1 usc information sciences institute gil & kim interactive knowledge acquisition tools: a...
DESCRIPTION
3 USC INFORMATION SCIENCES INSTITUTE Gil & Kim Our Previous Work in Knowledge Acquisition: The EXPECT Project at USC/ISI: EXPECT architecture for knowledge-based systems exploits highly declarative representations [Swartout & Gil, KAW-95], [Gil & Melz, AAAI-96] [Blythe et al, IUI- 01] Research focus: interactive knowledge acquisition (KA) tools that help end users to develop knowledge bases Deriving models of knowledge interdependencies to detect knowledge gaps and errors [Kim & Gil, AAAI-99] [Kim & Gil, IUI- 2000] [Kim & Gil, AAAI-2000] KA dialogue scripts to guide users by following up on effects of complex changes [Gil & Tallis, AAAI-97] [Tallis & Gil, AAAI-99] [Tallis, IJHCS-2001] Exploiting background theories to understand how new knowledge fits [Blythe, IJCAI-2001] [Blythe, AAAI-02]TRANSCRIPT
1USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Interactive Knowledge Acquisition Tools:
A Tutoring Perspective
Yolanda GilJihie Kim
USC/Information Sciences Institute
August 9, 2002
2USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Motivation: Investigate Synergies between Instructional Systems and Acquisition Tools
Instructional System
AcquisitionTool teaches
teaches
?
?
GoodTutoringPrinciples
GoodLearningPrinciples
SOFTWARE USER
3USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Our Previous Work in Knowledge Acquisition:The EXPECT Project at USC/ISI:
EXPECT architecture for knowledge-based systems exploits highly declarative representations [Swartout & Gil, KAW-95], [Gil & Melz, AAAI-96] [Blythe et al, IUI-
01] http://www.isi.edu/expect
Research focus: interactive knowledge acquisition (KA) tools that help end users to develop knowledge bases Deriving models of knowledge interdependencies to detect
knowledge gaps and errors [Kim & Gil, AAAI-99] [Kim & Gil, IUI-2000] [Kim & Gil, AAAI-2000]
KA dialogue scripts to guide users by following up on effects of complex changes [Gil & Tallis, AAAI-97] [Tallis & Gil, AAAI-99] [Tallis, IJHCS-2001]
Exploiting background theories to understand how new knowledge fits [Blythe, IJCAI-2001] [Blythe, AAAI-02]
4USC INFORMATION SCIENCES INSTITUTE Gil & Kim
EXPECT: A User-Centered Framework for Developing KBSs
Method instantiator
Knowledge BaseDomain
ontologiesand factual knowledge
Problemsolving methodsDomain
dependentKBS
KBS compiler
Knowledge-BasedSystem
InterdependencyModel (IM)
EXPECT Ontologies and Method libraries
KA toolsEMeD
Plans(PLANET)
Evaluations and Critiques
Evaluation PSMs
Resources(OZONE)
DomainontologiesCYC/Sensus
Upper
NL Editor
Instrumentation
Dialogue plans(KA Scripts)
PSMTool
5USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Brief Overview of Representative KA Tools (I)
CHIMAERA [McGuinness et al 2000] Acquisition of concepts, relations, instances Diagnoses faulty definitions
EXPECT [Blythe et al 2001] Acquisition of problem solving knowledge Exploits dialogue scripts, interdependency models, bg k
INSTRUCTO-SOAR [Huffman & Laird 1995] Acquisition of task models in Soar Situated NL instruction is mapped to PSCM [Newell et al. 1991]
KSSn [Gaines & Shaw 1993] Acquisition of concepts, rules, data Based on personal construct psychology [Kelly 1955]
PROTOS [Bareiss et al 1990] Acquisition and classification of new cases Learning indexes to categories
6USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Brief Overview of Representative KA Tools (II)
SALT [Marcus & McDermott 1989] Acquisition of constraints and fixes for configuration design Exploits Problem Solving Method/ Task (Role-limiting approach)
SEEK2 [Ginsberg et al. 1985] Acquisition of rules Uses verification and validation techniques
SHAKEN [Clark et al. 2001] Acquisition of process models User interaction based on concept maps [Novak 1977]
TAQL [Yost 1993] Acquisition of SOAR rules Editor for high level language for PSCM [Newell et al. 1991]
TEIREISIAS [Davis 1979] Acquires and classifies new cases Learning indexes to categories
7USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Open Challenges in KA Users remain largely responsible for the acquisition
process Decide where, what, when, how, why to enter knowledge System checks errors, may have some short-term acquisition
goals Ideally, KA tools should have student-like skills:
Formulate and pursue learning goals Keep track of lessons and progress Assess how much they are learning and how useful k is If teacher is not so great, still capable of learning
8USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Instructional Systems and Acquisition Tools: What Are the Synergies?
Instructional System
AcquisitionTool teaches
teaches
?
?
GoodTutoringPrinciples
GoodLearningPrinciples
SOFTWARE USER
SupplementStudent’s limitations
SupplementTeacher’s limitations
9USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Teaching/Learning principle Tutoring literature
Start by introducing lesson topics and goals
Atlas-Andes, Meno-Tutor, Human tutorial dialog
Use topics of the lesson as a guide
BE&E, UMFE
Subsumption to existing cognitive structure
Human learning, WHY, Atlas-Andes
Immediate Feedback SOPHIE, Auto-Tutor, Lisp tutor, Human tutorial dialog, human learning
Generate educated guesses Human tutorial dialog, QUADRATIC, PACT
Keep on track GUIDON, SHOLAR, TRAIN-TutorIndicate lack of understanding Human tutorial dialog, WHY
Tutoring and Learning Principles Relevant to KA [Kim & Gil, ITS 02] (I)
10USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Teaching/Learning principle Tutoring literature
Detect and fix “buggy” knowledge SCHOLAR, Meno-Tutor, WHY, Buggy, CIRCSIM
Learn deep model PACT, Atlas-AndesLearn domain language Atlas-Andes, Meno-TutorKeep track of correct answers Atlas-AndesPrioritize learning tasks WHYLimit the nesting of the lesson to a handful
Atlas
Summarize what was learned EXCHECK, TRAIN-Tutor, Meno-Tutor
Provide overall assessment of learning knowledge
WEST, Human tutorial dialog
Tutoring and Learning Principles Relevant to KA [Kim & Gil, ITS 02] (II)
11USC INFORMATION SCIENCES INSTITUTE Gil & Kim
ASSIMILATEINSTRUCTION
TRIGGERGOALS
PROPOSESTRATEGIES
PRIORITIZEGOALS &
STRATEGIES
PRESENTATIONDESIGN
USER INTERFACE
KNOWLEDGE ACQUISITION BACKEND
Five Main Functions of KA Tools
Knowledge Base
12USC INFORMATION SCIENCES INSTITUTE Gil & Kim
ASSIMILATEINSTRUCTION
TRIGGERGOALS
PROPOSESTRATEGIES
PRIORITIZEGOALS &
STRATEGIES
PRESENTATIONDESIGN
USER INTERFACE
KNOWLEDGE ACQUISITION BACKEND
Guidance Exploited by KA Tools Guidance fromKnowledge Base
Problem Solving & Task Knowledge
Domain Knowledge
General Background Knowledge
Example Cases
Guidance fromMeta Knowledge
Knowledge Repres.Model
Diagnosis & Debugging Principles
Tutoring & Learning Principles
13USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Acqu. goals
Acqu. strats
GuessGenerators
InteractionGuidelines
OperationalPrinciples
PrioritySchemes
GeneralTutoring
&Learning
Principles
KnowledgeEditor
Dialogue
- Goals & Strats- State- Suggestions- History
USER INTERFACE
KNOWLEDGE ACQUISITION BACKEND
Tutoring and Learning Principles in KA Tools: Basic Conceptual Framework
Knowledge Base
ASSIMILATEINSTRUCTION
TRIGGERGOALS
PROPOSESTRATEGIES
PRIORITIZEGOALS &
STRATEGIES
PRESENTATIONDESIGN
14USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Tutoring and Learning Principles Implicit in KA tools
KSSnAssess learned knowledge
Summarize what is learned
EXPECTPrioritize learning tasks
SEEK2Keep track of answers
Learn domain language
Learn deep models
EXPECT,CHIMERATAQLDetect and fix “buggy” K
INSTRUCTO-SOAR
INSTRUCTO-SOAR
Indicate lack of understanding
Keep on track
EXPECTTEIREISIASGenerate educated guesses
EXPECTTEIREISIASINSTRUCTO-SOARPROTOSImmediate feedback
PROTOS, SALTTEIREISIASPROTOSSubsumption to existing cog. structure
SALTEXPECTSEEK2SALTUse topics of the lesson as a guide
EXPECT, SEEK2Introduce topics & goals
Design Presentation
PrioritizeGoals & Strats
ProposeStrategies
TriggerGoals
AssimilateInstruction
Tutoring/Learning principle
Limit the nesting of lessons
15USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Tutoring and Learning Principles in KA Tools Observation: Some learning and tutoring principles
are used in some aspects of the dialogue by some tools
Opportunity: Incorporate principles more thoroughly in all aspects of the dialogue
Observation: These principles are implicit in the tool’s code and thus are limited
Opportunity: Exploit declarative representations of learning state, goals, and strategies
16USC INFORMATION SCIENCES INSTITUTE Gil & Kim
USER INTERFACE KB
Proactive Dialogue Window
Active Acquisition
Goals&
Strategies
Awareness Annotations
SLICK Dialogue Manager
KBState
Dial.History
SLICK (Skills for Learning and Interactively Capture Knowledge)
KNOWLEDGE ACQUISITION BACKEND
Tutoring&
LearningPrinciples
17USC INFORMATION SCIENCES INSTITUTE Gil & Kim
Conclusions Analysis of existing KA tools shows they use
tutoring/learning principles Sparsely Implicitly
Current capabilities of KA tools can be improved by: Representing tutoring/learning principles declaratively Organizing the dialogue around lesson topics Keeping track of how knowledge improves through dialogue Exposing what knowledge has been assimilated and what
areas need improvement or testing Assessing their competence and confidence on question
answering