We notion of argue that in domains can be where it a strong can be Mann and Moore [1981], on the other hand, while assembling texts dynamically to suit their audience, do so by "over-generating" the set of facts that will be related, and then passing them all through a special filter, leaving out those that are judged to be already known to the audience and letting through those that are new. McKeown [1981] uses a similar technique -- her generator, like Mann and Moore's, must examine every potentially mentionable object in the domain data base and make.