Semantic Retrieval Dynamics Are Shaped by Context and Creativity: Effects of Category Coherence in Verbal Fluency Tasks

Verbal fluency research has revealed much about how people search semantic memory, yet most studies use well-defined taxonomic categories (e.g., "animals") and rarely examine how retrieval changes when categories lack clear boundaries or shared features. In the present study, participants (N = 266) completed a verbal fluency battery based on Category Coherence Theory, spanning taxonomic, thematic (e.g., "items at a party"), goal-derived (e.g., "things to save in a fire"), and relational prompts (e.g., "barriers"). Because flexible semantic retrieval has been linked to creativity, we measured divergent thinking to test whether it shapes retrieval across category types, controlling for intelligence. Beyond fluency counts, we computed semantic search metrics using embedding-based approaches capturing local progression (forward flow), global semantic spread (divergent semantic integration), atypicality, and information-theoretic uncertainty (entropy). We found that category structure robustly predicted every metric. Fluency decreased monotonically as category coherence decreased (Taxonomic > Thematic > Goal-derived > Relational), reflecting easier access to exemplars when category boundaries are clear. In contrast, other semantic search metrics peaked for intermediate-coherence prompts (thematic/goal-derived), suggesting that some constraint supports semantic exploration, but too much or too little hinders it. Divergent thinking predicted higher scores across all metrics, but it only interacted with category structure when fluency was controlled, suggesting creative individuals show richer retrieval when category coherence provides limited guidance. These findings demonstrate that semantic retrieval reflects both stable individual differences and category-imposed constraints, and that embedding- and information-theoretic metrics offer useful tools for characterizing semantic search across fluency contexts.