The Effect of Generation and Interaction on Robust Learning

1
The Effect of Generation and Interaction on
Robust Learning

• Robert G.M. Hausmann
• Kurt VanLehn
• Pittsburgh Science of Learning Center
• Learning Research and Development Center
• University of Pittsburgh

2
Explaining Examples

• Prompting
• Paraphrase
• Self-explain
• Example Type
• Complete
• Incomplete

• The Generation Hypothesis
• The Coverage Hypothesis

• The Interaction Hypothesis
• The Coverage Hypothesis

3
The Interaction Hypothesis

• The interaction itself increases learning gains,
  even if the set of learning events covered by
  dyads and solos is exactly the same.
• Potential Explanations of the hypothesis (Rogoff,
  1998)
• Process of negotiating meaning with a peer
• Appropriating part of the peers perspective
• Building and maintaining common ground
• Articulating their knowledge
• Clarifying it when the peer misunderstands

4
The Coverage Hypothesis

• Learning should be equivalent for peers and solo
  learners, provided
• Both forms of instruction must cover the same
  information.
• The student must attend to that information.
• Similar proposals
• Transfer performance depends on mastery, not path
  (i.e., direct instruction vs. discovery learning)
  (Klahr Nigam, 2004)
• Different types of instruction lead to different
  knowledge structures but similar performance
  (Nokes Ohlsson, 2005)
• If within ZPD, then dialog monolog (VanLehn et
  al., in press)

5
Studies of Dyad vs. Solo

• Chi Roy (in press) example study problem
  solving
• Dyad gt solo when both solving and watching a
  video of a tutor/tutee pair solving the same
  problem.
• Many problem solving
• Self- vs. interactive explanations (Ploetzner,
  Dillenbourg, Praier, Traum, 1999)
• Newtonian Physics (Kneser Ploetzner, 2001)
• Conceptual Engineering (Hausmann, 2006)
• Hundreds more
• None example studying

6
Method

• Participants
• Physics LearnLab
• United States Naval Academy (N100)
• Materials
• Andes homework system
• Domain electrodynamics (electric magnetic
  fields)
• Robust Learning Measures
• Duration immediate (experiment), short delay
  (chapter exam), long delay (final exam)
• Transfer chapter final exam isomorphic
  problems
• Preparation for learning magnetism homework

7
Design

• Natural Solo prompts to keep working, but no
  processing advice (control for Hawthorn effects).
  
• Explain Solo prompts to self-explain
• Natural Dyads prompts to keep working together,
  but no collaborative processing advice.
• Explain Dyads prompts to generate joint
  explanations

Prompting
8
Procedure
Problem4 Immediate Posttest
Problem3 Intermed. Posttest
Problem2 Intermed. Posttest
Problem1 Warm-up Problem
9
Data Sources

• Andes log files Homework (before after)
• Andes log files Experiment
• On-screen activities Experiment
• Coded interactions (McGregor Chi, 2002)
• Novel or Repeated knowledge component
• Individual or jointly generated
• If individual, record speaker/listener (Hausmann,
  Chi, Roy, 2004)

10
Predicted Results
The Interaction Hypothesis
The Coverage Hypothesis
11
Learning event space
Process Line
12
How should prompting to explain affect path
choice?

• Read line (Solo)
• Explain ? Exit, with learning
• Not explain ? Exit, without learning
• Read line (Dyad)
• Neither explains ? Exit, with little learning
• A (B) explains
• B (A) comprehends ? Exit, both learn
• B (A) fails to comprehend ? Exit, A (B) learns
• A B co-construct an explanation ? Exit, both
  learning

Increase?
13
How should interaction affect path choice?
Accountability, so this decreases

• Read line (Dyad)
• Neither explains ? Exit, with little learning
• A (B) explains
• B (A) comprehends ? Exit, both learn
• B (A) fails to comprehend ? Exit, A (B) learns
• A B co-construct an explanation ? Exit, both
  learning

Probability of having the right knowledge
14
How should interaction affect path effects?

• Read line (Dyad)
• Neither explains ? Exit, with little learning
• A (B) explains
• B (A) comprehends ? Exit, both learn
• B (A) fails to comprehend ? Exit, A (B) learns
• A B co-construct an explanation ? Exit, both
  learning

Less feature validity
Partner not present at post-test, ? moderate
learning gains?
15
Questions/Feedback
16
Why might dyads choose the right paths more
frequently than solos?

• Collaborators may be more engaged than the solos
  
• ltsocial psych?gt accountable? Responsible?
• The union of the collaborators knowledge has
  fewer gaps, so they more often finds explanations
• Heterogeneous groups outperform homogeneous
  groups (Howe, Tolmie, Rodgers, 1992)
• Diverse knowledge increases probability of taking
  good paths.

17
Learning-event Space Solo

• Read line
• Explain Exit, with learning
• Not explain Exit, without learning

18
Learning-event Space Dyad

• Read line
• Neither explains Exit, with little learning
• A (B) explains
• B (A) comprehends Exit, both learn
• B (A) fails to comprehend Exit, A (B) learns
• A B co-construct an explanation Exit, both
  learning