Solomon: A Next-Generation QA System

1
SolomonA Next-Generation QA System
Unclassified//FOUO

• Problem 1
• Current QA systems are applicable only to early
  Information Gathering phases of the IA process.
• Answering substantive questions from later IA
  phases requires semantic understanding of the
  knowledge at hand, i.e. reasoning.
• The critical nature if IA compels answers to be
  accompanied by justifications ones that can be
  certified as normatively correct.
• Problem 2
• Current QA systems lack the capacity to answer
  questions about imaginary or hypothetical
  information and scenarios.
• Analysts work and reason with incomplete
  information. It is imperative that QA systems be
  capable of suppositional reasoning and questions
  answering in order to help focus the analysis.
• Problem 3
• QA systems are not integrated with other IA
  assistance tools, those focused on latter phases
  of the IA process.
• Integration provides the safeguard that answers
  are applied soundly during later analysis.

• Product
• A proof-of-concept QA system, Solomon, with the
  following five novel capabilities (and associated
  demonstrations thereof).
• Capabilities
• Knowledge Acquisition via Reading
• Human Computer Collaboration via Conversational
  Discourse.
• Discourse-driven Natural Suppositional Reasoning
• Generation of Defensible Answers, Rational
  Justifications, and Intuitive Explanations
• Integrated Reasoning over Visual and Textual
  Knowledge

• SOW milestones consist of a sequence of five
  demonstrations of Solomons novel capabilities
  and a sixth capstone end-to-end demonstration of
  the entire quintet.
• January 1, 2007 (completed)
• Demonstration of Knowledge Acquisition via
  Reading
• March 1, 2007 (on schedule)
• Demonstration of Knowledge Acquisition from
  Visual Data
• May 1, 2007
• Demonstration of Natural Language Generation of
  Justifications and Explanations
• July 15, 2007
• Demonstration of Interactive Conversation
  (initial)
• September 2, 2007
• Demonstration of Interactive Conversation
  (final)
• October 31, 2007
• Capstone End-to-End Demonstration

• Payoff
• The intelligence analyst has an integrated
  analysis assistant that marries the ease-of-use
  of dialog-based QA with power and certified
  correctness of automated reasoning.
• The interactive QA/IA model behind Solomon helps
  the analyst
• Avoid Bias
• Dodge deceptions
• Handle Novelty
• Produce Sound and Justified Conclusions
• Endorsements
• Related RD work in the RAIR lab is funded by
  AFOSR, DARPA and other agencies.

Unclassified//FOUO