When Does one Exemplar Behave Like Many?

• In Event: Poster Session 05
  In Poster Session: PS 05 Topic - Cognitive Processes

Abstract

Across visual and linguistic domains, infants have demonstrated their ability to abstract commonalities across multiple exemplars using comparison (Anderson et al., 2018; Ferry et al., 2015; Duh & Wang, 2019; Gerson & Woodward, 2012; Gerken & Bollt, 2008). There are many factors that facilitate and hinder the comparison process. We focus on two in this study: age and common motion. Over the course of the first year of life, changes in vision, attention, and encoding processes play a critical role in determining what comparisons are beneficial. For instance, younger infants have poorer visual acuity (Arterberry & Kellman, 2016) and longer habituation times (Colombo & Mitchell, 2009), and make more recognition failures (Mareschal & Johnson, 2003) compared to older infants. We predict that young infants may perceive a single repeated exemplar as several extremely similar yet unique exemplars, allowing them to abstract a common relation and generalize it to new objects. In contrast, older infants who have more mature encoding processes would recognize the single exemplar across trials. This narrow breadth of training would make generalization to new objects unlikely.

We tested this prediction in Experiment 1 by presenting 3-month-olds (n = 48) and 7- and 9-month-olds (n = 48) with a single exemplar of either a same or different relation repeated 6 to 9 times during the habituation phase. In the test phase, all infants saw novel and familiar relations of new objects and objects from the habituation phase (Figure 1). As predicted, 3- but not 7- and 9-month-olds were able to abstract the relation from a single repeated exemplar (Figure 2). Three-month-olds were able to discriminate between a novel and familiar relation when instantiated with new objects, t(45) = 2.33, p = .02, d = .34, and could recognize the habituation pair, t(45) = 2.80, p = .008, d = .41. In contrast, 7- and 9-month-olds did not generalize the relation to new objects, t(46) = .91, p = .37 d = .13, but could recognize the habituation pair, t(46) = 4.64, p < .001, d = .68.

To bolster the claim that less stable encoding was responsible for younger infants generalizing the relation, we increased the processing demands of the task by varying the motion of the objects during the training phase in Experiment 2. We predict that the increased variability will cause the younger infants to fail and the older infants to succeed in generalizing the learned relation. Data collection is currently underway.

Together, these results will lend new insights to our understanding of the ability to align across exemplars. Specifically, accounting for the learner’s level is crucial for abstracting relations. Experiment 1 suggests that for younger infants, one exemplar appears as many, while older infants recognize it as a repetition of a single exemplar. For Experiment 2, we predict that this pattern may be reversed when the single exemplar is presented using variable motions during the training phase. Together, these findings will demonstrate how relational learning interacts with visual, attentional, and encoding processes across development.

Authors

• Apoorva Shivaram, Northwestern University
  Presenting Author

• Erin Morgan Anderson, Indiana University - Bloomington
  Non-Presenting Author

• Susan J Hespos, Northwestern University
  Non-Presenting Author

• Dedre Gentner, Northwestern University
  Non-Presenting Author