I agree that localizing software can only be as good as the data that you get, that is why I hope for good results at least as long as the information is contained inside the localization. Adding computed data to the image will likely result in compromises, and the trick is to find out which.

I feel that somehow the architecture available or at least used today doesn’t put the power and the lack thereof close enough together, and I intend to fix that, too.

Feedback on l20n is welcome, either in the i18n newsgroup, in the wiki, or directly to me.

I don’t see the problem going away anytime soon, so if someone has put in some effort to make it better, I’d like to see what they did. We’re toying with the idea of making it better ourselves ( http://wiki.mozilla.org/L20n ) so it would be great to learn from other’s mistakes.

Thanks for the reply.

The problem is that the developer assumes that it’s possible to do a straight translation. But that’s not necessarily true. For example, who says that there will be always 1 argument to replace in “You have %d messages” ? Ok, maybe in this simple example, but the language might dictate otherwise.

As an example, I had a fellow programmer that actually solved the masculine/feminine/neuter problem above. Bzzzzzt ! Wrong ! Some languages have multiple forms (Czech has 4 forms). Some have less. In Portugese, a man might say “Muito obrigado” (Thank you very much), but a woman says “Muito obrigada”. Etc … etc … It’s an endless nightmare.

And no, I don’t have a good solution either – there are a few attempts, mainly in the AI area, but most are often unusable in C/C++.