Authors
Larry Zhang, Michael N. Jones
Abstract
Elucidating the mechanisms that underlie clustering and switching behavior is essential to understanding semantic memory search and retrieval. Hills, Jones, and Todd (2012) proposed a model of semantic foraging based on the observation that statistical signatures in memory search resemble optimal foraging in animal behavior. However, the original model was postdictive in explaining when a switch would occur, as opposed to predictive, and was agnostic as to the cues used by humans to make a decision to switch from local to global information. In this paper, we proposed a switching mechanism, Semantic Scent, as a predictive model underlying such behavior. Semantic Scent extends optimal foraging theory, reproducing the same switch behavior observed animal foraging behavior in memory search. We evaluated Semantic Scent against competing models including Random Walk and Fixed Count to determine its effectiveness in classifying switches made in fluency tasks. A quantitative model comparison between the switch models demonstrated Semantic Scent’s superior performance in fitting human data. These results provide further evidence of the importance of optimal foraging theory to semantic memory search.
Key Findings
-
Predictive Switch Mechanism: Developed the first predictive (rather than postdictive) model for determining when switches occur during semantic memory search, addressing a critical limitation in the field’s understanding of clustering and switching behavior.
-
Semantic Scent Model Performance: The Semantic Scent model significantly outperformed competing models, achieving a Bayesian Information Criterion (BIC) of 6903.93 compared to Random Walk (7419.88) and Fixed Count (34193.32), with a Bayes Factor > 10¹² indicating very strong evidence in favor of the model.
-
Optimal Foraging Extension: Successfully extended optimal foraging theory by incorporating a scent-based switching mechanism analogous to how animals decide when to abandon depleted resource patches, bridging animal ecology and cognitive science.
-
Accurate Cluster Boundary Prediction: Unlike the Random Walk model which predicted spurious switches within clusters, Semantic Scent accurately predicted switches at the beginning and end of stable categorical clusters, demonstrating superior alignment with hand-coded norms.
-
Theoretical Advancement: Provided evidence that memory search mechanisms may have been exapted from primitive foraging mechanisms that evolved for searching food resources in physical environments, supporting the hypothesis that cognitive and physical search share common underlying principles.
Methodology
The researchers developed and tested the Semantic Scent model using a comprehensive experimental design that addressed the predictive limitations of previous foraging models in semantic memory research.
Model Development: The Semantic Scent model operates as a power function based on residual proximity of items to the just-produced item in distributional semantic space, raised to a sensitivity parameter based on Shepard’s Universal Law of generalization. The model calculates switch probability as P(Switch|X,N) = 1/(1+∑cos(X,Yi)^λ) where Y corresponds to the N most similar items to the current item X in semantic space.
Comparative Analysis Framework: Three models were systematically compared: (1) Semantic Scent using proximity-based switching decisions, (2) Random Walk based on stochastic traversal of semantic networks with transition probabilities, and (3) Fixed Count as a baseline model switching after fixed numbers of items. All models were evaluated on the same distributional semantic space using Wikipedia Word2Vec embeddings to ensure fair comparison.
Data and Evaluation: The study utilized data from 141 undergraduate participants who completed semantic fluency tasks, producing 5,079 total item transitions. Performance was evaluated using extended Troyer Norms for switch classification, with optimization procedures tailored to each model’s parameters (neighborhood parameter N and sensitivity parameter λ for Semantic Scent; jump parameter ρ for Random Walk; fixed count K for baseline).
Quantitative Assessment: Model performance was assessed using Bayesian Information Criterion (BIC) across all item transitions, with additional qualitative analysis examining specific clustering examples to evaluate how well each model predicted switches at cluster boundaries versus within clusters.
Validation Approach: The researchers conducted both quantitative comparisons using likelihood-based metrics and qualitative analyses of specific fluency sequences to demonstrate how Semantic Scent better captures the temporal dynamics of cluster formation and abandonment compared to competing approaches.
Impact
The Semantic Scent model represents a significant theoretical and methodological advancement in understanding memory search mechanisms with broad implications for cognitive science, clinical assessment, and computational modeling.
Theoretical Contribution: This work resolves a fundamental limitation in optimal foraging approaches to memory search by providing the first predictive mechanism for switch behavior. The model’s success strengthens the theoretical connection between animal foraging and human cognition, suggesting that exploration-exploitation trade-offs are fundamental to both physical and mental search processes.
Clinical Applications: The predictive capabilities of Semantic Scent have important implications for clinical assessment, particularly given that clustering and switching metrics are sensitive diagnostic indicators for Alzheimer’s disease, Parkinson’s disease, and schizophrenia. A predictive model enables more sophisticated analysis of memory search patterns in clinical populations and could improve early detection of cognitive impairments.
Methodological Innovation: By moving from postdictive to predictive modeling, this research establishes a new standard for memory search models and demonstrates how computational approaches can capture the real-time dynamics of cognitive processes rather than merely explaining them after the fact.
Computational Modeling Advancement: The Semantic Scent mechanism provides a concrete instantiation of how semantic proximity influences decision-making during memory retrieval, offering insights that could be integrated into broader cognitive architectures and artificial intelligence systems that need to search through knowledge representations.
Future Research Directions: The model’s framework opens new avenues for investigating individual differences in memory search strategies, cross-linguistic variations in semantic organization, and the development of memory search capabilities across the lifespan. The predictive nature of the model also enables hypothesis-driven research about how various factors (fatigue, cognitive load, clinical conditions) might alter switching thresholds and search efficiency.