Dissociation between the Mnemonic Trace and Test Item Similarity in Recognition Memory Using Signal-Detection Theory

Memory is a complex cognitive system that can store large amounts of visual information with high accuracy (Brady et al., 2008). Various theories and models attempt to describe the structure of memory and take into account all possible factors affecting it in order to create accurate predictions about the effectiveness of tasks (Luck & Vogel, 1997; Wilken & Ma, 2004). Various forms of testing can be used to assess memory. One of the most popular is the n-alternative forced choice task (n-AFC), in which participants are first asked to remember a list of objects, and then, in the testing phase, should indicate the previously seen object (target) among the others (distractors). One of the important factors influencing the performance in an n-AFC task is the similarity between the target and the distractors. Brady and Störmer (2024) showed that random selection of stimuli is insufficient, since the effectiveness of the task decreases with high similarity between objects. It is important to select distractors that are as different from the target as possible for a more precise memory estimation, otherwise the similarity factor may affect the performance. This finding is in line with eyewitness identification studies, in which a participant, after watching a video of a robbery, must indicate the perpetrator among a set of suspects (Shen et. al., 2023). It has been shown that under certain conditions, choosing the most dissimilar fillers (distractors) allows to reduce the number of false alarms, while keeping the hit-rate intact (Colloff et al., 2021). The Target Confusability Competition model (TCC) suggests that two predictors, which are the memory strength and the psychophysical similarity function, play the main roles in memory tasks (Schurgin et al., 2020). It was demonstrated that the model predicts the data with high accuracy. Moreover, a recent work (Utochkin et al., 2024) indicates the possibility of applying TCC to real-world objects (table, turtle, etc.). Due to memorability, which implies that certain objects are remembered better (or worse) than others by various subjects highly consistently (Bainbridge, 2019), it becomes possible to build stable psychophysical similarity functions for real-world objects. Utochkin and et al. (2024) demonstrated that when the similarity function, that is operationalized as the mutual positions of the familiarity distributions within the Signal-Detection theory (SDT) which are extracted from 2-AFC tasks, is taken into account, it is possible to predict the probabilities of choosing each object in a 4-AFC very accurately. In this research, we continued the study of Utochkin et al. (2024), showing a high consistency between the TCC predictions and the real data: the psychophysical similarity function “shrinks” proportionally, as the memory strength is reduced. Furthermore, we show that performance is highly stable among participants both within and across memory strength levels. To sum up, this work, carried out within the SDT, indicates the importance of controlling for the similarity among the target and the distractors in recognition memory experiments, as we show that similarity functions are highly invariant across different levels of memory strength.