Memories of structured input become increasingly distorted across development - Working Paper

Abstract

As early as infancy, humans extract patterns from structured input, and demonstrate the ability to distinguish between reliably experienced patterns and new ones. However, the nature of memories that support these behaviors—and how their structure might change across childhood—remains unknown. Here, we ask what children and adults remember after exposure to a continuous stream of shapes: the particular sequence in which the shapes occurred, their higher-level group structure, or both? We showed 5- to 9-year-olds and adults (N=211) a stream of shapes comprising three triplets (groups of three shapes) that always occurred in a fixed order, followed by an old-new memory test including lure sequences that matched the exposure stream on a particular dimension (e.g., group structure). Given the early emergence of simple associative memories that increase in complexity over development, we predicted that the youngest children in our sample would remember specific shape-shape sequences, while older children and adults would additionally represent groups. After accounting for developmental improvements in overall memory, we found all ages were sensitive to specific transitions: Participants responded “old” to lures with intact shape-shape transitions at above-baseline levels. In contrast, order-independent group memory—as measured by “old” responses to shuffled triplets—was only observed in older children and adults. Our results show that while young children form memories for specific aspects of a structured experience, memory for commonalities across events is refined later—underscoring that even after identical experiences, adults and young children form different memories for those events.

Developmental Sequences Constrain Models of the Mind

“…This situation also needs investigation. We do know that some kinds of generalization appear only in middle childhood, when we see combination across multiple memories to generate new inferences, as when self-deriving new semantic knowledge (Bauer et al, 2012(Bauer et al, , 2021, inferring novel associations between items indirectly related through a common link (Schlichting et al, 2017(Schlichting et al, , 2022Shing et al, 2019), or learning general rules in statistical learning paradigms (Forest, Abolghasem, et al, 2023).…”