There are now a lot of machine learning based translation systems that could be used.
- Stack Overflow translations all questions and answers.
- Native speakers can edit translations
- The edited translations are fed back into the learning set for the translation engine.
One problem will be stopping non-English speakers from editing the meaning of an item when the system thinks they are editing the translations.
(I think this will need external funding.)
Ok, I was assuming that most people will be up-to-date on the research on machine translation.
The modem machine transitions systems work very well provided they have a large “learning set” that contains documents that have been translated by hand that have very similar language usage to what they are translating.
Microsoft research has got very good results with them for translating knowage base articles from English to Spanish. The most read articles are translated by hand and used as the learning set.
Given that Stackoverflow has proved what results can be got by harvesting the power of the users to build a question and answer site, it is not unreasonable to think that the power of a user based could be harvested to build up a training document set that colligates well with the questions and answers on Stackoverflow.
(The tabs may even help a transitions systems to know the context of the text)