General precedes specific in memory representations for structured experience.

Abstract

Decades of work has shown that learners rapidly extract structure from their environment, later leveraging their knowledge of what is more versus less consistent with prior experience to guide behavior. However, open questions remain about exactly what is remembered after exposure to structure. Memory for specific associations-transitions that unfold over time-is considered a prime candidate for guiding behavior. However, other factors could influence behavior, such as memory for general features like reliable groupings or within-group positions. We also do not yet know whether memory depends upon the amount of experience with the input structure, leaving us with an incomplete understanding of how statistical learning supports behavior. In 4 experiments, we tracked the emergence of memory for itemitem transitions, order-independent groups, and positions by having 400 adults watch a stream of shape triplets followed by a recognition memory test. We manipulated how closely test sequences corresponded to the input along each dimension of interest, allowing us to isolate the contribution of each factor. Both item-item transitions and order-independent group information influenced behavior, highlighting statistical learning as a mechanism through which we form both specific and generalized representations. Moreover, these factors drove behavior after different amounts of experience: With limited exposure, only group information impacted old-new judgments specific transitions gained importance later. Our findings suggest statistical learning proceeds by first forming a general representation of structure, with memory being later refined to include specifics after more experience.

Memories of structured input become increasingly distorted across development

“…Consistent with previously published findings (Finn et al, 2019;Jung et al, 2020;Raviv & Arnon, 2018;Schlichting et al, 2016), we demonstrate general age-related improvements in performance underpinned by gains in the ability to correctly identify exposure triplets as "old." We replicate past work (Forest et al, 2021) using a simplified, child-friendly paradigm to show that young adults reference both specific transition and general grouping but not within-triplet position information when making memory judgments. Crucial for understanding how learned representations shift with development, we demonstrate that while our youngest learners (age five) already use direct transition information in service of decisions, only beyond age eight do learners make use of derived groups.…”

“…There is an extensive body of literature documenting the ability of even infants to learn such statistical structures (Saffran et al, 1996); however, due to the nature of past memory assessments, the particular representations on which their behaviors are based remain ambiguous. Recent work in adults has shown that despite the consistency of the statistical stream and the many stimulus repetitions that might bias the system toward specifics (Kensinger & Schacter, 1999; a similar effect has been observed in children, see Brainerd & Reyna, 1996), mature learners nevertheless also express flexible, distorted memories for the triplet groupings (Forest et al, 2021). With respect to change over development, while past work has largely shown improvements in statistical learning ability throughout childhood (Janacsek et al, 2012;Jung et al, 2020;Raviv & Arnon, 2018;Saffran et al, 1997;Schlichting et al, 2016;Shufaniya & Arnon, 2018), the particular reason behind such behavioral gains remains unknown.…”

Attention Shifts to More Complex Structures With Experience

“…Because no prior research has directly addressed the question at hand, we did not have an a priori estimate of the effect size of the crucial time-by-complexity interaction for a formal power analysis. Still, we decided on a sample size of 75 prior to data collection, as this is enough to detect moderate effect sizes (Cohen’s d = 0.4 or more in 80% power; Batterink et al, 2015; Forest et al, 2021), which is conservative and smaller than the typical effects observed in various statistical-learning paradigms. It is also in line with past research, including studies designed to answer questions about individual differences (e.g., Siegelman, Bogaerts, & Frost, 2017).…”

“…Still, the information that is medium in complexity depends on a learner’s own experience and should therefore shift as a learner gains experience in their environment. This is because humans rapidly learn the statistical structure of their environments (Armstrong et al, 2017; Batterink, 2017; Forest et al, 2021; Saffran et al, 1996). Regardless of the structure prioritized initially, continued experience with structure should render it redundant as learning occurs.…”

Online abstraction during statistical learning revealed by neural entrainment from intracranial recordings

“…Finally, different aspects of learned structure can be measured. For example, memory for the temporal order of items within a statistical unit (e.g., triplet) can be dissociated from memory for the item groupings ( Park et al, 2018 ; Forest et al, 2022 ), and these distinct types of memory may be supported by different underlying neural representations ( Davachi and DuBrow, 2015 ; Henin et al, 2021 ). Although providing evidence of learning overall, the current study, and the basic neural entrainment design it employed, is insensitive to these differing underlying representations.…”