Transcript Animated Pedagogical Agents

Animated Pedagogical Agents:
Historical Figure Applications
Bob Heller, Athabasca University
[email protected]
Mike Procter, Athabasca University
[email protected]
Presented at the International Workshop on Intelligent and
Adaptive Web-based Education System (IAWES 2007) held in
conjunction with the 15th annual International Conference on
Computers in Education, Hiroshima, Japan, November 6, 2007
Acknowledgements
Daniel Ha – chat log analysis
 Lisa Jewell – chat log analysis, content developer, test chatter
 Meggie Yang – AIML builder
 Dean Mah – Web implementation
 Billy Cheung – Graphics, test chatter
 Julianna Charchun – Chat log analysis
 Jude Onuh – AIML programmer
Funding from the Athabasca University Mission Critical Research Fund

Outline


APA Review
Historical Figure Conversational
Agents



Animated Historical Figures


Freudbot
Piagetbot
Freudbot
Concluding Coments
Animated Pedagogical Agents
Definitions


programs deployable on the internet
with an embodied representation and
interactive capabilities through one or
more modalities
APAs have an Artificial Intelligence
backend that allows the designer to
simulate communicative agent
behaviour while guiding human-agent
interactions towards pedagogical goals
and objectives
Animated Pedagogical Agents
Johnson, Rickel, & Lester (2000)
Advantages of APA
 increase the “bandwidth of communication”
 increase computer capacity to engage and
motivate
The Persona effect
Animated Pedagogical Agents:
Persona Effect
Dehn & Van Mulken’s (2000) literature
review
Major Findings
1. User’s subjective experience
2. User’s objective behaviour
3. Performance outcomes

yes
?
?
Animated Pedagogical Agents:
Persona Effect
Research limitations




variability in animation/visual presentation
dependent measures (esp. social presence)
experimental control
constrained application space
Animated Pedagogical Agents:
Persona Effect
Historical Figure Applications




Stronger test of the engagement function
Expanded set of social roles
Opportunity to use narrative
Tighter synthesis between content and
persona
Historical Figure Conversational Agents
Freudbot


Developed using AIML
Design Principles
Narrative Theory
 Psycholinguistic Theory with emphasis on
pragmatics


Content development
Test chatters
 How much ALICE?
 Pilot study

Historical Figure Conversational Agents:
Freudbot
Heller, Procter, Mah, Jewell, & Cheung
(2005)


Proof of concept
Instructional set
Historical Figure Conversational Agents:
Freudbot
Method
•53 participants (81% women) recruited
through the web
•Random assignment to instructional set
condition
•Focused on self-report measures of chat
experience and analysis of conversational log
Historical Figure Conversational Agents:
Freudbot
Findings



Neutral evaluation overall
Those who would chat again (64%) had
significantly higher ratings than those
who would not
Freudbot experience

“It was pretty cool the way it felt like I was actually
interacting with Freud... he's deceased though, yeah,
but the picture, the fast answers... made me pay
attention to the answers alot more than if I had been
simply reading a text written by someone else. Plus it
was cool to feel like I could voice my own opinion with
the most well-known psychoanalyst of all time.”
Historical Figure Conversational Agents:
Freudbot
Findings




No effect of instructional set
Historical Figure and Quizbot applications
were highest rated
Expansion was strongly supported
Need for improvement
Historical Figure Conversational Agents:
Piagetbot
Heller & Procter (2007)



Developed Piagetbot based on Jean Piaget
Leverage development costs of Freudbot
Focused on learning outcomes under two
different interfaces: Text-based vs. Chatbased
Historical Figure Conversational Agents:
Piagetbot
Method
• 59 participants (85% women) recruited
through the web
• Random assignment to interface condition
(text vs. chat)
• Focus on self report measures and learning
outcomes
Historical Figure Conversational Agents:
Piagetbot
Text Interface
Historical Figure Conversational Agents:
Piagetbot
Chat Interface
Historical Figure Conversational Agents:
Piagetbot
Findings




Logical interface associated with better
learning outcomes.
Logical interface was associated with higher
ratings
Piagetbot and Freudbot were equally rated
Those who would chat again (60%) had
significantly higher ratings than those who
would not
Animated Historical Figures
Animating the conversational agents



Strong test of the engagement function of
APAs
Lower threshold of believability
Avoids many of the visual variables
Haptek Character Software – People
Putty

text to speech, lip sync, facial
expressions, personality parameters
Animated Historical Figures: Freudbot
Objectives



Effects of animation (no image, static
image, animated image)
Effects of targeting
Role of learning styles
Animated Historical Figures: Freudbot
Method
• 88 participants (81% women) recruited
through the web
•Random assignment to interface
condition (no image, static image,
animated image)
Animated Historical Figures: Freudbot
Preliminary Findings


The no image condition associated with
significantly higher ratings than the two
other conditions that did not differ from
each other
Those who would chat again (64%) had
significantly higher ratings than those who
would not overall and within conditions
Concluding Comments



Animation needs to better
integrated with communicative acts
Room for improvement in
conversational performance
Across all 3 studies, appox 2/3 will
chat again
Thank you
Animated Pedagogical Agents
IntelliMedia Centre for Intelligent
Systems – U of North Carolina
http://research.csc.ncsu.edu/intellimedia/



Herman the Bug
Cosmos
Whizlow
Animated Pedagogical Agents
Centre for Advanced Research on
Technology in Education (CARTE) –
USC
http://www.isi.edu/isd/carte/


STEVE
ADELE
Animated Pedagogical Agents
Teachable Agents – Vanderbilt U
http://www.teachableagents.org/overview.html
Autotutor – U of Memphis
http://www.autotutor.org/
Research on Innovative Technologies
for Learning – Florida State U
http://ritl.fsu.edu/_Website/index.asp
- Pedagogical Agent Learning Systems
Animated Pedagogical Agents
Embodied Conversational Agents Northwestern U
Media Lab – MIT http://agents.media.mit.edu/
Conversational Agents



AKA Chatbots, Chatterboxes, Talk
bots
SmarterChild
ALICE – Artificial Linguistic Internet
Computer Entity
http://www.alicebot.org/
Conversational Agents
Conversational Agents
Conversational Agents

ELIZA (1966)
“Its name was chosen to emphasize that it may be
incrementally improved by its users, since its language
abilities may be continually improved by a "teacher".
Like the ELIZA of Pygmalion fame, it can be made to
appear even more civilized, the relation of appearance
to reality, however, remaining in the domain of the
playwright.” Weizenbaum, 1966, p. 2


Like ELIZA, ALICE relies on simple
pattern matching and
iterative/incremental adjustments
Artificial Intelligence Markup Language
(AIML)
Conversational Agents

ALICE/AIML Strengths





Open source and XML
compliant
Numerous AIML parsers
Numerous front-end options
Loebner contest winner
Large user group
http://www.pandorabots.com


ALICE’s brain is free
Easy language to learn
Conversational Agents

ALICE/AIML Weaknesses




Content generation
Maintenance (AKA
Targeting)
Bending the will of ALICE
Hard language to master
Historical Figure Application
Emile
http://www.hud.ac.uk/hhs/research/emile/emileframeset.htm
Shakespeare
http://www.pandorabots.com/pandora/talk?botid=c6937cfb3e354738
Discussion


Historical figures vs. coaches
Course Assistant




Expanding the scope
Quiz pal
English as Second Language
Teacher
Speech Rehab Therapist
Animated Pedagogical Agents
Gulz and Haake (2006)


Focus on motivation & engagement as the promise of
APAs & the importance visual form and look
APA engagement can directly influence behaviour
and cognition
APAs must be believable & lifelike (like synthetic agents
in entertainment applications) to
maintain/strengthen the illusion of social exchange.
Animated Pedagogical Agents
Limited range of application
‘Constrained’ by backend systems
Historical Figure Applications
- greater potential to maximize
engagement