Second-order correlation learning in 2- to 4-year-old children, and its underlying mechanism

Area of Science:

• Cognitive Development
• Developmental Psychology
• Computational Neuroscience

Background:

• Second-order correlation learning, inferring indirect relations from direct ones, is well-studied in older children.
• Research on this ability in children aged 4 and younger is limited.
• Previous studies suggest 2- to 3-year-olds can use second-order correlations for causal inference in simple contexts.

Purpose of the Study:

• To investigate if 2- to 4-year-olds can detect and encode second-order correlations in complex, category-like contexts.
• To determine if children use these correlations for causal inference in demanding situations.
• To explore the potential mechanisms of second-order correlation learning using a computational model.

Main Methods:

• Behavioral experiments with 2- to 4-year-old children.
• Presentation of complex, category-like contexts with multiple objects.
• Development and application of a connectionist computational model.

Main Results:

• Children aged 2 to 4 years successfully detected and encoded second-order correlations among multiple objects.
• Participants utilized these correlations to make accurate causal inferences.
• The computational model accurately replicated children's performance, suggesting representational overlap as a potential mechanism.

Conclusions:

• Children as young as two can employ second-order correlations for causal inference in complex environments.
• This ability may stem from general cognitive processes involving representational overlap.
• Connectionist modeling provides insights into the underlying mechanisms of early correlation learning.