EXPECT effort in RKF
Yolanda Gil (firstname.lastname@example.org, (310)448-8794) Jihie Kim (email@example.com, (310)448-8769) Jim Blythe (firstname.lastname@example.org, (310)448-8251)
DARPA's Rapid Knowledge Formation (RKF) goal is to enable distributed teams of subject matter experts (SMEs) to enter and modify knowledge directly and easily, without the need for specialized training in knowledge representation, acquisition, or manipulation.
Within RKF, we are building the Knowledge Analysis module, called KANAL, which recognize possibly invalid statements entered by the user and point out what additional knowledge needs to be acquired. It accomplishes this by 1) relating different knowledge inputs among themselves and to the existing KB, and 2) detecting inconsistencies and knowledge gaps.
KANAL builds upon previous work on EXPECT using Interdependency Models (Kim and Gil AAAI-99, Kim and Gil AAAI-2000). The work on EXPECT analyzed problem-solving interdependencies, which model how different pieces of problem solving (procedural) knowledge relate to each other and to the class definitions in domain ontologies. Our initial work concentrates on process models, i.e., knowledge describing activities and subprocesses as well as their relationships.
KANAL will check knowledge in two stages. Posing questions about various features of the process model will perform static checks. Simulating the execution of the process model will perform dynamic checks. In order to perform static checks, we will maintain role, part-of relations, and type definitions. An explanation or trace of the answer to a query will be considered as a model of the interdependencies in that it reflects how different pieces of knowledge are put together to generate an answer. Dynamic checks will be done on the simulated execution of the process model. Using skolem instances, we will be able to check general scenarios of the animation of a process. The result of each animation will be used as a model of interdependencies among the sub-steps of the process model. KANAL will perform a variety of verification checks to process models, including unachieved conditions of steps in a process model, expected and unexpected effects, and disjunctive model branches. KANAL will also help users with validation checks of potential loops, redundancies, and resource use within a process model.
We are also building proactive dialogue capabilities to support knowledge acquisition from end users (see Proactive Dialogue section below).
1. Knowledge Analysis
- relating different knowledge inputs among themselves and to the existing KB
- detecting inconsistencies and knowledge gaps
2. SADL: Representing processes
The SHAKEN action language, for representing processes with loops, partially ordered events and conditional and disjunctive events in KM (Updated 2/14/01).
3. Proactive Dialogue
Recent publications on Proactive Dialogue (Updated 4/30/02)
dialog design (2001)
4. COA CP (Course of Action Challenge Problem)
Materials for CP (7/19/02)
Materials for SRI team meeting (6/3/02)
5. Our effort on EKCP (Expert Knowledge Challenge Problem) 6. Problem Solving Methods EXPECT
For more information, send mail to email@example.com.