Research Article: Integrating across memory episodes: Developmental trends

Date Published: April 22, 2019

Publisher: Public Library of Science

Author(s): Yee Lee Shing, Carsten Finke, Martina Hoffmann, Anna Pajkert, Hauke R. Heekeren, Christoph J. Ploner, Alessandra S. Souza.


Memory enables us to use information from our past experiences to guide new behaviours, calling for the need to integrate or form inference across multiple distinct episodic experiences. Here, we compared children (aged 9–10 years), adolescents (aged 12–13 years), and young adults (aged 19–25 years) on their ability to form integration across overlapping associations in memory. Participants first encoded a set of overlapping, direct AB- and BC-associations (object-face and face-object pairs) as well as non-overlapping, unique DE-associations. They were then tested on these associations and inferential AC-associations. The experiment consisted of four such encoding/retrieval cycles, each consisting of different stimuli set. For accuracy on both unique and inferential associations, young adults were found to outperform teenagers, who in turn outperformed children. However, children were particularly slower than teenagers and young adults in making judgements during inferential than during unique associations. This suggests that children may rely more on making inferences during retrieval, by first retrieving the direct associations, followed by making the inferential judgement. Furthermore, young adults showed a higher correlation between accuracy in direct (AB, BC) and inferential AC-associations than children. This suggests that, young adults relied closely on AB- and BC-associations for making AC decisions, potentially by forming integrated ABC-triplets during encoding or retrieval. Taken together, our findings suggest that there may be an age-related shift in how information is integrated across experienced episodes, namely from relying on making inferences at retrieval during middle childhood to forming integrated representations at different memory processing stages in adulthood.

Partial Text

Memory enables us to use information from our past experiences to guide new behaviours. An important part of this memory function entails the process of forming inference across multiple distinct episodic experiences [1]. For example, when seeing a woman (A) with a child (B) at the playground, then seeing the same child (B) on another day with a man (C), one may infer that the woman (A) and the man (C) are related (e.g. as partners). Evidence from the literature suggests that the hippocampus supports memory inference through novelty detection and pattern completion mechanisms, that is by detecting novelty in certain features of the newly experienced event (i.e. B), and reactivating previously stored, overlapping memory (i.e. A; [2]). At the same time, prefrontal cortex is implicated in memory inference. For example, inferior frontal gryus may be involved in making inferential judgments from premise associations during retrieval [3]. Also, medial prefrontal cortex may influence memory integration by representing relevant mental models (e.g., schema knowledge, see review by [4]).

Trials in which a response was given within 300 ms were discarded from the analyses as anticipatory response. Only five children, one teenager, and two adults had anticipatory responses (less than three such trials in each one of them). The key measures consisted of percentage of correct response for a given trial type (unique DE- vs. inferential AC-trials) and median of reaction times (RTs) of correct trials of a given trial type. Unique trials are better measures of associative memory compared to the direct trials because of the overlapping structure among the direct trials. Results also remained the same when using direct trials in the analyses. Importantly, for the inference trials, we restricted the calculation to inference trials for which both direct associations (AB, BC) were correctly remembered in the corresponding direct trials. This ensures that a possible inability to make memory inference was not due to having insufficient memory for the direct associations. We first checked for cycle effect, and if there was none, data was collapsed across cycles. The mean number of inference trial (with both direct associations being correct) was 17 for children (range: 9–26 trials), 24 for teenagers (range: 11–32 trials, and 28 for young adults (range: 19–32 trials). Significance level was set at p < .05. Post hoc analysis was performed when necessary with Bonferroni correction for multiple comparisons. In this study, we compared the accuracy and reaction time of children, teenagers, and young adults on tests of memory-based inferences and memory for associations. Our results showed that young adults outperformed teenagers, who in turn outperformed children in accuracies of both trial types. This was not in line with our hypothesis that predicted a task by age interaction, in which children would perform particularly worse on the inference trials. However, in the reaction time measure, we did find that children were particularly slower than teenagers and young adults in inference trials than in association (unique) trials. In the following, we will elaborate on the implications of these results.   Source: