Using Large amounts of Knowledge about Knowledge

Our proposed work centers around knowledge acquisition using natural language. This acquired knowledge is intended to be general purpose and to support a variety of applications. Although we believe that our technology will be of benefit to all the proposed applications of HPKB technology, we believe that our approach is of particular relevance to the applications which require a broad collection of knowledge about the political workings of the world.

The following scenario gives a sense of the type of application system that we imagine could be enabled by our technology:

James Smart, an intelligence analyst, reads several hundred documents a day. As he reads the documents he types a few English sentences describing the content of each document thought to be relevant. He also types a sentence or two describing what he sees as the implication of the document.

Today a document crosses his desk describing unusual small buildings which have been observed on the Confusnian side of the Confusnia-Drazni border. James enters his hypothesis that the Confusnians are tunneling under the border and using these building to shield the tunnels from observation. He soon receives back a message from the system

• Reporting that troop movements towards the border have been observed in Confusnia
• Reporting that relationships between Confusnia and Drazni have been deteriorating
• Suggesting that therefore a Confusnian attack might be imminent
• Suggesting the names of several other analysts who should be notified
• Offering to prepare a briefing:
  • stating the hypothesis of imminent attack
  • enumerating the warning signs which lead to the hypothesis
  • collecting the relevant documents which provide evidence for each of these warning signs
  • identifying the analysts who have already provided annotations corrobating and diagreeing with these conclusions

This scenario is science fiction, but not by much. To make it real the system must have a great deal of knowledge of the following types:

• Knowledge about standard geopolitical Crisis Scenarios
• Annotation knowledge describing documents
• Knowledge about geopolitics
• Knowledge about the roles of other members of the organization
• Know-How on how to raise alerts, find expertise and prepare reports

The HAWK system, which we propose to build as part of the HPKB program, will help the system to acquire such knowledge unobtrusively, from large numbers of experts, as part of their normal working activities.