A predictive framework for evaluating models of semantic organization in free recall

Area of Science:

• Cognitive psychology
• Computational neuroscience
• Human memory research

Background:

• Semantic associations influence episodic memory search.
• The precise structure and influence mechanisms of these associations are not fully understood.
• Existing models lack detailed accounts of semantic structure in memory search.

Purpose of the Study:

• To investigate the role of semantic structure in episodic memory search.
• To compare different models of semantic organization (LSA, GloVe, WAS).
• To determine the nature of cues driving memory organization.

Main Methods:

• Developed a likelihood-based model-comparison technique.
• Integrated semantic structure models into the context maintenance and retrieval (CMR) model.
• Evaluated model variants based on their ability to predict recall sequences.

Main Results:

• Word association spaces (WAS) demonstrated the highest predictive power for recall sequences.
• Evidence suggests distinct item and context cues drive semantic and temporal organization.
• A single context cue does not solely explain memory organization.

Conclusions:

• WAS provides a more accurate representation of semantic structure in memory search than LSA or GloVe.
• Memory search is organized by a combination of semantic and temporal cues, not a unified context cue.
• Findings offer critical constraints for developing comprehensive theories of human memory search.