How Do Simple Connectionist Networks Achieve a Shift From "Featural" to "Correlational" Processing in Categorization?

Area of Science:

• Cognitive Science
• Developmental Psychology
• Computational Neuroscience

Background:

• Infant categorization is often described as a developmental shift from "featural" to "correlational" processing.
• Existing empirical data suggests a change in how infants process information during categorization tasks.

Purpose of the Study:

• To analyze three developmental connectionist models simulating infant categorization.
• To evaluate the theoretical underpinnings of these models in relation to behavioral data.
• To investigate whether the purported shift in infant categorization processing is an accurate reflection of underlying mechanisms.

Main Methods:

• Simulation of behavioral data using three distinct developmental connectionist models.
• Analysis of how each model replicates key features of infant categorization.
• Evaluation of the theoretical commitments and processing strategies within each model.

Main Results:

• The analyzed models successfully simulate behavioral data associated with infant categorization.
• These models replicate the observed behavioral patterns by processing correlations from the initial stages.
• The models demonstrate that the apparent shift from featural to correlational processing might not occur mechanistically.

Conclusions:

• The developmental shift in infant categorization processing from "featural" to "correlational" may be an illusion.
• Connectionist models can replicate behavioral observations without a distinct processing shift.
• The underlying computational mechanisms may process correlations throughout development, challenging behavioral descriptions of a processing shift.