We stated earlier that it will be a long time before knowledge can be acquired by simply reading general texts such as encyclopedias or newspapers. This is why START focuses on understanding single sentences rather than full texts. However, we believe that there is a class of textual documents with valuable information which are explicit and literal enough to be handled correctly by the system.
The DoD is probably the world's largest owner of such documents because it has attempted to reduce its procedures to precise doctrine that troops can train against. There are also documents describing some of the ways in which intelligence information is combined; every DoD interpretation task seems to be accompanied by Indicators and Warnings (I&W) which might trivialize the full interpretative task but which are nevertheless useful.
The AI Laboratory has at least two other powerful natural language systems besides START: John Mallery's RELATUS system and Robert Berwick's Principles and Parameters based systems. We will experiment with using all three of these to assimilate the knowledge contained in such textual sources of knowledge. As part of this effort we will try to develop a reasonably precise definition of ``machine parseability.''