Rapid Knowledge Base Development Using Intelligent Agents

Period of Performance: 04/28/2000 - 01/28/2001

$99.9K

Phase 1 SBIR

Recipient Firm

Alphatech, Inc.
6 New England Executive Park
Burlington, MA 01803
Principal Investigator

Abstract

Building large knowledge bases is currently a time-consuming, labor-intensive process, during which highly trained knowledge engineers translate the knowledge of subject matter experts (SMEs) into formal knowledge representations. We propose to develop an intelligent agent that automates this knowledge engineering function. The agent will allow an SME with no special training in knowledge engineering to develop formal knowledge representations using natural language, enabling much more rapid knowledge base development.Key technical challenges include specifying the grammatical and semantic knowledge required to support knowledge entry, and developing an efficient, scalable approach to natural language interpretation. To address the first challenge, we will make our agent instructable: it will employ machine learning techniques to improve its knowledge of natural language by analyzing its interactions with an SME. To address the second challenge, natural language interpretation will be guided by knowledge entry widgets: structured templates for entering axioms in accordance with useful, recurring axiom schemas.Phase I will demonstrate the feasibility of the approach by constructing a small prototype and using it to model integrated air defense systems. Phase II will expand the prototype and extend it to support all stages of knowledge base development. The technology developed under this program will be of immediate benefit to DARPA's Rapid Knowledge Formation initiative, and will also have general applicability to rapid knowledge base development for a wide range of military and civilian problems.