Bridging CRU and CMR in Free and Serial Recall: A Factorial Comparison of Retrieved-Context Models

What the paper says

Retrieved-context theory posits that episodic retrieval is driven by a representation that evolves over time, tying each item to the contextual features present during encoding and later accessing those associations to guide retrieval (Howard & Kahana, 2002). Although the Context Maintenance and Retrieval (CMR) model (Morton & Polyn, 2016; Polyn et al., 2009) offers a flexible and well-tested retrieved-context implementation, addressing both free and serial recall (Lohnas, 2024), it is complex, incorporating mechanisms absent in some other retrieved-context models. In contrast, the Context Retrieval and Updating (CRU) model (Logan, 2018, 2021) provides a simpler, more streamlined specification of context-driven retrieval shown to excel in strictly ordered memory tasks such as serial recall. However, it remains unclear whether CRU's leaner architecture extends easily to unconstrained retrieval dynamics in free recall. It is similarly unknown whether CMR's added mechanisms confer meaningful advantages over CRU in serial recall. To investigate the gap between CRU and CMR, we systematically compare them, presenting them side by side and exploring how each can be viewed as a parameterized variant of the same foundational ideas. Using a factorial model selection approach, we selectively incorporate CMR-like features into CRU and compare each hybrid variant with standard CMR on free and serial recall data. We find that selectively incorporating CMR-like features substantially improves CRU's fit to free recall and that CRU's item confusion and recall termination mechanisms can help CMR capture serial recall data.