About this paper symposium

---

Panel information
Panel 1. Attention, Learning, Memory

Paper #1
Investigating the Dynamics Between Childhood Cognitive Control and Three Types of Statistical Learning
Author information	Role
Dr. Matthew Daniel Johnston, University of Edinburgh, United Kingdom	Presenting author
Jessica Dolan, University of Edinburgh, United Kingdom	Non-presenting author
Nicolas Chevalier, University of Edinburgh, United Kingdom	Non-presenting author
Abstract
As children grow older, they are expected to become autonomous and engage in increasingly complex behaviors. Such self-directed activities are maintained by the development of cognitive control – i.e., top-down processes that support goal-directed regulation of attention and actions. Despite its importance, underdeveloped cognitive control may be advantageous in early childhood, as less top-down guidance allows children to have diffuse attention, processing a wider range of environmental stimuli (Thompson-Schill et al., 2009). Relatedly, immature cognitive control may facilitate the ability to implicitly extract environmental regularities, i.e., statistical learning (Saffran, 2020). Empirical work with adults supports this notion that statistical learning interacts with cognitive control in a competitive manner, with latent executive function scores showing moderate and negative correlations with statistical learning (Pedraza et al., 2024). Though the past two decades has witnessed a proliferation of developmental research into statistical learning (Forrest et al., 2023), little work has extensively explored how cognitive control develops alongside statistical learning. The primary aim of this study was to explore statistical learning’s relation with cognitive control in childhood. Within developmental psychology, statistical learning has frequently been conceptualised as a unitary process. However, it has recently been suggested that statistical learning comprises different but interrelated processes, which differ in the input being extracted from the environment (Thiessen et al., 2013). Most developmental research has focused on children’s ability to learn ‘conditional’ relations between stimuli that are adjacent to each other (e.g., A predicts B), using triplet-based tasks (Arciuli & Simpson, 2012). However, people are also sensitive to conditional regularities between elements that are not immediately adjacent to each other (e.g., in XAY, XBY, and XCY sequences, learners detect that X predicts Y). Additionally, people are sensitive to ‘cue-based’ statistics (e.g., learning that certain perceptual cues indicate the presence of a hidden property). A secondary aim of our study was to explore the multifaceted nature of statistical learning in childhood by comparing performance in three tasks that tapped learning of different regularities: adjacent conditional, non-adjacent conditional, and cue-based statistics. 165 5-12-year-olds (43.6% female) completed these three statistical learning tasks as well as three tasks gauging different aspects of cognitive control (Figure A1). A composite cognitive control measure was obtained using principal component analysis. Across all three statistical learning tasks, we found that children were sensitive to the regularities (i.e., faster responses for predictable over unpredictable trials; Figure A2). Notably, differences between predictable and unpredictable trials in all three tasks appeared to be age-invariant. Unlike Pedraza et al., (2024) there was no association between cognitive control and statistical learning in the non-adjacent conditional task. Additionally, there was no relationship between cognitive control and cue-based learning. However, lower cognitive control scores predicted greater sensitivity to the regularities in the adjacent conditional learning task. Interestingly, there was no correlation across the statistical learning tasks. Our results can be viewed as an important step towards unravelling the multifaceted nature of statistical learning and understanding how its distinctive subtypes may hold separate developmental trajectories and relations to cognitive control.

Paper #2
Learning Information Outside of Task Focus: a Developmental Study of Associative Learning and Cognitive Control
Author information	Role
Rachel Foster, University of California, Davis, United States	Presenting author
Yuko Munakata, University of California, Davis, United States	Non-presenting author
Abstract
Successful implementation of cognitive control develops during childhood and supports learning of relevant information through the inhibition of irrelevant information (Diamond, 2013). Children’s reduced cognitive control engagement thus allows for increased learning of irrelevant information (Plebanek & Sloutsky, 2017). However, in real-world environments, children cannot process all task-irrelevant information. What cues drive children’s attention outside of the task focus? Given that children are sensitive to environmental regularities, children may use co-occurrence as a cue to drive attention beyond the task focus. Increased learning of information that co-occurs with task relevant information has been observed in other populations with reduced cognitive control engagement, such as older adults, adults who fail to pay attention during the task, and adults with high impulsivity (Campbell et al., 2010; Decker et al., 2022; Landau et al., 2012). We thus investigate whether children learn co-occurring information outside of task focus and to a greater degree than adults. Additionally, while most prior work capitalised on natural variations in cognitive control, we manipulate cognitive control engagement to test its causal role in learning of co-occurring information. Children (mAge = 6.4, n= 49, 46% female) and adults (mAge = 20, n=50, 52% female) sorted stimuli by shape or color. The task-irrelevant information on a given trial (e.g., the color of the stimulus if the task was to sort by shape) varied systematically: the color and shape pair co-occurred 84% of the time for high-contingent and 16% of the time for low-contingent stimuli. If participants learn co-occurring information, they should respond more quickly and/or accurately to high than low contingent stimuli. Participants first completed a Single Task block (sorting all stimuli by shape or by color) to validate and compare learning effects in children and adults. Participants then completed two Task-Switch blocks to test the causal role of cognitive control in learning of co-occurring information. In the Task-Switch Proactive Possible block, but not in the Task-Switch Proactive Impossible block, the sorting dimension was cued ahead of each trial, which increases engagement of proactive control in children and adults (Chevalier et al., 2020). In the Single Task condition, children and adults learned co-occurring information outside of task focus based on speed and accuracy (ps <.05). Children’s learning was marginally greater than adults based on speed (χ2 = 3.078, p =.079) but not accuracy (Figure B1). In the Task-Switch conditions, both children and adults showed decreased learning of co-occurring information when they could engage proactive control compared to when they could not, based on accuracy (z = 2.766, p <.01) but not speed (Figure B2). Thus, learned associations may be one factor that drives children’s attention outside of task focus. In addition, we find suggestive evidence that children demonstrate greater learning effects than adults, and some evidence for a causal role of increased cognitive control engagement on decreased learning of co-occurring information. We discuss remaining questions about distinct reaction time and accuracy effects, and more broadly about the role of implicit learning processes in the emergence of cognitive control.

Paper #3
Competition Between Predictive Processes and Prefrontal Cortex Functions During Development
Author information	Role
Dr. Dezso Nemeth, INSERM, France	Presenting author
Abstract
Human learning and predictive processing rely on multiple cognitive systems associated with distinct brain structures. These systems do not always interact cooperatively; at times, they may compete in optimizing performance. Previous research has demonstrated that reducing the engagement of prefrontal cortex-mediated attentional processes can enhance non-declarative learning and predictive processing. This antagonistic relationship has been supported by neuropsychological studies, non-invasive brain stimulation, and functional brain connectivity analyses (e.g., Ambrus et al., 2020; Park et al. 2022; Pedraza et al. 2024; Smalle et al., 2022; Virág et al. 2025). These studies form the basis of the competition theory. This theory provides a unique and useful framework for understanding developmental studies on implicit statistical learning and predictive processing. In our research, we aimed to elucidate the developmental trajectory of statistical learning through both cross-sectional and longitudinal studies. We measured statistical learning and predictive processes by a highly reliable probabilistic sequence learning task with non-adjacent second-order dependencies (Farkas et al. 2024). Our cross-sectional studies (N=421, N=270) showed better statistical learning performance before adolescence compared to later ages. Cross-sectional studies, while valuable, have limitations and are not ideal for answering developmental questions comprehensively. To address these gaps, we assessed statistical learning in the same participants -within-subject design - at ages 7, 8, 11, and 14 (N=135). Using linear mixed models and latent class analyses, we observed a decline in statistical learning with age and found that higher executive function levels were associated with this decline. Our developmental, neuropsychological, and neuroscientific findings together reveal a complex antagonistic relationship between prefrontal functions and implicit statistical learning, supporting the 'less is more' hypothesis in predictive processing during cognitive development (Janacsek et al., 2012; Juhász et al. 2019; Tóth-Fáber et al. 2024). These results highlight the competitive nature of brain systems in cognitive processes and may have significant implications for developing new strategies to enhance human learning and predictive processing.

Paper #4
How Working Memory Shapes Early Attention and Learning
Author information	Role
Qianqian Wan, Ohio State University, United States	Presenting author
Vladimir Sloutsky, Ohio State University, United States	Non-presenting author
Abstract
Early in development, attention tends to be more distributed, and learning is less goal-directed. While this distribution can result in less efficient learning, it also enables young children to excel in specific areas. For example, children learn languages more easily than adults (Newport et al., 2001; Toro et al., 2005), acquire implicit statistical sequences and structures better than adults (Janacsek et al., 2012; Turk-Browne et al., 2005), and are less susceptible to learning traps, such as the formation of stable false beliefs (Blanco et al., 2023). These advantages suggest that distributed attention may have significant evolutionary value in supporting early development and human intelligence. Historically, children’s difficulty with selective attention—the ability to stay focused and filter out distractions—has been primarily attributed to failures in inhibitory control (Pritchard & Neumann, 2011; Tipper et al., 1989; Unger & Sloutsky, 2023). However, we argue that selective attention may not be fully autonomous and requires a guiding “relevance map” – a model of the task and task dimensions that are relevant and which are not. At the very minimum, this guidance requires (a) learning the relevance map and (b) maintaining it in working memory. Therefore, it is possible that either failure to learn the map or to maintain it in working memory also contributes to distributed attention early in development. The two presented studies were designed to distinguish the “failure of filtering” from the “failure of guidance” possibilities. In Study 1 (N = 129, 48% female), conducted with 4-to-6-year-olds and adults, we introduced a category learning task, with category dimensions varying in relevance. To reduce filtering demands, all dimensions were initially occluded at the start of each trial. If children’s distributed attention is driven solely by filtering or inhibition failure, they should have succeeded in selectively uncovering only the most relevant dimension. However, 4-year-olds consistently uncovered most or all dimensions during the learning and testing phases, with selectivity increasing with age (see Figure 1). Importantly, children learned the relevance, as they consistently started uncovering the most relevant dimension (as shown in the x-axes of Figure D1), yet sampled more than needed. This finding suggested that the persistent distribution of attention does not stem exclusively from filtering failure or failure to learn the relevance map. In Study 2, we tested the possibility that distributed attention results from working memory immaturity (specifically, failure to maintain the relevance map). We imposed a working memory load on adults (N = 75, 43% female), where participants learned categories while performing a secondary task. Importantly, under the working memory load, adults tended to distribute attention, similar to young children under typical task conditions (see Figure D2). Together, these studies provide causal evidence that children’s distributed attention is not simply a failure of filtering but a result of cognitive control immaturity that includes working memory. The fact that children under standard conditions and adults under working memory conditions sampled more information than was needed highlights how early cognitive limitations support broad sampling that, while less efficient, may offer adaptive advantages in childhood.

⇦ Back to session