Contingency of Semantic Generalization on Episodic Specificity: Variations Across Development

Abstract

Popular computational models of memory have posited that the formation of new semantic knowledge relies on generalization from memories of specific but related episodes, at least when it occurs rapidly. This view predicts a contingency between new generalizations and episodic memory. However, very young children readily accumulate semantic knowledge at a time when their episodic memory capacities are fragile. This phenomenon challenges the notion that semantic knowledge acquisition and rapid generalization are necessarily gated by episodic memory. Here, we tested whether generalization depends on memory for individual episodes in children from 3 to 8 years of age and contrasted their performance with adults. We found that the interdependence of generalization and episodic memory changed across development. Young adults' generalization success was contingent on their memories for an item linked to its episodic context. In contrast, generalization by young children was contingent on memories of the specific identity of items and the availability of the conceptual common ground linking related episodes. This age-related contrast favors models of memory that can account for the relations between rapid generalization and episodic memory in immature systems.

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

“…While transition lure sequences were not seen as frequently as triplets, participants had indeed seen these sequences during exposure; thus, endorsing them as old is correct. In general, the finding that by age five children can remember infrequent, arbitrary associations is consistent with past work showing above-chance episodic memory in this developmental period (Drummey & Newcombe, 2002;Hudson et al, 1992;Johnson et al, 2020;Ngo et al, 2021;Saragosa-Harris et al, 2021). However, other work has emphasized that children's associative memory continues to improve well beyond this period into late childhood (Riggins, 2014) and adolescence (Lee et al, 2016).…”

“…With respect to simple associations, much work has demonstrated the protracted development of associative memory-or, the sorts of memories that connect multiple elements within an experience (Eichenbaum & Cohen, 2001)-across childhood (Lloyd et al, 2009;Raj & Bell, 2010;Sluzenski et al, 2006). Memory for associations experienced just a single time may begin to emerge in a basic form as early as 18-24 months of age (Gómez & Edgin, 2016), with above-chance memory for arbitrary associations in standard lab-based tasks being observed regularly among 3-and 4-year old preschoolers (Drummey & Newcombe, 2002;Hudson et al, 1992;Johnson et al, 2020;Naus et al, 1977;Ngo et al, 2021;Saragosa-Harris et al, 2021).…”

“…Importantly, such a mechanism appears to be available even to infants (Gómez & Edgin, 2016) given their demonstrated ability to extract regularities across experiences that unfold across months or years; and relatively lagging ability to encode precise details (Keresztes et al, 2018). However, here we focus instead on another mechanism that would support fast extraction of commonalities by maintaining the underlying memories for specifics (Ngo et al, 2021) while simultaneously supporting storage of higher-order links. This type of representation (Brunec et al, 2018; may be supported by the late-developing (Calabro et al, 2020;Demaster et al, 2016;Gogtay et al, 2004;Langnes et al, 2020) anterior hippocampus via its connections with medial prefrontal cortex .…”

“…Inspired by the idea that memory for specific associations precedes their higher-order relationships, we predicted that while even young children would show memory for their exact experiences from learning, only older children and adults would show memory for triplet groups that generalize to novel orderings of grouped shapes. In important contrast to prior work (Ngo et al, 2021), we used materials devoid of semantic information to assess the emergence of these representations in the absence of relevant prior knowledge. Moreover, participants' experience was such that specific transitions were entirely reliable and consistent across repetitions, meaning that any behavioral evidence of memory for shuffled groups of items that co-occurred repeatedly represents a distortion of the input.…”

Semantic relatedness retroactively boosts memory and promotes memory interdependence across episodes

“…Third, recursive reminders create interdependence between initial and later-learned information that preserves memory for individual temporal contexts. Interdependence -as measured by whether memories are remembered or forgotten together -can develop for pre-existing or newly learned associations and generally aids memory (87)(88)(89)(90)(91). Moreover, reminders also scaffold new learning, such that recollecting changes during the formation of new memories aids their retention (70,(77)(78)(79)(80).…”