Labels as features (not names) for infant categorization: a neurocomputational approach.

Area of Science:

• Developmental Psychology
• Cognitive Science
• Computational Neuroscience

Background:

• Labels significantly influence infant categorization processes.
• Mechanisms underlying label effects on infant categorization remain unclear.
• Two main theories: supervised name-based vs. unsupervised feature-based categorization.

Purpose of the Study:

• To investigate the mechanisms by which labels affect infant categorization.
• To present a neurocomputational model for infant visual categorization.
• To test the unsupervised feature-based approach using self-organizing maps.

Main Methods:

• Developed a neurocomputational model based on self-organizing maps.
• Implemented an unsupervised feature-based categorization approach.
• Validated the model against experimental data from Plunkett et al. (2008).

Main Results:

• The model successfully reproduced infant behavior in familiarization and testing phases.
• It mimicked infant responses using single stimulus presentations and minimal networks.
• Predicts transient learning leading to hierarchical structures and label impact influenced by similarity and sequence.

Conclusions:

• Before 12 months, labels function as additional features, not category invitations.
• This supports an unsupervised, feature-based model of early infant categorization.
• Label impact is modulated by object similarity and presentation order.