Abstract
Statistical regularities in our visual environment can influence memory formation. For actively maintained representations, the quality of memory representations may be distorted due to distractors, higher order structure, or previous trial exposure (e.g., Brady & Alvarez, 2015; Brady & Tenenbaum, 2013; Huang & Sekuler, 2010). Here, we investigated whether visual long-term memory (VLTM) is similarly subject to distortions driven by statistical regularities. Neural evidence supports the possibility that the same retention and retrieval operations occur for items in both long-term storage and active maintenance (Öztekin, Davachi, & McElree, 2010), suggesting that regularities may distort memory representations past the span of working memory. In the current experiment, participants studied 408 sequentially presented real-world objects and were then tested on their memory for the original color of each object using a continuous color report. During encoding, we biased the sampling of each object's color such that fifty percent of objects were selected from the same randomly determined 90° portion of a color wheel (i.e., rich quadrant). The remaining objects were randomly sampled from the other three quadrants of color space. No matter the original color of the object, participants were significantly biased towards the mean of the rich quadrant during testing. This bias was present regardless of either implicit or explicit awareness of the color manipulation. Probabilistic mixture modeling revealed that these errors toward the rich quadrant could be explained by a biased guessing distribution. In addition, in another experiment where the rich quadrant was rotated 180° for the second half of the study phase, objects continued to be biased towards the original rich quadrant. This was observed even for objects that originally appeared after the rich quadrant was rotated. These findings suggest that the quality of long-term memory representations is systematically and persistently biased by statistical regularities during learning.
Meeting abstract presented at VSS 2018