Implicit Category Learning of Crystal Images

1
Implicit Category Learning of Crystal Images
Joan A. Fisher and Paul J. Reber -- Northwestern
University
Experiment 1 Implicit Results
Experiment 2 Implicit and Explicit Results
Implicit Conditions
Explicit Conditions
Background

• Nondeclarative category learning has been
  studied in
• Dot Pattern Categorization
• (Reber, Stark Squire, 1998)
• Artificial Grammar Learning (Knowlton
  Squire, 1993)
• Probabilistic Classification (Knowlton,
  Squire Gluck, 1994)
• The memory systems involved in categorization can
  be dissociated. (e.g., in Knowlton Squire,
  1993)
• Perception and decision-making may function
  independently during categorization.
• (Maddox, Ashby Waldron, 2002)
• Reber, Gitelman, Parrish Mesulam (in press)
  compared implicit and explicit dot-pattern
  categorization. Both groups saw the same study
  and test items.
• In the implicit condition, the Categorical
  Fluency Effect (CFE) indicated less activity for
  categorical items in the posterior occipital
  cortex.
• In the explicit condition, more activity was
  found for categorical items in prefrontal and
  left occipito-temporal areas and the precuneus.
• To further explore the neural correlates of
  implicit and explicit category learning, we
  constructed a new categorization task based on 2D
  representations of 3D crystals.

Study Phase Explicit Instructions

Hex Instructions
Tet Instructions

• All groups performed above chance.
• Tet vs Hex, F(1,52)3.5
• Implicit vs Explicit, F(1,52)1.9
• Interaction, F(1,52)4.4, plt.05
• Implicit Tet similar to Hex
• Explicit Tet better than Hex
• Post-test Questionnaire
• Implicit results pooled (n54)
• Strategies reported
• Similarity to study items (n6)
• shape (n7)
• symmetry (n10)
• no conscious strategy (n15)
• No Conscious Strategy group
• performed above chance t(14) 5.5, plt.001
• suggests that they learned implicitly

• Study Phase Performance
• Hex group mean of 77.2 (2.8), t(11)10.7,
  plt.001
• Tet group mean of 81.3 (1.4), t(11)21.0,
  plt.001
• Category Test Performance
• Hex group mean of 56.0 (1.6) t(11)3.8, plt.01
• Tet group mean of 58.1 (2.8) t(11)2.9, plt.05

• Post-test Questionnaire
• Did you have a conscious strategy for determining
  which crystals were in the original category?
• Participants (2 to 4 for each dimension)
  mentioned
• symmetry
• similarity to study items
• number and shape of facets
• lines
• complexity
• dimensionality
• no conscious strategy
• Confidence was rated as low overall
• 2.5 on a 1-to-6 point scale
• These results suggest many participants learned
  implicitly. Most were unaware of what category
  information they had learned.

Experiment 2
As in previous categorization studies, changes in
processing should be seen by comparing activity
for categorical and non-categorical crystals. A
Categorical Fluency Effect might be observed in
visual processing areas related to object
identification (BA 19/37).

• Compared Implicit to Explicit category learning
  using crystals instead of dots
• (ref. Reber et al, in press Aizenstein et al,
  2000).
• Explicit group should
• -- see only two exemplars of category
• -- be given conscious strategy
• -- show similar performance to the Implicit group
• Instructions were based on real crystallographic
  category information.
• Methods
• Implicit Hex (8F, 7M) and Tet (8F, 7M) groups
  viewed the same study crystals as in Exp. 1.
• Explicit Hex (9F, 6M) and Tet (9F, 6M) groups
  viewed the training slides displayed above.
• All groups took the same categorization test.
• Post-test questionnaires were similar for all.
  However, Implicit participants described their
  conscious strategies in addition to rating which
  dimensions they had attended to.

Experiment 1 Methods
Preliminary fMRI data (n2) Event-related,
whole-brain fMRI at 3T. S1 is Implicit Hex, S2 is
Implicit Tet.

• Participants
• Undergraduate students (n24) participated for
  course credit.
• Materials
• Crystals from Shape software (shapesoftware.com)
  were presented in multiple orientations.
• Hexagonal crystals typically have 3-fold
  symmetry.
• Tetragonal crystals typically have 4-fold
  symmetry.
• Study and test items were pre-rated and balanced
  for complexity, number of facets, size, overall
  shape, brightness, and visual angles. Line
  thickness and symmetry co-varied with the
  category. None of the study phase crystals
  appeared in the categorization test.
• Experimental Procedure
• Participants were randomly assigned to one of two
  conditions.
• Hex condition participants were trained on
  hexagonal crystals.
• Tet condition participants were trained on
  tetragonal crystals.
• Study Phase -- Same-different judgments on 60
  pairs of crystals presented in different
  orientations. Participants were then told that
  all the study items came from the same category.
• Test Phase -- Participants categorized 43 novel
  categorical and 43 non-categorical crystals twice
  for a total of 172 items.
• Post-test questionnaire -- Assessed conscious
  awareness of category information.

Region of Interest S2 shows slight deactivation
in a similar area to S1
10
0
-10
R z-22
R z-22
F-value
x33
S1
S1
S2
Conclusions
Participants learned category information
incidentally from 2D representations of 3D
crystals by performing same-different judgments
on pairs of crystals presented in different
orientations. Questionnaire data suggests that
category learning relied on implicit processes.
Implicit learning conditions were matched with
explicit learning conditions in order to allow
for true comparisons using fMRI. Category
knowledge was demonstrated by participants
endorsing novel category members more often than
non-category members in a post-training
categorization test. Neuroimaging studies
are currently under way to compare implicit and
explicit conditions using fMRI. Other future
studies will likely compare implicit and explicit
category learning in children, amnesic adults,
and during the development of expertise of
crystal categories.
The authors and their research are supported by
NIMH R01 MH58748. The authors also wish to thank
Michael Levitt and Ken Paller for their help with
this project. Corresponding author
joanfisher_at_northwestern.edu
References
Knowlton, B. J. and L. R. Squire (1993). The
learning of categories Parallel brain systems
for item memory and category knowledge. Science,
262(5140) 1747-1749. Knowlton, B.J., Squire,
L.R. Gluck. M. (1994). Probabilistic
classification learning in amnesia. Learning and
Memory, 1 106-120. Maddox, W.T., Ashby, F.G.
Waldron, E.M. (2002). Multiple attention systems
in perceptual categorization. Memory
Cognition, 30(3) 325-339. Poldrack, R. A., J.
Clark, et al. (2001). Interactive memory systems
in the human brain. Nature, 414 546-550. Reber,
P. J., C. E. L. Stark, et al. (1998). Contrasting
cortical activity associated with category memory
and recognition memory. Learning Memory 5(6)
420-428. Reber, Gitelman, Parrish Mesulam (in
press). Dissociating implicit and explicit
category knowledge with fMRI. Journal of
Cognitive Neuroscience. Skosnik, P. D., Gitelman,
D.R., Parrish, T.B., Mesulam, M-M. Reber, P.J.
(2002). Neural correlates of artificial grammar
learning. NeuroImage, 17 1306-1314.