The Connection Between Associative Memory and Semantic Similarity: Evidence From Fan Experiments and Distributional...

Area of Science:

• Cognitive Psychology
• Computational Linguistics
• Neuroscience

Background:

• Memory retrieval is hindered by interference when multiple concepts match a retrieval cue, leading to slower and less accurate recall.
• The fan effect, studied via person-location pairs, demonstrates this interference by manipulating the number of facts linked to a concept.
• Current theories attribute interference to spreading activation, where linked memory traces reduce target retrievability.

Purpose of the Study:

• To investigate if spreading activation, causing memory interference, is triggered by semantic similarity in addition to explicitly learned associations.
• To formally connect vector-space models of meaning with the rational analysis of memory.
• To demonstrate that semantic similarity alone is sufficient to produce associative interference in memory.

Main Methods:

• Replication of the classical fan effect in two behavioral experiments using Dutch-language stimuli.
• Experiment 2 constructed semantic fans using pretrained word embeddings to test for interference elicited by semantic similarity alone.
• A simulation analyzed how similarity in embedding spaces predicts retrieval difficulty, aligning with rational memory models.

Main Results:

• The classical fan effect was replicated, confirming interference from learned associations.
• Items in higher semantic-fan conditions were retrieved more slowly and less accurately, mirroring the classical fan effect.
• Similarity in embedding spaces successfully predicted retrieval difficulty, consistent with rational memory models.

Conclusions:

• Semantic similarity is sufficient to produce associative interference in memory, challenging the necessity of explicit learning for such effects.
• Vector-space models of meaning can serve as formal implementations of the rational analysis of memory.
• This research formally bridges the gap between computational models of meaning and cognitive theories of memory retrieval.