The research described in this paper was motivated by recognition of the limitations in the traditional approach to authoring. The collection, structuring and presentation of information is a time-consuming, knowledge-intensive task, so that the considerable effort of producing a document has had to be carefully rationed. Authors do not have time to prepare different versions of their documents for different readers. As well, the high costs of printing and distribution effectively limit the possibility of revising documents for new audiences or when new information becomes available. The central problem with traditional authoring approaches is that the form and content of the document are determined without recourse to models of individual readers; the resulting document is therefore a poor fit to the needs and goals of these individual readers.
Recent technological advances have liberated us from the printing press, and a re-evaluation of traditional authoring is in order. Breaking the authoring process into its constituent agents (the author(s) and reader(s)) and knowledge sources makes it obvious where the bottlenecks are, and how they might be alleviated. Form and content are determined too early in the pipe between author and reader because specification and presentation of the document are intertwined and conflated. There needs to be a clear separation of specification-the author's compile-time task-, from presentation-a run-time task mediated by the author's specification and by models of individual readers.
We call the ensuing approach we have developed intent-based authoring, to emphasize the primacy of intention in authoring. See Figure .
This paper describes how minimal AI techniques can be used to implement intent-based authoring in a particular domain and medium. See Csinger et al.  for more on how intent-based authoring is distinguished from the traditional approach to authoring.