NITHIT SINGTOKUM^1,2 (nithitsingtokum136@gmail.com), Chaipat Chunharas^2,3,4; ¹Department of physiology, Faculty of medicine, Chulalongkorn University, Bangkok, Thailand, ²Cognitive Clinical and Computational Neuroscience Center of Excellence, Faculty of Medicine, Chulalongkorn University, Bangkok, Thailand, ³Division of Neurology, Department of Medicine, Faculty of Medicine, King Chulalongkorn Memorial Hospital, Thai Red Cross Society, Bangkok, Thailand, ⁴Chula Neuroscience Center, King Chulalongkorn Memorial Hospital, Thai Red Cross Society, Bangkok, Thailand

Visual working memory (VWM) capacity is limited due to the constrained cognitive resources.To overcome these limits , VWM can leverage statistical regularities and compress similar items together. Studies of real-world objects have demonstrated that semantically related items improve memory capacity compared with unrelated items. However, it remains unclear whether it improves memory for object identity alone or also for fine-grained object details. Here, we investigated how resource availability modulates the benefit of semantic relatedness in VWM, and at which level of memory precision this benefit occurs. Participants performed a VWM task in which we manipulated VWM load using set size (2, 4, 6 items) and encoding time (150, 500, 1000 ms) manipulation, along with changing the semantic relatedness between objects. Memory precision was assessed by requiring participants to identify the studied item shown in different states against two states of the foil object. We found that when objects are semantically related, participants can remember them better than when they have no semantic relationship (F(1,29)=104.22, p<.001). But this advantage emerged only when resources were depleted. We observed an interaction between the set size and semantic relatedness (F(1.66,48.09)= 36.07, p< 0.001), but no interaction between encoding time and semantic relatedness (F(1.96,56.96)=0.27, p= 0.759). However, we didn’t observe the improvement in object state precision on trials where participants correctly identified the object. (F(1, 27)= 0.50, p = 0.484) In summary, semantic relationships improve VWM performance for object identity, particularly when resources are limited. And, these findings might correspond with the underlying mechanism observed in low-level feature chunking, where surpassing capacity limits results in diminished detail for individual representations. However, absence of an effect on state precision might be because of the 4-AFC response, which could not clearly define effect on object identity and state separately. Our following work will examine these aspects independently.