Eight Reasons to Doubt the
Existence of a Geometric
Module
Nora S. Newcombe
Temple University
When Sociobiology Met
Cognitive Psychology

Modular mind

Adaptive pressure works to select specific
mental abilities



Massive modularity
Core knowledge
Innateness

These evolutionarily-selected modules are
(naturally) innately specified
Swiss Army Knife Analogy

“The Swiss Army knife is a flexible tool
[because] it is a bundle of tools, each welldesigned for solving a different problem –
scissors for cutting paper, corkscrew for
opening wine, toothpick for cleaning
teeth…..Similarly, the human mind …
contains a large number of programs, each
well-designed for solving a different
adaptive problem: choosing a good mate,
caring for children, foraging for food,
avoiding predators, navigating a landscape,
forming coalitions, trading, defending one’s
family against aggression, and so on”—
Leda Cosmides
Innately-Specified Modules Have
Proliferated





Language acquisition
Face processing
Theory of mind
Cheater detection
Geometric module
What Do We Mean By
Modularity?

Modular cognitive systems are domain specific,
innately specified, hard wired, autonomous, and
not assembled.

Fodor (1983, p. 37)
Neural Specialization Does Not Entail
Encapsulated Modularity

Brain areas generally need to ‘talk to’ one
another to support a function
Case Study of the Geometric Module


A representation of geometric information that
guides reorientation following disorientation
That does NOT use nongeometric information
even when doing so would be advantageous
Hermer & Spelke (1996):
Search Rates for Toddlers
White Room
F
R
C
F
C
.10
.39
.08
.31
.39
.12
.49
.12
N
C = Correct
N = Near
R
R = Reversal
F = Far
N
Hermer & Spelke:
Search Rates for Adults
White Room
F
R
C
.02
.57
.41
0
F
C
0
.04
N
C = Correct
N = Near
R
R = Reversal
F = Far
.96
0
N
Language-as-Bridge Hypothesis

Adults may have a further system of representation
that is uniquely human and that emerges over the
course of development. This system may connect to
many other systems of representation, regardless of
their domain-specific content. Its operation may be
governed by rules and principles allowing the
arbitrary combination of information from distinct,
domain-specific sources….The language faculty
appears to have all the right properties to serve as
this uniquely human combinatorial system of
representation. --Hermer-Vazquez, Spelke &
Katsnelson (1999, p. 34)
Support for Role of Language



Transition to feature use at 6 years is
correlated with productive use of left and
right
Training left and right leads to feature use
Adults who do linguistic shadowing task
concurrently do not use features
Adaptive Combination Models

Various sources of spatial information



Ego-referenced: response learning and path
integration
Allo-referenced: cue learning, place learning
Weighting depends on



Salience
Certainty and variability with which information is
encoded
Validity


probabilities of finding objects given use of the
information, derived from interaction with the
environment
Weighting develops both in real time and in
developmental time
Point 1: Evidence Against Encapsulation
from Non-Human Animals



Monkeys use colored
walls and large but not
small features (a
sensible choice given
likely cue validity)
Other species
 Chickens
 Pigeons
 Fish
See Cheng &
Newcombe, PBR 2005,
for review
Without cue
With cue
46 %
4.7 %
75 %
3.5 %
3.3 %
46 %
1.5 %
20 %
Point 2: Featural Cues Are Only
Neglected in Tiny Rooms
100
90
80
70
60
50
40
30
20
10
0
-10
Toddlers
3-4 Years
5 Years
6 Years
Adults
Larger Room
Small Room
Cheng & Newcombe
(Psychonomic Bulletin & Review, 2005)
Note: Perfect Performance = 100
Point 3: Use of Features Varies for
Several Reasons

Activity



Active motion focuses spatial attention
Active motion leads to remodeling of
hippocampal firing
Nature of landmarks


More distal landmarks provide more useful
and ecologically valid information
Larger landmarks may be more salient and
more likely to be stable
Room Within Room Studies



Small waist-high
enclosure (HermerSpelke size)
centered within
“large” room
(Learmonth et al.
size)
Large room had one
colored wall
Children stay within
small enclosure
Learmonth, Newcombe, Sheridan & Jones
(Developmental Science, 2008)
How the Data Fit
An Adaptive Combination Model
Distal
Action
Target Proximal
Feature? Possible? to Feature?
Yes
Age at Success
Hermer-Spelke
No
No
6 years
Learmonth
et al.
Yes
Yes
Yes
Study 1
Yes
No
No
6 years
Studies 2 & 3
Yes
No
Yes
4 years
18 months
(earliest tested)
How the Data Fit
An Adaptive Combination Model
Distal
Action
Target Proximal
Feature? Possible? to Feature?
Yes
Age at Success
Hermer-Spelke
No
No
6 years
Learmonth
et al.
Yes
Yes
Yes
Study 1
Yes
No
No
6 years
Studies 2 & 3
Yes
No
Yes
4 years
18 months
(earliest tested)
How the Data Fit
An Adaptive Combination Model
Distal
Action
Target Proximal
Feature? Possible? to Feature?
Yes
Age at Success
Hermer-Spelke
No
No
6 years
Learmonth
et al.
Yes
Yes
Yes
Study 1
Yes
No
No
6 years
Studies 2 & 3
Yes
No
Yes
4 years
18 months
(earliest tested)
Point 4: Featural Cue Use is
Easy to Get When Lacking
Room within Room Switch-No Pen: 3 Years
% of possible instances this box
was selected
100
90
80
70
67.71
60
1st 4 Trials
2nd 4 Trials
50
40
30
20.83
20
10
4.17
7.29
0
Correct
Opposite
Near
Far
Box Selection
Learmonth, Newcombe, Sheridan & Jones
(Developmental Science, 2008)
Similar finding: Twyman, Spetch & Friedman,
(Developmental Psychology, 2007)
Point 5: Spatial as Well as Verbal Shadowing
Reduces Feature Use in Adults
Ratliff & Newcombe, Cognitive Psychology, 2007
Also--Hupbach et al., Spatial Cognition & Computation, 2007
F
*
Usual results with white room
and with colored wall but no
concurrent task
R
C
.06
.52
.33
.10
N
Point 6: New Evidence from Conflict
Paradigms


When features are moved, subjects must
choose a location based either on features or
on geometry
These paradigms reveal the fundamental
similarity of human adults to children and nonhuman animals

Ratliff & Newcombe, Psychological Science, 2008
Conflict Procedure
“There are four hiding spots in this room, one at each corner”
“I will hide a pair of keys in the same place every time”
Conflict Procedure
►4
practice trials (target & landmark stable)
► Leave the room
► Brief delay ~ 2 minutes (drawing task)
► While the participant waits outside, the
experimenter goes back into the room to
move the landmark clockwise to the next
adjacent wall
► Two conflict test trials
Where are the keys?
A
D
B
C
Experiment 1
TRAINING (Between Subjects)
N = 32
TESTING
Indirect Landmark
E
G
G
L
E
G
G
L
Small room (4x6ft)
n=8
Large room (8x12ft)
n=8
Direct Landmark
Small room (4x6ft)
n=8
Large room (8x12ft)
n=8
(Landmark = L, Geometrically appropriate = G, and Error = E)
Adaptive Combination Predictions

When forced to choose one cue over the other
(geometry vs. features), conflict test will result
in a room size effect

Distal landmarks are more valid in the larger room


Corners related to feature cues will be more likely to be
chosen in the larger room
Geometric cues are more salient in the smaller room

Geometric cues will be chosen most often in the smaller
room
Results

Significant Room Size effect (p < 0.01)


Geometric information guided reorientation in the small room
Features guided reorientation in the larger room
SMALL ROOM
TRAINING & TESTING
E
LARGE ROOM
TRAINING & TESTING
E
0
.43
G
G (.06)
.16
0
(.06)
.38
G
(.04)
.19
(.05)
L
G
.28
.56
(.06)
(.06)
(Landmark = L, Geometrically appropriate = G, and Error = E)
L
Experiment 2



How does prior experience impact feature use in a
conflict situation?
Replicates experiment 1 but switches room sizes
between training and testing
Predictions:
 Replicate no effect of landmark location
(direct/indirect)
 Training in the larger room will increase feature use
among adults reorienting in the small room
Results

Cross-experiment comparisons:


Significant effect of training
Feature use increased in small room testing from Exp. 1 (small
room training) to Exp. 2 (large room training)
LARGE ROOM TRAINING/
SMALL ROOM TESTING
E
0
.03
G
SMALL ROOM TRAINING/
LARGE ROOM TESTING
G
E
(.02)
G
.03
.94
(.02)
(.03)
.10
0
(.04)
L
.10
.81
(.04)
(.04)
G
(Landmark = L, Geometrically appropriate = G, and Error = E)
L
Conclusions


Feature use is not associative:
 No effect of feature location (direct vs. indirect) in
Exp 1 or 2
Experience is important


As expected, large room experience boosts feature use for the
small room test
Reorientation depends on process of combining
weights associated with features and geometry
 Not simply relying on the most salient cue
 Although the conflict procedure does not allow direct
assessment of combination
Point 7: Geometric Information May Be a
Special Case of Relative Information


The modularity position
predicts failure to
reorient in the absence
of geometric information
Is reorientation a more
general discrimination
of relative quantity task?
Huttenlocher and Lourenco, 2007
Non-geometric, Relative Cue
Non-relative Cues
Why Might Relative Cues Be Easiest?

Multiple cues


Area, spatial density, and number
Nominal versus ordinal scale

“Cues which specify order along a magnitude
scale (more, less) may be more easily
mapped onto spatial position (left, right) than
two distinct unordered properties (red, blue)
which are mapped separately onto spatial
position” (Huttenlocher & Lourenco, 2007)
Is This Pattern True Across Species?

Since the initial claims of
modularity were made from
animal research…

Will mice replicate this
pattern of results?
C57 mice – 4 per group
Non-geometric
Non-geometric
Non-geometric
Relative Cue
Non-Relative
Non-Relative
Results: Accuracy
Twyman, Newcombe & Gould, J. Comparative Psychology, in press
Small - Large
Yellow - Blue
Dots - Gray
81%
69%
69%
Even though there is a trend for the relative cue
group to be more accurate, there were no
significant differences between groups
Results: Trials to Criterion
Twyman, Newcombe & Gould, J. Comparative Psychology, in press
Small versus Large
12 trials (0.00)
Yellow versus Blue
38 trials (5.77)
Dot versus Gray
33 trials (6.81)
The small-large group learned the task significantly
faster than either the yellow-blue or the dot-gray
group.
Point 8: Use of Features is NOT
Merely Associative

One might have thought 7 points were
enough—but the modularity theorists have a
comeback--
The Module Strikes Back


Lee, Shusterman &
Spelke (2006) argue
that—
Search behavior
following disorientation
depends on two distinct
processes: a modular
reorientation
process…and an
associative process that
directly links landmarks
to locations (p. 581)
Return of the Jedi

Problems with Lee et al.



Small moveable landmarks
Defining quite small area
Alternative way to test the two step
hypothesis

Use of colored wall in an octagon with
alternating short and long sides to
discriminate among 3 all white corners
The Octagonal Space
A
D
B
C
Photo of the Octagon
Starting with An All-White Octagon

Questions

Can children use geometry in a more complex figure?


Can children use geometry in a figure without an axis of
principal symmetry?


YES if total GC choices > 50%
YES if total GC choices > 50%
Methodologically, have children been successfully
disoriented?

YES if correct choice NOT > average of other GC
choices
Data from All-White Octagon
Newcombe et al., Developmental Science, in press
100
100
50
0


2 3
yr yr
Total
GC
Chance
PC
50
0
2 3
yr yr
At left, we see that both 2- and 3-year-old children were
reliably greater than chance in choosing geometrically
correct corners
At right, we see a slight and non-reliable ability to choose
the correct corner from among the 4 geometrically correct
corners
Mean
GC
Data from Octagon with Colored Wall
Newcombe et al., Developmental Science, in press
100
50
0


3 5
yr yr
Total
GC
Chance
100
PC
50
0
3 5
yr yr
At left, we see that both 3- and 5year-old children were
reliably greater than chance in choosing geometrically
correct corners
At right, we now see a reliable ability to choose the correct
corner from among the 4 geometrically correct corners
Mean
GC
What About the Targets in All-White
Corners?

Correct choices reliably greater than
average of other GC corners



35% versus 14% at 3 years
38% versus 10% at 5 years
These data show that young children do
in fact use features to reorient
Can Children Use Features ALONE?
Newcombe et al., Developmental Science, in press
Three Hiding Boxes
Two Hiding Boxes
X
X
X
On these 2 conditions, children
averaged 50%, reliably greater than
chance (33%)
In this condition, children averaged
64%, reliably greater than chance
(50%)
Specific Conclusions




Strong evidence for coding of geometric information
Little evidence for an encapsulated geometric module
or the role of language in puncturing it
Good evidence that different types of spatial information
are routinely (though not invariably) combined
Combination process depends on
 Cue salience
 Encoding variability and certainty
 Acquiring information on cue validity through action
and experience
General Conclusion

We can analyze
spatial navigation
and orientation as an
evolved mental skill
without postulating
 Encapsulated
modularity
 Highly specific
innate endowment