Embed Size (px)
Citation preview
From the Structure of Experience to Concepts of Structure
Many object categories (i.e., most artifacts) criticially rely on causal properties – one reason they include members with differ-ent surface features. Here we ask:
1. What aspect of experience do causal properties of objects come from? Can they be extracted from predictive information presented in naturalistic event streams?
2. How spontaneously and automatically do we form generaliz-able causal categories?
Experiment 1: How do we assign causal properties to objects
in naturalistic event streams?
Experiment 2: Do learners sponta-neously form generalized
causal categories?
generalization testwhich objects is this most similar to?
same motion & same causality
different motion & same causality
same motion & different causality
different motion & different causality
cause+ motion+
cause+ motion-
cause- motion+
cause- motion-
0 0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
Percent of Subjects Choosing
n=18
Motion match F(1,17) = 15.36, p < .001Causal match F(1,17) = 5.52, p < .05
Why is this important?
Predictive structure is a pervasive part of experience that can be ex-tracted using straightforward learning mechanisms. But it can also be leveraged to gain abstraction, as predictive relations can be gener-alized across participating events and sensory features. Together, this could account for bot-tom-up abstraction of sensory experience and the formation of novel kinds generalizing across sen-sory features. do
wnl
oad
me!
similarity judgments using a spatial sorting task
Linear regression with motion model : c cells = 0 , others = ; causal model : m cells = 0 , others = 1, fit on individual data, betas subjected to t-test.
Linear regression with a single factor:mixture model: c cells = .5 and m cells = .5, others = 1
more sim
ilarm
ore di�erent
light causer 1
light causer 2
light reactor 1
light reactor 2
light causer 1light causer 2light reactor 1light reactor 2
C
C
m
m
Casual modelt(17) = 2.42, p < .05
Motion model t(17) = 2.39, p <.05
Mixture model t(17) = 3.53, p < .01
normalized screen distances
University of PennsylvaniaAnna Leshinskaya & Sharon L. Thompson-Schill*
other-object effect
random 3
.28
.26
cause
effect
.96
.21
.23
non-
caus
al
mov
emen
tra
ndom
2
.19
.19
.15
.31.31
.31
.29
.31
.14
.27
.27
.30
.30
.30 cause
effect
.96
.21
other-object effect
.23
random 3
.19
.19
.15
.31
.28
.31.31
.29
.31
.14
.26
.27
.27
.30
.30
.30
non-
caus
al
mov
emen
tra
ndom
2
Blue Object Green Object
Task: determine what each object causes.
Stimuli: sequence of 250 animated visual events order governed by markov chain.
1.2s
Sentence Acceptability Measure
1
1.5
2
2.5
3
3.5
4
4.5
5 n = 77
1. Ca
usali
ty (e
ffect
)
Acce
ptab
ility
Ratin
g (1
- 5)
2. Ca
usali
ty
(oth
er-o
bject
effe
ct)
Freq
uenc
y
(effe
ct >
othe
r-obje
ct ef
fect
)
3. Fr
eque
ncy
(effe
ct in
this
> ot
her o
bject
)
******
***
1 2 3 4 5definitely false definitely trueunsure
“The green object seemed to cause the multi-colored stars to appear”
1. Causality: E�ect Event
“The green object seemed to cause the bubbles to appear”
2. Causality: Other-Object E�ect Event
“When the green object was present, multi-colored stars appearing happened more often than when the blue object was present”
3. Frequency: E�ect vs. Other-Object E�ect Event
Each object appeared with all ambient events, but only one of those events also depended on its movements.
Causality can be assigned on the basis of higher-order event structure: an event de-pendency that depends on the object’s presence.
Causers Reactors
Shape to condition assignments and effect events counterbalanced
1 2 1 2cause cause cause cause
effect effect effect effect
Participants used a combination of motion and causality/predictive structure to group objects.
Task: learn as much as possible about the object and press a button when anything unexpected happens.
Experiment 3: Is causal generalization automatic?
1.2s
Task: hit a key whenever an event repeats
cause effect
random 1 random 2 random 3
cause effect
random 1 random 2 random 3
same effect object 2
random 1 random 2 random 3
cause effect
different effect object
same effect object 1
familiarity forced choice test
Event assignments counterbalanced
When the same event was predictable (v.s., radom), learning was automatically facilitated, indi-cating automatic generalization.
vs
example
Same Different0
0.2
0.4
0.6
0.8
1
* t(167) = 2.16, p = 0.032
