Context, cortex, and associations: a connectionist developmental approach to verbal analogies

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Developmental Psychology

Background:

• Verbal analogy problems (A:B as C:[D1|D2]) are crucial for assessing analogical reasoning.
• Developmental shifts from associative to relational responding are observed in children.
• Brain damage, particularly in temporal and frontal lobes, impacts analogical reasoning.

Purpose of the Study:

• To develop a PDP (Parallel Distributed Processing) model for binary choice verbal analogy problems.
• To explain the developmental shift in reasoning strategies.
• To model the effects of specific brain damage on analogical reasoning performance.

Main Methods:

• Training a recurrent neural network on item-relation-item triples.
• Simulating anterior temporal damage by deleting network connections.
• Simulating frontal damage by altering activation maintenance mechanisms.

Main Results:

• The model demonstrates an emergent developmental shift from associative to relational responding based on environmental statistics.
• Learned statistics allow gradual acquisition of relational knowledge, overcoming association frequency biases.
• The model replicates performance degradation after simulated anterior temporal damage and associative dominance after frontal damage.

Conclusions:

• The PDP model offers a unified computational account of developmental changes and neurological impacts on analogical reasoning.
• Emergent relational reasoning arises from learning environmental statistics, not explicit analogy training.
• The model highlights the distinct roles of temporal and frontal structures in analogical processing.