This paper brings two research traditions together: classic experiments that measure interference from explicitly learned pairs, and modern models that represent word meaning as points in space. The authors show that a mathematical account of how activation spreads in memory matches a common measure used in distributional semantics. They then use Dutch-language tests and simulations to probe whether similarity in meaning alone—without explicit learning—can create the same slowing and errors. The results indicate that semantic closeness in embedding spaces predicts retrieval difficulty in ways that mirror human performance.
That connection has practical implications for education, user interfaces, and tools that rely on search and recommendation. If semantic overlap produces interference, then organizing content and choosing words can reduce confusion and speed recall. The article invites readers to explore how models of meaning map onto human memory and to consider how this link could expand access to learning and communication. Click through to see the experiments, the simulations, and what this means for designing information that supports human potential.
Abstract
Memory retrieval is prone to interference: when multiple concepts in memory match a given retrieval cue, recall becomes slower and less accurate. This has repeatedly been studied in fan effect experiments in which participants learn facts that are combinations of person–location pairs. These experiments manipulate the fan of a concept—the number of facts linked to it—establishing interference. The standard theoretical account invokes spreading activation: when a cue is linked to multiple memory traces, activation spreads across them, reducing the target’s retrievability. We study whether this spreading activation is triggered only by explicitly learned associations or also by semantic similarity. We show that spreading activation in the rational analysis of memory is pointwise mutual information and that similarity in at least some vector-space models of meaning approximates the same quantity, which makes such models potentially formal implementations of the rational analysis of memory. In two behavioral experiments using Dutch-language stimuli, we first replicate the classical fan effect. Experiment 2 tests whether this interference effect can be elicited through semantic similarity alone, using pretrained word embeddings to construct semantic fans. We find that items in higher semantic-fan conditions are retrieved more slowly and less accurately, mirroring patterns from Experiment 1. In a simulation, we show that similarity in embedding spaces predicts retrieval difficulty in a manner consistent with rational models of memory. Together, these results formally connect vector-space models of meaning with the rational analysis of memory, and demonstrate that semantic similarity is sufficient to produce associative interference in memory.
