New evidence for systematicity in infants’ curiosity-driven learning

Ke, H., Westermann, G., & Twomey, K. E. (forthcoming; 2018, June). New evidence for systematicity in infants’ curiosity-driven learning. In Eiterjoerge, S. (chair) Using Innovative Methods to Understand Infants’ Curiosity-Driven Learning. Symposium to be presented at the XXI ICIS International Conference on Infant Studies, Philadelphia, PA, USA.

Decades of research demonstrate that infants’ exploration is sensitive to features of the environment. For example, infants from 12 months old show high sensitivity to correlations between object features (Younger & Fearing, 1998). Importantly, complexity in the task environment interacts with learning. For example, a study by Kovack-Lesh and Oakes (2007) using 3D plastic replicas of animals found that 10-month-old infants showed a novelty preference after familiarization with pairs of different animals, but not with pairs of the identical animals. In related work using a looking time paradigm, infants have been shown to learn categories when category exemplars are presented such that between-exemplar differences are maximized, but not when differences are minimized (Mather & Plunkett 2011).

Importantly, these and related infancy studies typically employ carefully-designed experiments with complexity determined a priori by the experimenter. However, the majority of infants’ learning happens in the unconstrained, noisy, real-world environment, in which exploration and information selection are intrinsically motivated by their own curiosity. While recent empirical and computational investigations of visual curiosity-driven learning (Kidd, Piantadosi & Aslin, 2012, 2014; Twomey & Westermann, 2017) suggest that infants will systematically generate intermediate task complexity in visual curiosity-driven learning, whether infants systematically generate a particular level of difficulty during everyday object exploration is unknown.

In the current work, we explored the possibility that infants will also systematically prefer intermediate task complexity when allowed to freely explore a set of 3D objects. To test this hypothesis in a naturalistic environment, we developed a shape priming paradigm using 3D-printed toy-like stimuli and head-mounted eyetracking. We tested 12-month-old (N=18), 18-month-old (N=18) and 24-month-old (N=18) infants. Each infant was presented with 3 sets of 3D-printed stimuli (see Figure 1). Each set was a category of five objects with edges that differed in a continuum from corners to rounded (for example, from a cone to a square pyramid). Differences between exemplars were manipulated such that each object was perceptually distinct from every other object, but differences between successive stimuli in the continuum were controlled. During the experiment, on each trial infants were first allowed to play with one exemplar from one of the extreme ends of the category for 15 s, followed by the rest of the objects in the same set for 30 s. Infants’ first touches and subsequent exploratory sequences were coded offline.

Data show that regardless of age group infants show an identical pattern of object selection, first selecting exemplars of greatest perceptual difference from the prime (see Figure 2). Thus, infants’ explicit object selection diverges from their implicit exploratory behavior in existing looking time studies, but is nonetheless systematic, raising the important question for future work of how visual and physical exploration are related in curiosity-driven learning. Overall, this study offers new evidence that infants as young as 12 months actively impose structure on their learning environment outside the constrained lab environment.