It would also be a big work. Internationalization is already a
difficult task just because of finding translators. Finding translators
who _also_ know a bit about accessibility and have an idea on how they
should translate strings is even harder, I'd say "square as much" :)
That's why having a technical solution that just uses what exists is
the most pragmatic solution for now.
That being said, having a different output for sighted and non-sighted
users is not so handy, in particular when you have both kind of users
working together: if they don't have the same output, they will not
understand each other.