Nested Timescales of Looking and Learning: Visual Exploration and Simulations of the Shape Bias

• In Event: 3-036 - 30 Years of Shape Bias Research: Lessons for Word Learning, Atypical Development, Categorization and Beyond

Integrative Statement

The “shape bias” refers to the tendency of children and adults to generalize novel names for novel solid objects by similarity in shape. The accumulation of over 30 years of research on the shape bias has resulted in a relatively clear theoretical picture linking the development of the bias to the statistical regularities of the early noun vocabulary. The idea is that similarities among the individual categories of things children learn to name early build a set of statistics that train their attentional systems and facilitate the learning of even more words—children learn how to learn words. This rich literature presents a compelling illustration of the multiple nested timescales of cognition. In this talk I will use empirical data from multiple studies with children as well as computational simulations to argue that understanding of these nested timescales is fundamental to understanding cognition and cognitive development.

Specifically, I will begin with data examining the moment-by-moment decision processes of young children in a novel noun generalization task. We use a looking-while listening procedure to track the length and sequences of 15- to 40- month old children’s looks in a novel noun generalization task with 3-Dimensional, solid exemplars (rather than images on a screen as in some prior work). Data collection is ongoing but preliminary findings suggest that although many children demonstrate a strong shape bias, there are revealing differences in their patterns of visual exploration as the decision process unfolded. For example, children who demonstrate a stronger shape bias (Shape Responders) spent more time examining the shape-matching compared to material-matching test object from the start of a trial (Figure 1). Children who did not show a systematic bias (Chance Responders), on the other hand, looked equally to the shape and material test objects. Once data collection is complete we will examine the relation between the detailed statistics of children’s vocabularies and their visual exploration to understand how associations between perceptual properties, category organization and naming support noun generalization in-the-moment.

These new data will be discussed in the context of other empirical and computational work demonstrating how the details of task structure such as the number of test objects, the nature of the warm-up trials, and the specific question asked, change noun generalizations. These data illustrate that the task can change how a learner connects current stimuli with prior individual representations and results in behaviours that are smartly tuned to context and history. Likewise, at longer timescales, data and model simulations will illustrate how these processes change as new representations are added on the basis of prior learning, altering what subsequent input can draw upon. Together this work sheds light on several central issues in the field: 1) the mechanisms that support learning from the statistics of a richly-structured environment, 2) the nonlinear interaction of real-time decision processes and stored representations, and 3) how these processes change over learning and development. This is the rich contribution of the extensive shape-bias literature to our understanding of cognitive development.

Authors

• Larissa K Samuelson, University of East Anglia
  Presenting Author

• Megan Lorenz, The University of Iowa
  Non-Presenting Author

• Joshua G Dickinson, University of East Anglia
  Non-Presenting Author

• Rachel Tremlin, University of East Anglia
  Non-Presenting Author

• Jessica Bates, University of East Anglia
  Non-Presenting Author

• John Spencer, University of East Anglia
  Non-Presenting Author