An Agent-Based Model of Semantic Memory Search: Disentangling Cognitive Control and Semantic Space Organization.

Area of Science:

• Cognitive Science
• Computational Neuroscience
• Psychology

Background:

• Verbal fluency tasks show semantic clustering and switching.
• Existing models often overlook interactions between search strategies and semantic memory structure.
• The superiority of strategic versus stochastic search models remains debated.

Purpose of the Study:

• To develop and validate a novel Area Restricted Search (ARS) agent-based model for semantic memory retrieval.
• To compare ARS against Lévy Walk (LW) and Random Walk (RW) models under varying semantic structures.
• To investigate how different search mechanisms interact with semantic organization.

Main Methods:

• Developed an ARS agent-based model with incremental decision-making and local information use.
• Systematically varied semantic structure parameters (clustering, density, dispersion).
• Evaluated model performance against 50 participants using clustering, switching, and temporal metrics.

Main Results:

• ARS excelled in moderately clustered semantic structures.
• LW performed best in sparse structures, RW in dense structures, but RW deviated from human data.
• ARS demonstrated human-like performance across configurations when semantic density was constrained, unlike LW and RW.

Conclusions:

• Human-like semantic memory retrieval necessitates dynamic, adaptive search strategies.
• No single stochastic process (LW, RW) universally explains retrieval across all semantic structures.
• The Area Restricted Search (ARS) model provides a more flexible and human-like account of semantic memory retrieval.