PresentationPDF Available

New Evidence for Systematicity In Infants’ Curiosity-Driven Learning

New Evidence for Systematicity
in Infants’ Curiosity-Driven Learning
Han Ke, Gert Westermann, Katherine Twomey
Han Ke
Psychology Department, Lancaster University
Infants learn from everyday
Research demonstrate that infants’ exploration is sensitive to
features of the environment.
-The correlations between object features
Familiarization / novelty preference: rich amount of previous
literatures suggested infants from 2 weeks to 12-month-old, infants
prefer familiar objects or novel objects (Hunter, Ames, & Koopman,
-The task environment interacts with learning
Ten-month-old infantslearnt categories when familiarized with
pairs of different items, but not when familiarized with pairs of
similar items (Kovack-Lesh& Oakes, 2007).
Ten-month-olds learnt categories only when between-exemplar
differences are maximized but not minimized (Mather & Plunkett,
Infants can also actively sample information, driven by their
own curiosity.
-The Goldilocks Effect (Kidd et al., 2012, 2014).
Seven- and 8-month-old infants show lowest
looking-away rate to intermediate level of
Suggests infants implicitly seek to maintain
intermediate rates of information absorption
and avoid spending cognitive resources on
overly simple or overly complex events.
Complexity (predictable)
Twomey & Westermann’s (2017) neural-computational model of infants'category learning in
looking-time tasks. Exp1, stimuli are pre-arranged, showed the model learns better when
the maximised. Exp2 the model learns even better when they can choose the stimuli
themselves. The sequence chosen by the model also showed intermediate preference.
Infants will generate intermediate task complexity (cf. Kidd, Piantadosi & Aslin, 2012).
Maximal Learning in curiosity
“Goldilocks” effect
Infancy studies typically employ carefully-
designed experiments with complexity
determined a priori.
Whether infants systematically generate a
particular level of difficulty during
everyday, curiosity-driven exploration is
therefore unknown.
A task which allows
exploration, and also lets
us control learning history
and quantify complexity
Shape priming - free play
3D printed morphed
Quantify complexity: distance between exemplars
Prime (15s)
Test (30s)
Participants: 12mo (N=18), 18mo (N=18), 24mo (N=18)
Paradigm: 6 prime-test pairs
DVs: First Touch
Touch Sequence
Looking Time
First Touch
χ2(3) = 20.91, p= .007
χ2 (3) = 22.84, p < .001
χ2(3) = 8.05, p= .045
* * *
22 10 23 33 23 11 13 39 24 17 14 31
from prime
Touch Sequence
(from the 2nd touch)
12 mo 18 mo 24 mo
type 1 type 2
type 1
type 2
type 1
Type2: Entropy, p =.018
MeanDist, p=.054(marginally significant) Type2: MeanDist p=.012,
TouchNum, p<.001
Type2: TouchNum p<.001,
Entropy, p<.001
Distance from previous touch
1 2 3
type 2
Infants show Systematic pattern in free-exploration.
First touch usually to most novel or most similar objects
Exploration patterns characterized by length, and distance between
successive objects
Overall intermediate levels of exploration sequence complexity
(Goldilocks Effect) (Kidd et al., 2012; Twomey & Westermann, 2016) )
Further possibilities for analysis:
1) Looking time from head-mounted eye-tracking.
2) Sequence of gaze switches
3) Possible computational modelling
Any other good ideas? ….
Thank you!
We appreciate all the parents and infants who took
part in our project!
ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.