Preschoolers' information search strategies: Inefficient but adaptive

Abstract

Although children's sensitivity to others' informativeness emerges early in life, their active information search becomes robustly efficient only around age 10. Young children's difficulty in asking efficient questions has often been hypothesized to be linked to their developing verbal competence and growing vocabulary. In this paper, we offer for the first time a quantitative analysis of 4- to 6-year-old children's information search competence by using a non-verbal version of the 20-questions game, to gain a more comprehensive and fair picture of their active learning abilities. Our results show that, even in this version, preschoolers performed worse than simulated random agents, requiring more queries to reach the solution. However, crucially, preschoolers performed better than the simulated random agents when isolating the extra, unnecessary queries, which are made after only one hypothesis is left. When additionally isolating all the unnecessary queries, children's performance looked on par with that of the simulated optimal agents. Our study replicates and enriches previous research, showing an increase in efficiency across the preschool-aged years, but also a general lack of optimality that seems to be fundamentally driven by children's strong tendency to make unnecessary queries, rather than by their verbal immaturity. We discuss how children's non-optimal, conservative information-search strategies may be adaptive, after all.

Keywords: active learning; cognitive development; efficiency; information search; strategy.

Figure 1

Stimuli display and illustration of the search path. These 16 cards are grouped into three hierarchical levels in a 4-by-4 grid display: 2 superordinate levels (i.e., animals and plants), 4 basic levels (i.e., fish, birds, trees and flowers), and 8 subordinate levels (i.e., owls, parrots, clownfish, shark, pine/Christmas trees, apple trees, roses and sunflowers).

Figure 2

Expected number of queries as a function of age. The shaded areas represent 95% confidence intervals. The blue dotted lines represent the performance of the simulated optimal agents, and the red dashed lines represent the performance of the simulated random agents. Note that the performance of the simulated optimal and random agents are always significantly different from each other. Only the left figure was fitted by a Negative Binomial model, while the other figures were fitted by Poisson models.