Rai Bari¹ (), Linda H. Lidborg¹, Holger Wiese¹, Anna Grubert¹; ¹Durham University

Visual working memory (vWM) capacity has been shown to decrease with increasing stimulus complexity. Faces are among the most complex stimuli, and prior work suggested that only one unfamiliar face can be actively maintained in vWM (Towler, Kelly, & Eimer, 2016). Here, we examined whether this limit increases for famous faces, for which long-term memory (LTM) representations might support vWM maintenance. Participants performed a lateralised change detection task. A memory array (200ms) contained one versus two (Experiment 1) or two versus three (Experiment 2) face identities. After an 800ms retention interval, a test array appeared (200ms), and participants indicated whether all identities were repeated or whether one had changed. In addition to behavioural accuracy, we measured the contralateral delay activity (CDA) of the event-related potential during the retention period as an electrophysiological index of the number of items actively maintained in vWM. Behaviourally, accuracy declined with increasing set size, confirming higher memory demands in higher load conditions. In Experiment 1, CDA amplitudes were larger for load-2 than load-1 trials, indicating that participants maintained two identities when required. A control experiment confirmed that these load effects were independent of low-level pictorial cues. However, in Experiment 2, CDA amplitudes were virtually identical in load-2 and load-3 trials, suggesting that vWM capacity was limited to two face identities. Overall, our findings indicate that familiarity can boost the number of complex objects maintained in vWM beyond the previously reported limit of one. Nonetheless, capacity remains sharply restricted and well below the typical 3-4-item limit observed for simple features or basic objects. Even for highly overlearned, meaningful stimuli such as famous faces, vWM capacity appears limited to two identities, highlighting a boundary on the extent to which LTM can support the active maintenance of complex visual information in vWM.

Acknowledgements: This work was funded by research grants of the Leverhulme Trust (RPG-2020-319) awarded to AG and the ESRC (ES/X002063/1) awarded to HW.