To give credit where it's due, the project was originated by Sandy Trevor (now here) and was led throughout by Dr. Mary Flanagan (now here). (I'll send an e-mail inviting Mary to correct or elaborate in the comments). The MT technology used was a now-vanished rule driven engine called Transcend, by Intergraph. (Update/Correction: Transcend was first sold to Transparent Language and then resold to SDL which still supports it.) A contemporary trade press description of the project is here. Fair warning: Some of the systems created were patented (US Pat #5,715,466).

We first set up the machine translator to run on a technical support forum for the service's Mac based proprietary client. As commenters below noted, restricting the translation domain, particularly into a technical field with a lot of jargon, makes the translation more effective. The translator would run through the lists of posts every few minutes, and translate every new English post into French and German and insert them on parallel forums in those languages, also grabbing new ones originating in French and German and translating and reposting them in English. (Note we never had the non-English language pairings - all of these projects were 'stars' around English.)

This went pretty well, in spite of some squawking and mocking by a group of professional translators on the service, and we moved on to bigger things. Adding Spanish to the list of languages, we create four parallel forum bulletin boards which were collectively called the 'World Community Forum'. We seeded them with some guesses of topics that people would want to talk about across cultures - e.g., Sports, Politics, Travel. We were now venturing out of safe technical topics, and the technology was going to be challenged more severely. One of the joys of online services then, as with the Internet now, was the ability to just 'go live' with your experiments. And since the revenue model of the day was pay-per-connect time, we could even try to make money doing so. After running a few tests, we make the forums public, and started promoting them on the "What's New" part of the main CompuServe menu.

The World Community Forum was a roaring success. It attracted hundreds and soon thousands of visitors, went profitable within a few weeks, and soon was regularly appearing in the top 50 products list of the entire service. (You'd have thought the profits would have been added to my R&D budget, but noooo....) After some weeks of dealing with translation accuracy and with scaling issues - we'd brought the whole project up on a single server in a field office - there was enough time to run some analytics and dig into the conversation patterns and topics that were evolving in the forums.

The results were very interesting. While there was significant use of the translators, the majority of posts and responses were in the same language, even though we could tell from the idents of the users that many were posting from different native language territories. Digging into the texts, we found some of them ridiculing and trying to correct the machine translations, but that more were people deliberately using their second langauges. Germans tried out their schoolbook English on the Americans and Brits, and Americans used their rotten Spanish to try and converse with Latin Americans. And all sides cheerfully corrected each other's mistakes as the conversation flowed.

What had begun as a technology trial had in fact resulted in a marketing experiment. Just by the way we promoted the forums, we had drawn in those most interested in cross-cultural meetups, from a service that was already the most international of that time. Topic drift being multicultural, of course they proceeded to talk about anything that came to mind, largely foiling any ideas to use specific vocabularies in particular topic areas. (I clearly recall reading one thread in which German hausfraus and wives of US troops posted there had a lengthy and detailed discussion of their different approaches to drying and folding towels. Go figure.)

I take these results as optimistic for the idea of translated blogs as cultural bridges, in several ways. The known deficiencies of machine translation didn't kill the experiment. Real people were able to repair the dialog in almost all cases of error, and found some common cause in doing so. Providing the facility attracted motivated cultural 'bridgers' who in many cases supported each other in language skills directly, as well as carrying on enjoyable and wide ranging conversations. Few of them were on world saving topics, but a number of close friendships resulted.

'Translating' this experience from a closed, proprietary, and highly structured system into the loosely linked, open, and dynamic blogosphere is going to be a bit of a challenge. It's likely to become a social system as much as a technology infrastructure, so they need to start modestly and co-evolve. But this long ago and already paid for experiment suggests it is workable and worthwhile.