Is it possible to count tokens which matched documents?

This discussion may be relevant then.