How to say ‘no’ to a false memory: Leaky and noisy evidence accumulation during associative read-out

The associative-read-out model (AROM) is an interactive activation model (IAM) of semantic and episodic memory, which has already predicted behavioral and neural data during recognition memory. The core AROM is a deterministic model containing the original IAM and a semantic layer, where long-term associations between words are defined by word co-occurrence statistics. Episodic traces on the other hand are reflected by the resting level activations of semantic word units. However, the AROM lacks an explicit decision layer. Here, we extend AROM with a layer encompassing linear leaky competing accumulators, where evidence accumulation is driven by the activation from the episodic-semantic layer. We fit the resulting AROM+ with individual episodic memory traces and decision parameters and show that it can account for accuracies and response times in a recognition memory task. Further it mechanistically shows that false and veridical memories are driven by the combination of the semantic and episodic memory signals. When exploring decision-related vs. mnemonic explanations of the N400, we find that the amplitude of the N400 is driven by episodic-semantic activation at frontal, central and posterior locations. This amplitude modulation is strongest over frontal electrodes supporting the view that the frontal FN400 is also driven by memory signals. In addition, the amplitude of the FN400 was influenced by the response criteria. In sum, the AROM+ provides a convenient tool for disentangling episodic, semantic, and executive processes in recognition memory.