, it offered a map showing the locations of the last 10 users to take part in the experiment. But would they come?

How did it work?

The popularity of the message board was higher than we expected, with close to 12,000 messages posted during the event - that's roughly a message every two seconds over the six hours it was live.

Topics ranged from "Is the web a right or a luxury?" to "Is it possible to find love online?". Quite a few users also commented on the experiment itself, like this message from Rozana in Brazil (translated from Portuguese): "What a fantastic, brave new world, did not think I would live to see the man crossing the barrier language. The simply wonderful!"

While we only provided translations of the supported languages (English, Spanish, Portuguese, Arabic, Chinese, Persian, Indonesian) we didn't stop users from posting in any other language. By the end of the day messages had been posted in 51 different languages, the top ones being English, Spanish and Portuguese. As for the geographic spread of our users, we had visits from 142 countries with the top ones being United Kingdom, Brazil and the United States.

By treating this day as a big experiment we had greater flexibility around the running of the event. We were keen that as we couldn't predict the format, fierceness or topic of debate we should do everything possible to encourage our users to post messages. By dropping the register-to-post and pre-moderation barriers, users fired-up by the discussion could weigh into the debate quickly. We're very pleased to say that of the 12,000 messages posted only 1.8% of those had to be moderated - a tiny fraction!

User experience design

The main challenge was to create an interface that users from around the world could understand quickly and interact with - without having to learn a whole new method of interaction. Thus we looked at sites that already dealt with large numbers of international users in a message stream format such as and . Users are likely to have come into contact with these sites before and we were looking to re-use interactions that people already understood to encourage them to post, rather than unnecessarily re-inventing the wheel.

We provided the ability to 'reply' to any comment on the page, which we thought was an important part of the conversation idea, giving lots of entry points for users. However, the sheer popularity of the event meant that a lot of the replies were being missed by users as so many new messages were coming in at the top of the stack. It was akin to being friends with everyone on Facebook! So whilst our initial problem was trying to get people to post in the first place, the success created its own unforeseen issues. Surfacing new replies is something to figure out for the next iteration. (View a full size image of the GUI.)

mundo_594.jpg

How it was built

We developed the message board software in Python and hosted it on , a cloud based web application platform leveraging Google's infrastructure. Our time-scales were extremely tight, we had just under three weeks to complete not only the audience facing message board, but also the administration and moderation facilities. Hosting it on Appengine meant we didn't have to worry about server setups and scalability issues, and could instead focus all our energy on implementing features.

The core feature of the system is the automatic translation. During the day we presented only seven of the World Service languages, but we built it to be able to handle all languages supported by Google Translate. In fact our audience, as diverse as it was, actually took the opportunity to post messages in over 50 languages.

A user would post or reply to a message on our site which would be routed through and stored on Appengine. This message would sit in a queue awaiting language recognition and translation. On average messages were translated six different ways and returned back to our audience subscribing to the live feed within 30 seconds. By the end of the day the system had performed over 100,000 translations.

Because we removed the need pre-register to post a message, we were clearly vulnerable to spam attacks as well as use of vulgar and obscene language. In order to protect ourselves from this we built in a profanity filter for each of the supported languages, support for IP banning and , into the administration screen. Use of CAPTCHAs was seen as a last resort because it can be problematic for users with non-latin keyboards.

With the message board being reactively moderated, it was essential that during the six hour debate we had an editorial representative from each of our seven supported languages checking the content of the messages, moderating those which were felt to break the Ö÷²¥´óÐã's House Rules in addition to feeding back interesting quotes to editorial colleagues.

What's next?

It was fantastic to be able to bring our large, sometimes disparate, multi-lingual audience together in a universal conversation. The popularity and performance of our message board exceeded expectations and whilst the translations were sometimes a little dubious, the underlying automated machine translation technology will only get better with time. With tools such as the and , the language barriers on the internet will become less of an issue.

In the future it would be good if the conversation we spark could continue after our involvement has come to an end. One idea we had was to host the debate entirely on using a simple . Our sites would serve as a portal to the discussion, presenting the translated messages and featuring discussion points, but it would mean the discussion could take on a life of its own. Of course moderation of Tweets carrying our hashtag would be absolutely necessary to avoid the potential for embarrassment.

We see this experiment as the start of something much bigger. With our 32 different language newsrooms and a truly global audience we believe we are ideally placed to experiment more in this field. As all-pervasive internet continues to make inroads into some of our less-well connected language markets, we see a great opportunity to bring global events such as the and closer to our audience and our audiences closer to each other.

Henrik Pettersson, Tom Leitch, Matthew Isherwood and Alexi Paspalas
Ö÷²¥´óÐã World Service Future Media

Comments

Ìý

More from this blog...

Ö÷²¥´óÐã iD

Ö÷²¥´óÐã navigation

Ö÷²¥´óÐã © 2014 The Ö÷²¥´óÐã is not responsible for the content of external sites. Read more.

This page is best viewed in an up-to-date web browser with style sheets (CSS) enabled. While you will be able to view the content of this page in your current browser, you will not be able to get the full visual experience. Please consider upgrading your browser software or enabling style sheets (CSS) if you are able to do so.