Program

Heather Kosakowski¹, Michael Cohen², Atsushi Takahashi¹, Boris Keil³, Nancy Kanwisher¹, Rebecca Saxe¹¹MIT, ²Amherst College / MIT, ³Mittelhessen University of Applied Science

Philosophers and psychologists have long debated the relative roles of built-in structure versus learning in the development of the human mind. Yet it is only relatively recently that evidence from whole-brain measurements of awake infants has become available to inform these debates. Previous functional magnetic resonance imaging (fMRI) research found neural responses to faces and scenes in human (n=9) and macaque (n=3) infants. But, unlike adults, these face and scene responses did not appear to reflect specialized processing for these categories because objects evoked similar responses in the same regions. Thus, initial fMRI evidence seemed to support the idea that category-specific structure emerges slowly, presumably from learning. In adults, brain regions engaged in high-level visual perception are located on the ventral surface of the brain, a difficult part of the brain to image. Thus, it is possible that category-selective responses were present in infants but went undetected due to the challenges of awake infant fMRI. Here, we built a new coil and used a high-resolution acquisition sequence to collect fMRI data from infants (2-9 months) while they watched videos of faces, bodies, objects, and scenes. We collected enough low-motion data for a functional region of interest (fROI) analysis from 19 infants. For fROI analyses, we used anatomical constraint parcels for face-selective FFA, scene-selective parahippocampal place area (PPA), and body-selective extrastriate body area (EBA) generated from adult data. Infant fROI analyses revealed that the face response in FFA was greater than the response to bodies, objects, and scenes (all Ps<0.001); the scene response in PPA was greater than the response to faces, bodies, and objects (all Ps<0.01); and the body response in EBA was greater than the response to faces, objects, and scenes (all Ps<0.05). Thus, infants have face-, scene-, and body-selective responses in the FFA, PPA, and EBA, respectively. Why do category-selective responses emerge in stereotyped locations across individuals? According to the proto-architecture hypothesis, category-selective responses emerge in areas of cortex that maximally respond to corresponding co-occurring low-level stimulus properties. Specifically, face-selective responses emerge in areas of cortex that preferentially respond to images that are high in low spatial frequency and curvilinear content whereas scene-selective responses emerge in areas of cortex that maximally respond to rectilinear and high spatial frequency content. Thus, we next measured the curvilinear, rectilinear, and spatial frequency content of our stimuli. The pattern of response in FFA, PPA, and EBA did not correspond to the pattern of low-level statistics in our stimuli, indicating that the proto-architecture account is not sufficient to explain category-selective responses in infant cortex. Although these results do not rule out a role of experience in cortical development, they powerfully constrain accounts of development: If experience is necessary for the emergence of face-, scene-, and body-selective regions of cortex, then just a few months of experience is enough.

Alexandra Rett¹, Jamie Amemiya¹, Micah Goldwater², Caren Walker¹¹University of California San Diego, ²The University of Sydney

Understanding any phenomenon requires knowing how its causal elements are related to one another. Further, recognizing shared causal structure across phenomena may be useful for understanding novel processes. For example, understanding how multiple factors contribute to cost-of-living estimates may facilitate understanding of how multiple causes also contribute to the rise of carbon dioxide in the atmosphere (Rottman et al., 2012). Although prior research provides evidence that children can learn individual causal relations (Schulz, 2012), no previous work has explored whether children recognize similar causal structures across distinct events. Here, we examine whether children match causal narratives based on shared causal structure. In Study 1, we present 64 4- to 7-year-olds (n = 16 per age group) with three-variable causal stories over Zoom. For each story, illustrated events unfold according to either a common effect or a causal chain structure (e.g. “The sun shone very bright and warm, which made the dirt dry. The dry dirt made the flowers droopy.”). We then ask children to make judgments about which stories are the most similar. Results indicate that the ability to recognize and use abstract causal structure as a metric of similarity develops gradually between the ages of 4 and 7: While we find no evidence that 4-year-olds recognize the common causal structure between events (M = .59, 95% CI [.35, .84], p >.05), 7-year-olds demonstrate a relatively mature understanding of shared causal structure when making similarity judgements (M = .91, 95% CI [.76, 1.0], p < .001). Five- and 6-year-olds show mixed success. Despite this initial evidence, it remains possible that children may have used subtle language cues within individual stories to support their judgments (e.g., noticing similar language among stories of the same structure). We address this possibility in Study 2 (ongoing), which aims to replicate the findings from Study 1 with 5- to 7-year-olds. In Study 2, we precisely control the language of each story to remove any low-level cues that children may have used in Study 1, and include short, animated videos of each to ensure that children are representing the intended causal structure. Additionally, to measure causal understanding, we ask children a counterfactual question about the events in each target story. While data collection is ongoing (current N = 33; pre-registered N = 96), the current trend is consistent with the claim that, by 7 years of age, and potentially as early as 6, children can use abstract causal structure as a metric of similarity, even in the absence of any superficial cues. That is, 6- and 7-year-olds (n = 9, M = .72, 95% CI = [.43, 1.0] and n = 15, M = .77, 95% CI = [.55, .98], respectively) tend to match stories based on their causal structure, while 5-year-olds do not (n = 8, M = .50, 95% CI = [.15, .85]). This suggests that children are using causal structure information to match stories, rather than a low-level language cue. We also find that 6- and 7-year-olds perform well on counterfactual questions (M = .89, 95% CI = [.68, 1.0] and M = .83, 95% CI = [.64, 1.0], respectively), providing additional evidence that they are using information about causal structure to match stories, rather than relying on low-level cues. These results extend the literature on causal reasoning, and open new avenues for exploring how the recognition of causal structures may support learning and transfer in children.

Damien Foinant¹, Jérémie Lafraire², Jean-Pierre Thibaut¹¹LEAD CNRS UMR-5022, Université Bourgogne Franche-Comté, ²Institut Paul Bocuse Research Center, Ecully, France

Distinguishing between foods from non-foods is a crucial human ability (Rozin, 1990). Misclassifications of foods as non-edible can lead to unnecessary rejections (e.g., an unfamiliar dragon fruit). More critically, misclassifications of non-foods as edible can lead to poisoning (e.g., household products; Schwebel et al., 2015). There is an extensive body of research describing how children’s food knowledge develops across the preschool years (see Cognition Development, Special issue on Children’s Food cognition, 2020). Here, we will focus on children’s understanding of the relation between edibility and food processing because it has been claimed that food processing (i.e., signs of human intervention such as slicing) conveys information about food edibility (Fischler, 1988). We studied this influence of food processing and its interactions with food familiarity and children’s food neophobia (i.e., the fear of uncertain eating situations) in a food versus non-food categorization task. We hypothesized that, first, children would categorize more often processed stimuli (sliced) as foods than unprocessed stimuli (whole) and, second, that children with high food neophobia scores would have poorer performance compared to other children. 137 children, aged 57.14-72.02 months, performed a computerized categorization task. They were shown a series of 16 food and 16 non-food items, whole or sliced. Each food was paired with a non-food perceptual match (i.e., similar in both shape and color). Children’s food neophobia scores were assessed with the CFRS food rejection scale (Rioux et al., 2017). Children were asked to categorize whether or not each item was a food or not by pressing a “Yes” or “No” button. Children’s hits and false alarms were recorded to compute discriminability indices and decision criteria. Results indicated that children were better at discriminating foods from non-foods in the whole condition than in the sliced condition (F = 18.63, p < .001, d = 0.74). Actually, children considered sliced items as foods more often than whole items (F = 32.75, p < .001, d = 0.98). Furthermore, children with higher food neophobia scores had poorer discrimination abilities than children with lower food neophobia (Spearman’s r = – 0.205, p = .017). Interestingly, as sown in Figure 1, the highly neophobic children displayed a more conservative criterion of decision than their more neophilic counterparts (r = 0.354, p < .001), categorizing more often actual edible items as non-foods. In conclusion, children distinguish more easily foods from non-foods when whole, and that food processing appears to be an important cue for edibility that can even lead to dangerously incorrect categorization. We also discuss the costs and benefits of being a neophilic or neophobic individual. Indeed, even if children with higher food neophobia had poorer food categorization abilities than children with lower food neophobia, highly neophobic children displayed a more conservative criterion of decision that could protect them from dangerous false positives.

Andrew Flores¹, Philip Huebner¹, Jon Willits¹¹University of Illinois at Champaign-Urbana

Children’s ability to learn distributional statistics is well established and is hypothesized to be important for many aspects of cognitive development. One question that arises for statistical learning theories and models is, what are the units over which statistics are calculated? The units that are used to calculate distributional statistics can have a significant impact on what ends up being learned. In the current work, we tested the question of whether language that is segmented into whole words or into morphosyntactic units (i.e. treating units like -s, -ed, and -ing as separate units) better predicts child vocabulary development and semantic category learning from distributional statistics. Previous work has shown that if infants segment morphological endings like -ing, this better predicts and explains the age at which children can recognize verbs (Willits, Seidenberg, & Saffran, 2014). Does treating morphological inflections as “units” in distributional learning help predict other aspects of language development? In both of the experiments described below, we used the CHILDES corpus (MacWhinney, 2000) obtained through the childes-db website (Sanchez et al., 2019). We created two conditions, a “raw” condition leaving the corpus standardly transcribed (with each token being a traditionally defined word) and a “parsed” condition splitting the inflectional morphemes -s, -ed, -ing, -ie, and -y from nouns and verbs and leaving them in the corpus as separate tokens. In Experiment 1, we compared these two versions of the corpus in terms of their ability to predict child vocabulary acquisition (using MCDI data obtained from the Wordbank website, CITATION). Previous work has shown that distributional predictors like word frequency, contextual diversity, and the proportion of known words with which a word co-occurs all significantly predict the proportion of children of a given age who produce a word (Goodman & Dale, 2004; Braginsky et al, 2020, Flores et al., 2021). We found that overall ability to predict vocabulary development is significantly higher in the morphologically parsed condition. In Experiment 2, we compared ability of the two versions of the corpus to be used by a distributional learning model to learn words’ semantic categories. Previous research has shown that distributional models learning from the CHILDES corpus can predict the semantic categories of words with high accuracy (Huebner & Willits, 2020). In this experiment, we found that the distributional models learned significantly better from the parsed version of the corpus. In fact, the parsed version of the corpus reached the level of performance of the raw version of the corpus in about one-third the time, requiring only about 2 million words of input compared to six million words. Across two experiments, we show that taking into consideration the units over which statistical learning is operating can significantly affect the predictions one makes about the capacities of distributional learning models. In support of previous research (Willits et al., 2014), predicting both child vocabulary development and semantic category learning from distributional statistics significantly improve if children are treating morphological inflections as separate units during distributional learning.

Jinyun Lyu¹, Yijin Fang¹, Stella Christie¹¹Tsinghua University

Recent evidence shows that free play can bring about learning, even as well as didactic experience can (Sim & Xu, 2017). While the mechanism behind it is not fully understood, it is believed that an exploratory component is necessary for play to effect learning. If so, a culture that favors rigidly defined play over free exploration may not be conducive to learning from play. We test this expectation in the Chinese culture, given evidence that Chinese parents act and perceive themselves as teachers and not playmates for their children (Lin & Li, 2018). In three studies we tested Chinese three year-olds’ generalization learning from play. We used an identical paradigm to Sim & Xu (2017), where children saw a machine that made a sound when certain blocks were put atop it. A rule governed the sound activation (e.g., same-shape rule: the machine made sound if the block’s shape matched the machine’s). The question was whether children can learn and generalize this rule–at test they had to select one out of three blocks to activate a previously seen machine (First-order test) and a new machine (Second-Order test). Study 1 compared children (N = 60) in Free-Play vs. Didactic conditions. Children in Free-Play were given the machines and blocks to play by themselves for 5 minutes, while children in the Didactic condition were explicitly shown which blocks activated the machines (without being told the rule). Children learned well in the Didactic condition (First-order test: 77% correct, t-test against chance .5, t(29) = 3.40, p = .002), Second-order test: 73% correct, (t(29) = 2.84, p = .008). However, children did not learn from Free-Play: accuracy of First-order test was 53% (t(29) = 0.36, p = .722), and 47% in Second-order test, t(29) = -0.36, p = .722). These results differ from Sim & Xu (2017) where U.S. three-year-olds learned the rule from both Didactic and Free Play. In Study 2 (N = 31), we asked whether promising children with reward would change their play behavior and subsequently learning outcome. It did not; 3-year-olds’ accuracy in Free-Play-Reward condition was still at chance (First-order test 39%, t(30) = -1.270, p = .214; Second-order test 45%, t(30) = -0.533, p = .598). There is evidence that Chinese parents respond to their children’s failures and mistakes, but not to successes (Ng, Pomerantz, & Lam, 2007). If so, it’s possible that children’s exploratory play depends on children’s sense of success and failure. To test this, we modified Study 1’s Free-Play condition where instead of giving a predetermined rule for the machine’s activation, the rule was determined by the child (but unbeknownst to them) by their very first try/action. Children in Study 3 were randomly assigned to either the Play-Success-First where the child’s first action became the rule, or the Play-Failure-First where the opposite happened. Everything else was identical to Study 1 Free-Play–children played freely by themselves for 5 minutes. Amazingly, initial success determines whether play effects learning: Play-Success -First children were 77% accurate (t(29) = 3.395, p = .002), while Play-Fail-First children did not learn from free play (37% accuracy, t(29) = -1.490, p = .147), just like children in the Free-Play conditions in Studies 1 and 2. Overall, our results revealed that across cultures, children can learn from free play. Culture, however, may constraint exploratory tendency during play, resulting in (at times) non-learning from play.

Andrew Young¹, Jan Rodriguez – Cruz¹, Jackelyn Castaneda¹, Michele Villa¹, Samantha Macksey¹, Kayla Nuszen¹, R.B Church¹¹Northeastern Illinois University

Instructional gestures have been shown to benefit learning in a number of domains, including mathematics, science, and language. For example, instruction about the meaning of the equal sign (i.e., mathematical equivalence) that includes gesture improves student understanding more than speech-only instruction (Cook et al., 2016; Singer & Goldin-Meadow, 2005). Despite the overall efficacy of instructional gestures, many children fail to learn from either instructional format. The present research examines whether individual differences in cognitive resources previously shown to correlate with mathematical equivalence understanding (i.e., executive functions and cognitive reflection; Young & Shtulman, 2020) predict students’ learning from instructional gestures. 2nd-5th graders (N = 162) participated in an online pretest-intervention-posttest study via Zoom. Students completed 12 mathematical equivalence problems with operations on both sides of the equal sign (e.g., 2 + 3 + 8 = _ + 6) at pretest and posttest (unique items for each test). Students also defined the meaning of the equal sign, a conceptual measure of mathematical equivalence understanding. Most U.S. elementary students provide incorrect operational definitions (e.g., “it means put the answer”) rather than correct relational definitions (e.g., “it means the same amount”; Hornburg et al., 2018). Students were randomized to speech-only or speech+gesture instructional videos used by Koumoutsakis et al. (2016). In a classroom setting, Koumoutsakis found both videos improved overall student understanding, but speech+gesture led to greater learning. Prior to the pretest, students also completed measures of set shifting, working memory, and a developmental measure of cognitive reflection (i.e., the tendency to reflect on one’s own thinking; Young & Shtulman, 2020). To assess learning, we examined the posttest performance of students who had poor problem solving (0 or 1 correct; N = 129) and an incorrect definition (N = 86) at pretest. We fit Bayesian regressions with condition (speech-only vs. speech+gesture) and children’s age, cognitive reflection, set shifting, working memory, and their interactions with condition as predictors. There were two main results. First, there was a condition by cognitive reflection interaction for problem solving, such that more reflective children benefited more from speech+gesture (over speech-only) than less reflective children. Second, there was a condition by working memory interaction for definitions, such that children with higher working memory capacity benefited more from speech+gesture (over speech-only) than children with lower capacity. Bayesian model selection suggested both interactions were important for out-of-sample predictive performance. The present results suggest individual differences in cognitive reflection and working memory predict who will learn from instructional gestures. Overall, children with greater cognitive resources were more likely to benefit, which is in contrast to suggestions that instructional gestures might be particularly supportive of students with fewer resources (e.g., Begolli et al., 2018). A better understanding of how cognitive reflection and working memory interact with features of the learning environment (e.g., speech+gesture vs. speech-only instruction; solving problems vs. defining the equal sign) will allow researchers to design more effective instruction to improve academic achievement.

Arithmetic is a complex topic involving the application of four basic operations to various types of number, including whole numbers, decimals, and fractions. Arithmetic development is similarly complex, with each type of number posing distinctive challenges to both children and researchers seeking to explain development. For example, single-digit addition involves a transition from counting to fact retrieval, multi-digit whole number arithmetic involves column algorithms based on place value, decimal arithmetic involves modified versions of these column algorithms, and fraction arithmetic involves an entirely different set of procedures. Cognitive models have been proposed to explain development in each of these domains of arithmetic, but a unified theory that explains all of them does not yet exist. To address this challenge, we extended a domain-specific theory of arithmetic to a new domain. Braithwaite, Pyke, and Siegler (2017) previously developed FARRA, a computational cognitive model, to explain children’s fraction arithmetic. We hypothesized that FARRA’s key assumptions would also apply to decimal arithmetic. To test this hypothesis, sixth and eighth grade children (N = 92) solved 12 decimal addition and multiplication problems; strategies were coded based on written work and think-aloud protocols. Consistent with the hypothesis, children displayed three phenomena analogous to ones previously observed in fraction arithmetic and explained by FARRA: (1) most errors involved overgeneralizing strategies that would be correct for problems with different operations or types of number; (2) accuracies on different types of problems paralleled the frequencies with which the problem types appeared in textbooks; and (3) children displayed four patterns of strategy use predicted by the theory. To account for these similarities between different domains of arithmetic, we created UMA–a Unified Model of Arithmetic. UMA is a theory of arithmetic learning and development implemented as a computational model. The model’s theoretical assumptions are shared with FARRA, but unlike FARRA, UMA simulates arithmetic not only with fractions, but also with single- and multi-digit whole numbers and decimals. Simulations performed with UMA displayed each of the three phenomena observed in our empirical study of children’s decimal arithmetic. UMA generates these phenomena by virtue of three assumptions: (1) arithmetic errors are caused by small deviations from standard correct procedures, of which strategy overgeneralization is the most common type; (2) the likelihood of selecting a correct procedure for a given problem depends–via a reinforcement learning mechanism–on how often similar problems have been encountered previously; and (3) continuous variation among children in the parameters governing learning leads to discrete differences in patterns of strategy use. The present findings, together with those of Braithwaite et al., (2017; see also Braithwaite at al., 2019) show that these assumptions can explain empirical phenomena in children’s arithmetic with both fractions and decimals, despite large differences in the details of arithmetic procedures in these two domains. More generally, we argue for the viability of UMA as a unified theory of arithmetic development.

Children possess remarkably rich intuitions about quantity. Without counting or measuring, they can rapidly judge which of two jars has more candy (number), which piece of licorice is longer (length), or piece of chocolate is bigger (area). Through development, they learn to attach language (number words) to these perceptual magnitudes, allowing them to verbally quantify their intuitive, imprecise representations (e.g., judging that there are “one-hundred-and-two” candies in a jar). What mechanisms support children’s interface between number words and perceptual magnitudes? We test two competing mechanisms. Under a structure mapping (SM) account, if children understand the logic that underlies how number words are attached to perceptual representations, then the very moment children can verbally estimate in any one dimension (e.g., number), they can readily estimate in others (e.g., length, area). Instead, if their interface relies on creating item-specific associations between number words and specific perceptual states, consistent with an associative mapping (AM) view, then children would need to learn to map number words to each dimension independently, with knowledge in one dimension not transferring to another without additional learning. Ninety 5- to 11-year-olds completed two tasks. In the Verbal Estimation (VE) task, children judged “how many” items they saw in number, length, and area, with novel units provided for each dimension (Figure 1). We calculated the coefficient of variation (CV; variability) and the slope (accuracy) of their mean estimates across target values shown. To test if differences in VE could be accounted for by differences in children’s underlying perceptual representations, children also completed a Perceptual Sensitivity (PS) task. Here, they judged which side of a display had more dots (number), which line was longer (length), or which blob was bigger (area), with ratio (difficulty) varying within each dimension, and accuracy (%correct) as a measure of individual differences in perceptual acuity. Across ages, children readily linked number words to novel units and these dimensions, despite some not having had formal experience in length/area estimation. They were even able to map number words to completely novel number units (i.e., “one” unit represented by 3 dots; Figure 1), which is rather remarkable considering that children had potentially competing associative links that had to be inhibited (e.g., “one” and 1 item) and that our youngest participants had not yet learned to formally multiple/divide. Further, any differences in VE were not explained by mere perception, suggesting that SM mechanisms likely support this interface. In ongoing work, we examine the extent to which children can flexibly link number words to different number units (e.g., “one” represented by 1 dot, 3 dots, or 5 dots) and demonstrate intuitive multiplication/division. Understanding this link between number words and perceptual magnitudes is especially important from perspective of development, considering children have access to perceptual representations from birth and acquire number words slowly through development. Our findings show that the link between number words and perceptual magnitudes is a major developmental achievement and one that is not built in piecemeal: once children can map language to one perceptual dimension, they can easily map number words to other dimensions and units..

Rahma Mbarki¹, Dora Kampis², Jinjing (Jenny) Wang¹¹Rutgers University-New Brunswick, ²University of Copenhagen

Humans and a variety of non-human species have demonstrated the ability to perceive numerical quantities since early infancy, suggesting that these abilities may be rooted in biological inheritance instead of explicit learning and instructions (Dehaene, 2011). In contrast, recent research using verbal descriptions and responses revealed biases in both adults and young children to think basic cognitive abilities including quantity approximation to be late emerging in humans and require learning and instruction (Berent et al., 2019; Wang & Feigenson, 2019). However, these results are limited by the method, which requires explicit reasoning about others’ mental capacities based on verbal descriptions, leaving it unclear whether such biases remain when people are asked to make intuitive judgments of others’ behavior. How do children and adults predict numerical behaviors of other agents? Specifically, do children and adults ascribe the same kind of numerical judgments they are capable of making to agents that differ in age or species? In three experiments, we explored this question in children and adults by showing participants an agent facing two different quantities at a time (e.g., 9 vs. 27 blueberries), and asking participants to predict which of these the agent will choose. The agents varied in age and species (a human adult, a human child, a human infant, an adult chimpanzee, and an adult ant). Additionally, participants completed baseline trials with no agent present and they judged which quantity was more. In Experiment 1 we found that 6- to 8-year-old children (N = 58; 33 female) performed at ceiling in these baseline trials (M = 0.98, 95% CI [0.97, 1]. Critically, a Generalized Linear Mixed-effects Model predicting participants’ choice (with a logistic link) during the test trials revealed a significant effect of Agent Type, χ2 (4) = 392.82, p<.001: children predicted the human adult and human child to choose the larger amount (M = 0.95, 95% CI [0.93, 0.98]; M = 0.84, 95% CI [0.80, 0.89]). In contrast, they predicted the human infant, adult chimpanzee, as well as the ant not to choose the larger amount reliably (M = 0.42, 95% CI [0.36,048]; M = 0.64, 95% CI [0.58, 0.69] ; M = 0.46, 95% CI [0.40, 0.52], respectively) (Fig. 1). We then asked whether this tendency to think human infants and non-human species cannot make reliable quantity judgments is specific to children in Experiment 2. We tested adults (N = 50) with the same questions and found that adults show a similar pattern, predicting that the adult and child would choose the larger amount more consistently than the other agents (M = 0.89, 95% CI [0.85, 0.93]; M = 0.84, 95% CI [0.80, 0.88]). In Experiment 2b (N = 50) we found that there was no difference between adults’ judgments for the human infant and the non-human species and the random coin flip, suggesting that participants believed these agents would pick the bowls at random. Our ongoing experiments are investigating whether children and adults attribute specifically an inability to differentiate quantities to human infants and nonhuman species, or whether they rather feel unsure about these agents’ behavior in general. Together, these studies so far point to an early emerging intuition of others’ numerical judgments varying by age and species, which may be informative for our reasoning about other minds.

Eliza Congdon¹, Elizabeth Wakefield², Miriam Novack³, Susan Goldin-Meadow⁴¹Williams College, ²Loyola University Chicago, ³Northwestern University, ⁴University of Chicago

Gestures – hand movements that accompany speech and express ideas – can help children learn how to solve mathematical equivalence problems (e.g., 5+3+2=_+2), and flexibly transfer that learning to novel problem-solving contexts (Novack et al., 2014). Yet not all children benefit equally from gesture instruction. Here, we consider two factors that may impact whether gesture leads to learning in a given instructional scenario: (1) whether a child spontaneously produces gesture when they explain their solutions to equivalence problems (Congdon et al., under review); (2) and whether the child is seeing the gesture during instruction, or doing the gesture themselves (e.g. Dargue et al, 2019). 145, 3rd grade children (Mage = 9.19 years) participated in a training study aimed at teaching children how to correctly solve mathematical equivalence problems. No children solved any pre-test questions correctly. They were asked to explain their (incorrect) problem-solving solutions as a measure of spontaneous gesture. One-third of the sample (31%) produced no spontaneous gesture on any of the six pre-test explanations (see Table 1). Next, children were randomly assigned to one of four one-on-one training conditions: Doing Action (DA), Seeing Action (SA); Doing Gesture (DG); or Seeing Gesture (SG). Children in all conditions expressed an equalizer strategy in speech (“I want to make one side equal to the other side”) and either observed or produced a grouping strategy in gesture, or a grouping strategy in action (indicating that the first two addends can be summed to arrive at the correct answer). After training, children were given a post-test with two types of problems — “trained” problems, which matched the form of problems children saw during training and measured learning directly, and “transfer” problems, which were novel in form and measured children’s ability to flexibly generalize any conceptual gains to novel contexts. Raw average scores are shown in Figure 1. Overall, posttest performance indicated that children’s spontaneous inclination to gesture during pre-test interacted with their training condition, particularly for transfer problems. Two separate models were built to predict learning performance (Model 1) and generalization performance (Model 2). Follow-up analyses on Model 1 (trained problems) revealed that pretest gesture status did not significantly predict performance in the gesture training conditions (Χ2 (1) = 2.349, p = .125) or the action training conditions (Χ2 (1) = 0.176, p = .675). However, for Model 2 (generalization problems) results showed that within the Do Action training condition, pre-test non-gesturers performed significantly better on transfer problems than pre-test gesturers (Χ2 (1) = 7.164, p = .007). The opposite pattern emerged in the Do Gesture condition: pre-test gesturers performed significantly better on transfer problems than pre-test non-gesturers (2 (1) = 4.305, p = .038). These differences did not emerge in either the See Action (Χ2 (1) = 2.056, p = .151) or See Gesture (Χ2 (1) = 1.321, p = .250) conditions. Results have implications for predicting which individuals may benefit most from gesture-based instruction, and suggest that there may be benefits to aligning instruction with children’s natural communicative tendencies.

Diqi Zeng¹, Hyesung Hwang¹, Nicole Burke², Amanda Woodward¹¹University of Chicago, ²New York University

Starting early in life, language is an important social category that children use to divide the social world: monolingual English-speaking children privilege English over foreign language speakers as friends (Kinzler et al., 2007) and informants (Corriveau et al., 2013). However, it is unclear whether children prefer English speakers or dislike foreign language speakers when questions are asked in a forced-choice comparison. Further, why children privilege information from English speakers is unclear: is it because they think foreign language speakers are not knowledgeable and/or because they like English speakers more? Moreover, how does this selectivity relate to children’s exposure to linguistic diversity? Thus, the current study used rating scales to examine children’s social preferences for and epistemic perceptions of foreign language speakers and quantified the linguistic diversity in children’s neighborhoods and social networks. Eighty-two 6-year-old English-speaking children (52% females; 49% White; 40 monolinguals; 26 exposed to other languages; 16 bi/multilinguals) with no exposure to Korean participated in the study. Children completed (1) the Liking task and (2) Knowledge tasks in counterbalanced order. In each task, children saw and heard four foreign-language (i.e., Korean) speakers and four English speakers in random order. In the Liking task, after listening to each speaker, children’s liking was assessed on a 3-point scale (1=don’t like very much, 2=maybe like, 3=like very much). In the Knowledge task, children saw along with the speaker a picture of either familiar objects (e.g., ball) or novel objects (e.g., unfamiliar objects from the novel noun database), and their judgments were assessed on a 3-point scale (1=speaker doesn’t know what this object is for, 2=maybe knows, 3=does know). In addition, children’s home zip codes were used to extract neighborhood demographics from the U.S. census, and parents were interviewed to collect demographic information about the people in children’s social networks. Children rated liking English speakers (M=2.5) more than Korean speakers (M=2.0), F(1,79)=59.4, p<.01. Children showed a significant preference above the ambivalent choice (“maybe like”) for English speakers (p<.01), but not for Korean speakers (p=.9). Children judged English speakers (M=2.4) as more knowledgeable than Korean speakers (M=2.2) regardless of object types, F(1,79)=14.6, p<.01. Both Korean and English speakers were judged as knowing familiar objects (ps< .01) but at chance (not different from “maybe know”) for knowing novel objects (Korean: p=.33; English: p=.17). Children’s liking and knowledgeability ratings of Korean speakers were correlated, r=.35, p<.01, but not so for English speakers, r=-.33, p=.8. We also found marginal evidence that monolingual children living in neighborhoods with a high percentage of foreign-language speakers rated Korean speakers as more likeable than children in neighborhoods with a low percentage of foreign language speakers (b=-.3, p=.05) and this neighborhood effect was most prominent in children who have no linguistic diversity in their social networks. Our results suggest that the foundation for social bias against foreign language speakers may be rooted in both judgments about likability and knowledgeability, and that linguistic diversity in children’s neighborhoods could play an important role in shaping children’s perceptions of foreign language speakers.

Lauren Girouard-Hallam¹, Yu Tong², Fuxing Wang², Judith Danovitch¹¹University of Louisville, ²Central China Normal University

Nearly 98% of American households with a child under 8 (Rideout, 2017) and 94% of Chinese households (CNNIC, 2019) have access to the internet, yet our understanding of children’s attitudes and beliefs about the internet is limited. The current studies explore how children’s views of the internet vary with age, experience, and culture. We developed and validated a new measure – the Children’s Internet Attitudes Scale – in order to examine these questions. Six- to 10-year-old children (N=413; Mage = 8.43) from China and the United States indicated their experience with the internet by stating if they had engaged in each of 6 different types of internet-based activities, such as watching a video or finding the answer to a question. Children then indicated their degree of agreement with each of 16 statements using a four-point scale. Upon confirming that these data were appropriate for an exploratory factor analysis (EFA), four factors were extracted based upon Kaiser’s rule (eigenvalues > 1). Four factors explained 52% of the total shared variance in the model. The four latent variables were: comfort using the internet, perceived accuracy of the internet, perceived scope of the internet, and comparison of print sources to the internet. All factor loadings were significant (ps<.001). A confirmatory factor analysis (CFA) was then performed with a new sample of 6-10 year old children from both countries (N=417; Mage=8.17) who completed the same procedure. The final model fit was adequate (Χ2 p value=.15, RMSEA=.028), confirming the loading of factors into the four latent variables. All factor loadings were significant (ps<.001). Finally, the same CFA model and sample were applied to a structural equation model (SEM) that was used to explore the factors that contribute to differences in children’s responses to questions about the four latent variables. Children’s age positively predicted their comfort using the internet, but negatively predicted their beliefs about the accuracy of the internet. Experience positively predicted children’s comfort when using the internet, and negatively predicted their preference for print sources. Culture predicted children’s beliefs about the scope of the internet, such that American children believed the internet had broader scope than Chinese children. Culture also predicted children’s beliefs about the internet’s accuracy and their preference for print sources, where Chinese children indicated that the internet was more accurate and they had a stronger preference for print sources, despite reporting similar levels of comfort with the internet as American children (all ps<.001). These results demonstrate that culture, age, and previous experience all play a role in shaping children’s understanding of the internet. Considering these factors together allows researchers to garner a robust understanding of children’s beliefs and attitudes. Age and culture are related to experience, but their contributions to children’s beliefs and attitudes regarding the internet are not mutually exclusive. These results also set the stage for future research using the Children’s Internet Attitudes Scale to explore individual differences in internet attitudes and how they relate to children’s trust in and learning from the internet and internet-based devices.

Sehrang Joo¹, Sami Yousif¹, Frank Keil¹¹Yale University

Are children naturally predisposed to believe in God(s) or other supernatural beings? Some theories suggest that children are intuitively theistic (Kelemen, 2004) and may even have a broadly agentic worldview (i.e., over-ascribing intentionality to all of nature; Rose, forthcoming). These perspectives are based in part on how children are inclined to *explain* the world around them; children’s explanation preferences in response to ‘why’ questions may suggest an affinity for understanding nature as designed by some (supernatural) agent (e.g., Kelemen, 1999). Yet existing research has largely overlooked how children may understand the *questions* that precede these answers. Here, we explore how children’s pragmatic understanding of ‘why’ questions may shape their explanation preferences, with downstream consequences for these broader theories of intuitive human cognition. Indeed, adults’ pragmatic expectations about ‘why’ questions shape their preferences for the explanations that follow. Adults (1) interpret “Why?” as asking a specific implicit question (i.e., *either* “What is the purpose of x?” or “How did x come to be?”); and (2) endorse the explanation that happens to answer the most relevant question (Joo, Yousif, & Keil, 2021, PsyArXiv). Thus, when adults endorse teleological answers (e.g., that giraffes have long necks so that they can reach the leaves of tall trees), they think that this answer can address the question (e.g., what the giraffe’s neck is for) that someone was really meaning to ask in the first place. These findings raise a critical question about children’s understanding of explanations: Do children (1) share these pragmatics about ‘why’ questions, and (2) if so, how might these expectations affect their explanation preferences (and the broader theories that revolve around them)? We investigate this question across three experiments. First, we find that children develop an adult-like understanding of ‘why’ questions around age 6 (Experiment 1), at the same time at which they develop consistent explanation preferences (Experiment 2). Collectively, these experiments demonstrate that children have specific expectations about ‘why’ questions that may relate to their explanation preferences: They understand ‘why’ questions about different kinds of objects to imply *either* a ‘how’ or a ‘purpose’ question. But do these implicit questions actually shape how children subsequently reason about the explanations that follow? Here, we find that children do think of particular kinds of explanations as answers to specific, disambiguated questions (‘how’ vs. ‘purpose’; Experiment 3). By the time they develop specific expectations regarding ‘why’ questions, children also treat teleological vs. mechanistic answers as resulting from *different* questions. These findings therefore suggest that children, like adults, may prefer ‘explanations’ that answer the most relevant implied question–regardless of whether or not these answers are best able to actually explain why something exists. In other words, children may be drawn to teleological explanations *not* because they innately think nature was intended by God(s) for this purpose, but simply because they want to know (and think others want to know) what different things may be for. This view of children’s explanation preferences thus has critical implications not only for the study of explanation, but for the broad theories of human cognition that build upon it.

Isobel Heck¹, Jesús Bas², Katherine Kinzler¹¹University of Chicago, ²Pompeu Fabra University

There are wide divides in what people look for in leaders. Research with adults underscores the extent to which people differentially prioritize strength and power vs. competence and prestige in leaders, and suggests that powerful leaders are more preferred in contexts of actual and perceived competition (e.g., Kakkar & Sivanathan, 2017; Sprong et al., 2019; Lausten & Peterson, 2015). Here, we investigated the developmental origins of preferences for powerful vs. prestigious leaders during times of competition vs. cooperation. Across three pre-registered experiments, we introduced 3-6-year-old children (N = 288; from 34 U.S. states; with racial and socioeconomic diversity) to a fictional group (the Dotis) facing a problem (a water shortage) and varied the way this problem was framed. Across experiments, children learned that the Dotis were choosing between two possible leaders: The “biggest and strongest” Doti that is “seen as being powerful” (powerful); and the “most skilled and smartest” Doti that is “seen as being respected” (prestigious) (order counterbalanced). We asked children which Doti (powerful, prestigious) they thought would be the leader. In Experiment 1, we framed the problem (between-subjects) as involving within-group cooperation (“The Dotis need to work together as a group to get more water”) or between-group competition (“The Dotis need to work against another group to get more water”). By age 3, children’s leader choices depended on framing: Whereas children in the within-group cooperation condition tended to choose the prestigious leader, children in the between-group competition condition tended to choose the powerful leader (Wald X^2(1) = 11.80, p < .001). Independent from condition, children’s choices also shifted with age: Across ages 3 to 6, children were increasingly likely to choose the prestigious (vs. powerful) leader (Wald X^2(1) = 7.92, p = .005). In Experiment 2, we replicated these findings using framings that always involved another group (between-group cooperation vs. between-group competition). In Experiment 3, we asked whether these results would extend to indirect framings of competition vs. cooperation–the relative scarcity vs. abundance of resources. Following Rhodes & Brickman (2011), we told children that to get more water, the Dotis go to a well to which many other groups also go. Between-subjects, children heard that the well had “a lot of water; always enough for all of the groups” (resource abundance) or “only a little water; never enough for all of the groups” (resource scarcity). Children more often chose the powerful leader in the scarcity (vs. abundance) condition (Wald X^2(1) = 5.62, p = .018). Children were again increasingly likely to choose the prestigious leader with age (Wald X^2(1) = 9.16, p = .002). We further examined whether children’s parents’ political orientation may influence their choices. In Experiments 1 and 2 (but not Experiment 3), older (but not younger) children with more conservative parents were more likely to choose the powerful (vs. prestigious) leader (regardless of condition) (Experiment 1, age x political interaction: Wald X^2(1) = 4.16, p = .041; Experiment 2, age x political interaction: Wald X^2(1) = 4.89, p = .027). Together, these findings suggest the impact of competition vs. cooperation on leadership preferences has deep developmental roots and that preferences for power vs. prestige may have different developmental trajectories.

Joshua Confer¹, Hanna Schleihauf¹, Jan Engelmann¹¹University of California, Berkeley

Beliefs are foundational to our lives. Currently, beliefs about the danger of a virus and the existence of climate change lead to critical personal and social consequences and spark hostile conflicts in public discourse. We often get angry at others when they do not possess the beliefs we think they should hold. But how responsible are we for our beliefs? Can people freely decide to choose one belief over another? In philosophy, whether beliefs are voluntary is controversial (James, 1937; Ganapini, 2020). Yet, recent psychological work indicates that adults endorse the ability to control one’s mental states, especially beliefs (Cusimano & Goodwin, 2019). However, much is still unknown about how we judge the controllability of beliefs, particularly if different kinds of beliefs influence judgments of control. For example, can people decide to think it’s sunny outside if they have been inside all day without windows or access to the weather report? Can people decide to think it’s sunny if they look outside and see that it’s storming? Additionally, even less is known about how children judge these cases. While research has investigated children’s judgments of the controllability of actions (Kushnir et al., 2015), to our knowledge nothing is understood about their judgments of the controllability of beliefs. In two pre-registered studies, we tested how 5-6-year-olds, 7-8-year-olds, and adults judge the controllability of others’ beliefs. Specifically, how perceptions of control may depend on the available evidence for a belief (Study 1) and the morality of a belief (Study 2). In Study 1 (N = 120), children and adults judged that people have control beliefs that are not supported by evidence or contradicted by evidence. In other words, a person could freely choose to hold other beliefs if they wanted to. These judgments of control were identical to that of choosing possible actions. However, children and adults judged that people have much less control over beliefs that are supported by strong evidence. These evidence-backed beliefs were judged as less controllable than possible actions, but more controllable than impossible actions. In Study 2 (N = 120), children and adults judged that people have control over opinions and immoral beliefs, identical to possible actions. Yet, children and adults judged that people have less control over moral beliefs. Similar to evidence-backed beliefs, moral beliefs were judged as less controllable than possible actions, but more controllable than impossible actions. Additionally, developmental differences emerged in judgments of the controllability of moral beliefs, such that adults viewed these beliefs as more free than younger children. Thus, our results suggest that judgments of how much of a choice people have over their beliefs greatly depends on the belief in question. Moreover, it seems already at a young age, children have advanced intuitions of the controllability of beliefs and how these beliefs are constrained.

Gesture is versatile in form and function.  Under certain circumstances, gesture can substitute for speech, and when it does, it embodies the properties of language that children themselves bring to language learning, and underscores the resilience of language itself. Under other circumstances, gesture can form a fully integrated system with speech.  When it does, it both predicts and promotes learning, and underscores the resilience of gesture in thinking. Together, these lines of research show how much of our minds is hidden in our hands.

Rodolfo Cortes Barragan¹, Andrew Meltzoff¹¹University of Washington

Cross-cultural work has shown that the expression of prosociality is influenced by cultural experience (Callaghan et al., 2011). Yet, there has been little empirical work on how cultural variations within in the U.S. may related to the expression of infants’ prosociality. Cultural-ethnic groups are known to vary in the socialization values, goals, and behaviors that they display towards their children, even within mainland U.S. (Tamis-LeMonda et al., 2008). Specifically, cultural psychology studies suggest that Asian-American and Hispanic-American parents have socialization practices that generally differ from the practices of White-American parents (Markus & Moya, 2010). Both Asian and Hispanic/Latinx groups in the U.S. emphasize an importance of noticing and actively helping others to achieve their goals, rather than a more self-oriented pursuit of their own individual goals at the expense of strangers (Rogoff et al., 2017). Yet, little is known about whether cultural factors are significantly influencing prosocial behavior in very young U.S. Asian and Hispanic children. Prior research constructed a situation that afforded 19-month-old infants with the opportunity to share valuable, personally valued items (e.g., their own teddy-bear) with a stranger (Barragan et al., 2020; Barragan & Meltzoff, 2021). Here, analyses focused on specifically comparing Asian-American infants and Hispanic-American infants to White-American infants using this large sample (N = 192). We find that, despite being exposed to the same broad environment of the metropolitan Seattle area, infants of Asian-American and Hispanic-American were significantly more likely than White-American infants to share valued items with strangers. This suggests the interesting cultural hypothesis that already at 19-months, cultural-ethnic experiences may be shaping U.S. infants’ behavior towards strangers. These findings reveal certain “strengths” in the infants of Hispanic/Latinx-American families, and thus we wish to counter the “deficit view” often offered for this stigmatized group. Discussion focuses on developmental mechanisms within Asian-American and Hispanic-American families that may explain the (obtained) elevated group-level generosity and sharing shown by these infants. Moreover, we posit how these mechanisms within Asian-American and Hispanic-American families may specifically inform both theories of social development and also theory-based interventions designed to promote prosociality within the child population as a whole.

Matthew Johnston¹, Teresa McCormack¹, Agnieszka Jaroslawska¹, Sara Lorimer¹, Sarah Beck², Christoph Hoerl³, Aidan Feeney¹¹Queen’s University Belfast, ²University of Birmingham, ³University of Warwick

Counterfactual thoughts and their associated emotions have received considerable attention from developmental psychologists. In particular, studies have explored children’s ability to experience and understand the emotions of regret and relief. The literature suggests these emotions emerge relatively late in childhood given the sophisticated counterfactual abilities required to experience them. However, recent theories suggest relief also occurs following the end of an unpleasant episode (e.g., Hoerl, 2015). In these so called ‘temporal relief’ instances, the emotion may not arise from complex counterfactual reasoning, but simply from an ability to appreciate that a negative event is now over and in the past. Until recently, developmental psychologists had only examined counterfactually mediated relief. Moreover, although previous work has explored children’s ability to use past events to infer current emotional states (see Lagattuta, 2014), less is known about children’s ability to understand how others feel immediately after events (positive or negative) are over. The current study explored children’s ability to integrate temporal information into their emotion judgements and contributes to a literature which heretofore has only focused on counterfactual relief. Nine vignettes were presented to 100 4-6-year-olds (40% female). In each story, one protagonist went through an episode that subsequently ended. Character’s attitudes were manipulated such that there were temporal relief stories, in which the episode was very unpleasant for the character, and temporal disappointment stories, in which the episode was very pleasant for the character. To gauge children’s understanding of how the cessation of events can impact emotions, we asked participants to judge whether the characters felt better, worse, or the same after the event was over. We reasoned that characters in the temporal relief and temporal disappointment stories should feel better and worse than before respectively. We also compared judgements in these trials to control stories in which the character felt neutral towards episodes and as such should feel indifferent when the event ends. We found that even 4-year-olds were above chance levels at judging the characters in the temporal relief stories as feeling happier than before. The main analyses did, however, indicate that the 6-year-olds made significantly more target judgements than the 4-year-olds, OR = 4.40, SE = 2.94, p <.001, suggesting that the tendency for children to make these relief attributions increases with age. Unexpectedly, participants’ judgements were more accurate in the temporal relief stories than in the disappointment and control stories (both p values <.001). This may be a result of individual differences in expectations about how people feel at the end of positive and neutral events. Overall, this study suggests that children are capable of attributing relief to others earlier than previously estimated in studies of counterfactual relief. This has implications for theories suggesting relief is best thought of as an emotion with two subtypes that have distinctive antecedents and cognitive prerequisites. Moreover, we provide evidence that 4-year-olds can integrate knowledge about the cessation of events into emotion judgements. This adds to a body of research which suggests young children are capable of taking past events into account when making judgments about how people feel in the present.

Meltem Yucel¹, Marissa Drell², Vikram Jaswal², Amrisha Vaish²¹Duke University, ²University of Virginia

Children are sensitive to (un)fairness. But little is known about the extent to which children perceive fairness as a moral vs. a conventional norm. Through five studies with 193 children and 540 adults in the U.S., this work establishes how children understand fairness norms, how children’s fairness understanding changes with age, and the role of harm in the moralization of fairness norms. Across a series of studies, 4-year-olds rated moral transgressions (e.g., hitting) as more serious than fairness (e.g., taking more chalk than another child) and conventional transgressions (e.g., wearing pajamas to school), but importantly, they rated fairness and conventional transgressions as similarly serious. In contrast, 6- and 8-year-olds and adults rated moral transgressions as more serious than fairness and conventional transgressions, and fairness as more serious than conventional transgressions. An additional, forced-choice procedure revealed that most 6-year-olds also categorized fairness with moral rather than conventional transgressions; 4- and 8-year-olds’ responses did not show systematic patterns. Conversely, adults categorized fairness with conventional rather than moral transgressions. In a series of follow-up studies, the role of harm salience in perceptions of fairness violations was investigated with 4-year-olds and adults. We randomly assigned 4-year-olds and adults to Baseline or Harm Salience conditions. The Baseline condition was identical to the previous study. In the Harm Salience condition, the four Fairness scenarios emphasized the harmful outcome (i.e., “This child took more chalk than this other child, and this other child was sad because he/she got less chalk.” ). Harm was manipulated only for the Fairness transgression type. Young children’s perceptions of harm influenced their norm judgments, such that fairness scenarios with high harm salience (Harm Salience condition) were evaluated significantly worse than scenarios with no additional harm emphasis (Baseline condition). In contrast to the 4-year-olds, adults in the Harm Salience condition evaluated unfairness marginally more seriously than those in the Baseline condition. Across both conditions, adults rated moral transgressions as more serious than fairness and conventional transgressions, and fairness as more serious than conventional transgressions. Furthermore, the forced-choice procedure revealed that 4-year-olds’ responses did not show systematic patterns even with the harm emphasis. Once again, adults categorized fairness with conventional rather than moral transgressions in the baseline condition. Interestingly, adults were at chance for grouping fairness transgressions with moral or conventional transgressions. These studies revealed that the amount of emphasis placed on harmful outcomes plays an important role in adults’ (and possibly also children’s) perceptions of unfairness. This is the first evidence that children and even adults may not equate norms of fairness in resource distribution with harm-based moral norms. This points to the need for a more nuanced understanding of children’s developing perceptions of social norms.

Fan Yang¹, Margaret Wolfson¹¹The University of Chicago

Human beings depend on groups to survive and thrive throughout history, but to what extent do we care about the groups we are a part of? Extensive research has shown that even young children display a remarkable capacity to act prosocially interpersonally, but little is known about our early tendencies to benefit groups. Our research aims to understand how children (aged 4-9) tradeoff their own interests versus the interests of their group, to shed light on early altruistic potential toward groups. Across four studies (three preregistered, total N = 274 children), we examined 4-9-year-old children’s valuation and promotion of group interests, in terms of both harm and benefit. Across studies, we developed and adopted a novel paradigm in which children could decide to gain benefit (e.g., a medal made by a 3D printer) for a certain target at the expense of harm (e.g., listening to an annoying noise from the printer) for another target. Study 1 revealed that in the absence of a cost to the self (e.g., forced choice between only benefiting/harming the self versus benefiting/harm both the self and the group), the majority of children benefited (M = .90) and avoided harming (M = .72) their group. In Study 2, children were asked how long they would inflict harm (i.e., an annoying noise, 1= short, 4 = super long) on themselves or their group in order to gain a benefit (i.e., a medal) for themselves, and they decided to impose more harm on themselves (M = 3.07) compared to on their group (M = 2.70), p = .04. They also judged that they should impost more harm on themselves (M = 2.75) than on the group (M = 2.36), p = .014. Study 3 examined the extent to which children would endure cost in order to benefit themselves or the group. We found that children decided to benefit the group (M = 3.32) more than the self (M = 2.91), p = .03, and they judged that they should benefit the group (M = 3.10) more than the self (M = 2.50), p = .0003. In Study 4, children directly traded off decisions for harming and benefitting the group versus themselves. We found that overall children’s tendency to avoid harming their group (for their own benefit) was stronger than benefiting their group (at their own cost). Younger children (aged 4-6) chose to benefit themselves over the group (M = .12, p = < .0001), and were at chance for harm decisions. Older children (aged 7-9) were at chance for benefit decisions, but they were more likely to imposed harm on themselves than on the group (M = .69, p = .0003). Our results reveal that children value their group’s interest more than their own interests, especially when allocating harm. With age, children increasingly put more weight on group interests, when directly trading off the self and the group. The findings contribute to a better understanding of the early altruistic potential toward groups.

Madeline Reinecke¹, Larisa Heiphetz²¹Yale University, ²Columbia University

Can moral rules change? Several lines of work in moral psychology suggest that children (and adults) may view moral truths as unchanging. However, a separate body of research concerning the cognitive science of religion finds that many people view God as all-powerful, which implies that God could change moral truths. This brings with it an unsightly conclusion, however: God could have made morality, including seemingly “absolute” aspects of morality, entirely differently (Murphy, 2019). Instead of making murdering a child morally wrong, God could have made murdering a child for fun morally right. We recruited a sample of 132 children from the United States to investigate their beliefs about whether God could change widely shared moral propositions (e.g., “it’s not okay to call someone a mean name”), controversial moral propositions (e.g., “it’s not okay to tell a small lie to help someone feel happy”), and physical propositions (e.g., “fire is hotter than snow”). We provided children with an opportunity to express their belief (and their confidence in that belief) concerning a given proposition. For example: “This person thinks that it’s okay to stomp on someone’s foot really hard. This person thinks that it’s not okay to stomp on someone’s foot really hard. Which person do you agree with more?” We then asked whether God could make the opposite of their choice become true (e.g., “Do you think that God could make it not okay to stomp on someone’s foot really hard?”), followed by a second certainty judgment (e.g., “How sure are you that God could/couldn’t do that?”). We observed an emerging tendency to report that God’s ability to change morality is limited. Children between ages four and six years old did not distinguish among God’s ability to change widely shared moral, controversial moral, and physical propositions (ps ≥ .032, Cohen’s ds ≥ .21). In contrast, children between seven and nine years old became increasingly confident that God could change physical propositions (t(261.63) = 3.54, p < .001, Cohen’s d = -.44) and could not change widely shared moral propositions (t(251.09) = 2.72, p = .007, Cohen’s d = .34). Critically, older children–like younger children–denied that God could change widely shared aspects of morality (younger: t(63) = -6.40, p < .001, Cohen’s d = .80; older: t(67) = -10.46, p < .001, Cohen’s d = 1.27). These findings converge with a well-documented tendency for morality to shape non-moral cognition (Knobe, 2010), including children and adults’ judgments of possibility (Phillips & Cushman, 2017; Shtulman & Phillips, 2018), as well as research on the development of God concepts (Heiphetz, Lane, Waytz, & Young, 2018).

Layla Unger¹, Hyungwook Yim², Olivera Savic¹¹Ohio State University, ²Hanyang University

Recent years have seen a flourishing of models that can mimic vital aspects of human language, such as completing sentences and forming analogies (Landauer & Dumais, 1997; Mikolov et al., 2013; Penninton et al., 2014). These models capitalize on a simple, decades-old idea: It is possible to learn a lot about what words mean based on regularities in the way they are used with each other in language. This idea is commonly referred to as the distributional hypothesis, which states that regularities of word use are informative about word meanings (or semantics) because words with related meanings such as “chicken” and “duck” tend to reliably co-occur with each other, as in “Chickens and ducks lay eggs”, and in similar contexts of other words, as in “Chickens lay eggs” and “Ducks lay eggs”. Word knowledge that captures these similarities in meaning is vital for language processing in both models and humans (Borovsky et al., 2016; Kutas & Federmeier, 2000; Nation & Snowling, 1999). Does this rich source of information also shape the development of knowledge about word meanings in humans? Although this possibility has been noted for decades (Borovsky & Elman, 2006; Ervin, 1961; Sloutsky et al. 2017), it faces two important challenges. First, the language input to models that capitalize on these regularities differs greatly from language input to young children in both content and sheer amount. Second, models use learning mechanisms that may be absent or immature for much of human development. Specifically, young learners are indeed sensitive to simple regularities with which words co-occur with each other (e.g., Fisher, 2010; Wojcik & Saffran, 2015). However, models can capture similarity in word meaning even when words do not co-occur with each other, and instead only occur in similar contexts of other words. This capacity requires learning that spans entirely different experiences, such as hearing a sentence about chickens laying eggs on one day, and a sentence about ducks laying eggs on another. Critically, abilities to learn these regularities may develop only gradually throughout childhood and into adolescence (Bauer et al., 2012, 2020; Schlichting et al., 2017; Shing et al., 2019). Equipped with only the limited input and learning mechanisms available during human development, can young language learners develop semantic knowledge about words just from regularities of word use? To answer this question, we investigated whether semantic knowledge can develop just from the simple regularity with which words co-occur with each other in language input to infants and children. In Study 1, we provide a proof-of-principle that that language input to young learners is rich in co-occurrence regularities that can support learning relations between words similar in meaning, such as words for animals, colors, or body parts. Study 2 builds upon this evidence by showing that these regularities can explain early semantic knowledge about words that has been observed across several prior studies. Finally, Study 3 goes further to show that these regularities can account for the strength of children’s spontaneous associations between words similar in meaning. Findings from all three studies provide robust evidence that even simple co-occurrence regularities in infants’ and children’s language input can play a powerful role in shaping their development of knowledge about words.

Ruthe Foushee¹, Mahesh Srinivasan²¹University of Chicago, ²University of California, Berkeley

An extensive literature suggests that children learn best from the speech that caregivers direct to them (child-directed speech, or CDS), compared to other speech that they overhear. However, the practice of speaking to young children is rare in many cultural contexts. Moreover, some researchers have argued that language development unfolds similarly, both in contexts where children receive little CDS and contexts where CDS is common (e.g., Casillas, et al., 2020), raising the possibility that infants adapt their learning strategies to the specific sources of language available in their environments. Here, we seek early evidence of two forms of language knowledge in Tseltal Maya infants–who receive very little CDS, in part because they are almost always carried on their mothers’ backs. An adapted looking-while-listening task used infants’ relative looking between pairs of pictures to test their knowledge of a set of (1) Common Nouns and (2) Greetings, terms used as greetings between community members, depending on the age and sex of the addressee (e.g., “tatik” is used to greet an older man, while “metik” is for an older woman). Though very frequent in infants’ early environments, such greetings would never have been used to greet infants themselves. In the study, infants sat on mothers’ laps before side-by-side screens in a sound-dampening tent. On each trial, mothers⁠–whose eyes were closed throughout the study–heard a target noun or greeting over headphones, then repeated it for the infant. Common-Noun trials presented item-pairs of one animate and one food-related image (e.g., baby-corn, horse-soda, 32 trials total), while Greeting trials presented pairs of faces corresponding to distinct terms of address (e.g., older-man–younger-woman, 8 trials). A camera recorded infants’ gaze, manually coded later for left/right direction. 24 infants (5.3-15 months, M=11) and caregivers participated (5 in only Common-Nouns, 3 in only Greetings, and 14 in both tasks, order counterbalanced). Previous work suggests Tseltal infants in this age range rarely (<15%) receive CDS and spend the majority of each day overhearing from their mothers’ backs. This was confirmed in interviews with caregivers and ongoing analysis of daylong audio-recordings collected at the same time. Following Bergelson & Swingley (2012), we defined an analysis window from 367-3500ms post-target-word onset, and calculated the mean difference in infants’ proportion looking to each image when it was the target, relative to when it was not the target. Positive scores for item-pairs and for individual subjects suggest a tendency to prefer the target image. Notably, based on evidence from Western samples, positive scores on the Greeting trials are unlikely, as they contrast two highly similar and semantically related visual stimuli (i.e., two faces)⁠–not typically distinguished until 24 months (Kartushina & Mayor, 2019). Our results provide evidence of early context-specific language knowledge in infants who rarely hear CDS. Means for all Common-Noun and Greeting item-pairs were positive (Common-Noun range: 0.01–0.24; M=0.09, 95% CI:[0.055, 0.14]; Greeting range: 0.003-0.22; M=0.11[0.0035, 0.22]). At the individual level, 14 of 19 infants’ mean Common-Noun scores (-0.57-0.40; M=0.09 [-0.01, 0.18]), and, remarkably, 10/14 Greeting scores were positive (-0.76-1; M=0.13 [-0.09, 0.37])–consistent with infants adapting their learning strategies to their environments.

Melina Knabe¹, Christina Schonberg², Haley Vlach¹¹University of Wisconsin-Madison, ²IXL

Constraints, sociopragmatic, and domain-general theories have proposed the cognitive and social processes critical to children’s word learning, leading to a corpus of replicable research findings. However, we do not know whether adults’ understanding of early word learning aligns with key research findings. This is a critical gap in research because adults serve as the architects of children’s learning environments (e.g., Hirsh-Pasek et al., 2015; Rowe et al., 2017; Weisleder & Fernald, 2013). Alignment with research findings is therefore pivotal in supporting children’s development. The goal of Experiment 1 was to examine adults’ intuitions about the science of children’s word learning. A sample of undergraduates (n = 89), adults from the general public/non-parents (n = 93), general public/parents (n = 88), and Speech-Language Pathologists (n = 77) completed a survey with 11 word learning principles. For each question, participants were prompted to select an answer based on the perspective of a preschooler. Questions tested key concepts from early word learning research, such as the mutual exclusivity assumption, whole object assumption, shape bias, and taxonomic bias. Additional concepts included the role of sociopragmatic cues (i.e., pointing, eye-gaze) and the ability to learn from overheard speech. Finally, participants were asked whether children could learn from massed vs. spaced presentation schedules, in varied vs. repeated contexts, and cross-situationally. We predicted that adults’ intuitions would be less aligned for principles derived from domain-general theories, regardless of their experience with language acquisition or language instruction. This prediction is based on prior studies revealing adults’ inaccurate perceptions of domain-general principles in human memory, such as spaced learning (Kornell & Bjork, 2008). Results supported our prediction: Adults intuitions were aligned for all principles except those derived from domain-general theories (Figure 1), regardless of experience with language development. A follow-up experiment (Experiment 2; N = 91) was designed to test whether the perceived difficulty of an item (5-point Likert scale; 5 = “Extremely difficult”) impacted adults’ reasoning about word learning processes. Indeed, domain-general theory principles (M = 2.76, SD = 0.48) were rated as significantly more difficult than sociopragmatic theory (M = 2.49, SD = 0.65), p < .001, and constraints theory principles (M = 2.12, SD = 0.55), p < .001, especially when adults thought a preschooler could not complete the task. These findings suggest that adults use task difficulty as a guide to what will aid children’s language learning. Finally, Experiment 3 (N = 97) ruled out level of confidence and interest in the test items as mechanisms to explain the results. Taken together, this study highlight disconnects in knowledge between the research community and general public. Efforts must therefore be made to assess what the public knows and does not know about children’s word learning. After all, comparing adults’ intuitions to research findings that scientists consider robust is a critical first step in developing an evidence-based plan for disseminating research. The talk will conclude with a discussion of how to address adults’ biases against difficult learning conditions, thereby supporting the key stakeholders of children’s language development.

Keith Apfelbaum¹, Jamie Klein-Packard¹, Bob McMurray¹¹University of Iowa

It is well known that the processes that support efficient word recognition undergo rapid development in early childhood, and substantial research on this has focused on the first few years of life. However, later childhood offers an intriguing context for continued development of word-level skills: the school-age years include richer language input and the onset of reading education. As a result, children’s spoken- and written-word recognition abilities may continue to be refined throughout later childhood. Moreover, the demand for flexible and efficient word recognition is heightened in these ages, with the increase in linguistic complexity that occurs with the onset of schooling and the need for precise phonological representations to support reading. The first part of this talk presents an overview of recent work that shows surprisingly slow development of language skills that were often thought to develop in the earliest years. For example, the efficiency with which children recognize a spoken word and suppress competitors improves between 9 and 16 years. This development is mirrored in written-word recognition. Even more foundational speech perception skills show considerable refinement through age 18, suggesting that these years are a period of considerable development of seemingly low-level skills. Building on this work, our research team recently launched the Growing Words Project, a longitudinal study that traces the development of spoken- and written-word recognition and speech perception during the school-age years. The project aims to characterize the nature of these ongoing changes, and to identify factors that relate to this growth. This project will help further refine theories of school-age language development and investigate individual differences in development across a broad range of children. The second part of this talk reports results from the first year of data collection for Growing Words. Growing Words measures the dynamics of lexical processing for both spoken and written words using eye-tracking in the Visual World Paradigm (VWP). Speech perception is assessed with a continuous-rating-scale task to measure sensitivity to within-category phonemic information. These abilities are related to other cognitive processes, such as cognitive control; to standardized measures of oral language, phonological processing and reading; and to measures of the home language and literacy environment. The battery of measures was collected with 241 children between first and third grade, each of whom will complete follow-up sessions for the next three years. The Growing Words sample includes a broadly representative sample of children with a broad array of language abilities and backgrounds across the developmental spectrum. Initial results support the idea that the dynamics of word recognition (both spoken and written) reflect a complex intersection of language, cognitive skills and the language environment. In line with previous research, speech perception and word recognition show improvements across grades, even within the narrow timeframe of 1st to 3rd grade. However, these developmental patterns go beyond simple age effects. I discuss how these measures interrelate with cognitive control, standardized assessments and language input in cross-sectional comparisons. These data are used to refine theories of school-age language development, and to detail predictions from longitudinal data collection in Growing Words.

Molly Scott¹, Jessica Lawson-Adams², Emily Hopkins³, Haley Weaver⁴, Jacob Schatz⁵, Rebecca Dore⁶, Marcia Shirilla⁷, Molly Collins², Tamara Spiewak-Toub¹, David Dickinson², Roberta Golinkoff⁷, Kathy Hirsh-Pasek¹¹Temple University, ²Vanderbilt University, ³University of Scranton, ⁴University of Wisconsin-Madison, ⁵New York University, ⁶Ohio State University, ⁷University of Delaware

Many vocabulary interventions aimed at children from disadvantaged backgrounds have exhibited limited success (Wasik et al., 2016; Dickinson et al., 2011), perhaps because they typically teach words through shared book reading, a method shown to have moderate impacts on vocabulary acquisition (NELP, 2008). Teaching words solely through book reading may not be powerful enough to improve the vocabulary ability of at-risk children (Neuman et al., 2011). Moreover, a Matthew effect has been noted by literacy researchers (Stanovich, 1986; Duff et al., 2015); children who demonstrate heightened word knowledge learn a greater number of vocabulary words compared to children who know fewer words. What if we could increase the effectiveness of interventions for children with limited vocabulary? Perhaps we could boost word learning by taking advantage of research showing that children learn best from playful and interactive contexts (Fisher et al., 2013; Weisberg et al., 2016) in which they are active, engaged, and participating in meaningful interactions with adults and other children (Harris et al., 2011). The Language for Reading project did just this, by comparing teaching words through book-reading and play. Teachers in state- and federally-funded, low-income, preschool programs taught 60 difficult words (e.g. caution, gunwale) to students (N = 238; Mage = 54.03 months) over seven months. Vocabulary words were embedded into storybooks that teachers read aloud to students and into playful learning activities that included small-group games, large-group games, music, socio-dramatic play, and a touchscreen game. Some words were taught only in book-reading, others only during play. Students’ growth in knowledge of the target vocabulary items was assessed before and after the intervention. Children’s self-regulation ability and general vocabulary knowledge (PPVT) were measured prior to start of the intervention (Dunn & Dunn, 2007). A logistic regression predicting correct responses on an experimenter-created receptive measure of vocabulary demonstrated that, after controlling for the effects of self-regulation, children’s growth from pre- to post-test on words taught in book-reading was moderated by their PPVT score indicating that children with greater vocabulary learned more target words, B = 0.005, SE = 0.002, p = 0.054. However, after controlling for self-regulation ability, this moderator effect of PPVT score on children’s target word growth was not found for words learned from any of the play activities, B = 0.003, SE = 0.002, p = 0.214. That is, children’s vocabulary skill did not impact their learning of the target words that were taught through the playful learning activities. Playful learning permitted children to experience and retain the meanings of the new words. These results suggest that playful and interactive pedagogies used to teach words may be particularly effective for children with low vocabulary, who are most at risk for future academic problems. Future research should continue to examine play as a method for word learning and which types of play might be the most effective.

Hilary Miller-Goldwater¹, Melanie Hanft¹, Lucy Cronin-Golomb¹, Patricia Bauer¹¹Emory University

Book reading is a potential means of promoting science learning early in development (e.g., Ganea et al., 2011). Past research informed the type of information children learn from book reading. Yet there is limited understanding of the factors influencing science learning from books prior to formal schooling and the extent to which these early factors predict science achievement into elementary school. We bridge these gaps in two experiments. In Experiment 1, we assessed how caregiver’s reading styles and characteristics of science books themselves related to preschool-age children’s science learning. Caregiver’s elaborative talk during reading (providing details, definitions, and examples beyond the text of books) predicts children’s language skills (e.g., Price et al. 2009). Whether it also predicts science learning is unknown. Additionally, young children’s science books vary greatly in the extent to which their text includes elements expected to facilitate learning, namely, Cohesion (elaborates on concepts to support limited prior knowledge and working memory), and Demand (includes questions and interactive prompts to support active engagement and underdeveloped memory strategies) (Miller et al., 2020). How caregivers’ reading styles and books’ structural features interact to influence science learning was also unknown. We tested 38 caregiver and their 4-5-year-old children (Mage =4.81 years). Caregivers read to their children four books that were crossed on levels (high, low) of Cohesion and Demand. We measured caregiver’s elaborative talk, children’s open-ended recall of science facts from the books, and children’s verbal comprehension. When controlling for verbal comprehension, Cohesion most strongly predicted children’s science fact recall (F1,243.62 = 30.10, p < .001, d =.70). Additionally, Demand interacted with caregiver’s elaborative talk (F1,267.54 = 7.72, p = .006, d =.34), such that when Demand and caregivers elaborative talk were high, children recalled more science facts from the book. These findings show that properties of the books strongly influence children’s early science learning and interact with caregiver’s talk to influence such learning. To understand factors influencing later science knowledge, in Experiment 2 we tested the same children longitudinally 2.5-3-years later. We assessed how individual differences in children’s recall of science facts from books at 4-5-years and caregiver’s reading styles predicted children’s science achievement at 7-years of age. Our preliminary sample includes 24 children (Mage=7.49). We assessed children on standardized tests of science achievement and verbal comprehension. In terms of concurrent predictors at 7 years of age, children’s science achievement was highly correlated with their verbal comprehension (r22 = .65, p < .001). Longitudinally, children’s recall of science book facts at 4-5-years predicted both science achievement (t21=7.50, p=.003) and verbal comprehension (t21=22.11, p<.001) at age 7, whereas caregiver’s elaborative talk was not significant (ps >.625). These preliminary results suggest that children’s learning of science facts from books relates to their later science achievement and language skills. In sum, these findings highlight that reading high quality science books prior to formal schooling can be an avenue for supporting science learning into elementary school.

Christina Schonberg¹, Xiaozhu An¹, Bozhidar Bashkov¹¹IXL Learning

This study applies a new theoretical framework (Yeh, 2018) to investigate the potential of using individualized instruction to narrow the academic opportunity gap between White and Hispanic elementary and middle school students. Under Yeh’s framework, differential academic outcomes are hypothesized to stem from the way that the current educational system treats “at-risk” students differently from other students (e.g., through class placement), which can lead to demotivation and disengagement. Students may instead benefit more from individualized work and assessments offering frequent formative feedback, which can lead to developing a sense of mastery, increased motivation, and feelings of ownership of their education (see Yeh, 2006). Method. We investigated the effectiveness of technology-based individualized instruction–specifically, IXL, a platform used by 1 in 5 students in the U.S.–in supporting Hispanic students’ academic achievement. IXL’s design is rooted in learning sciences research (Bashkov et al., 2021) and provides students with exactly the type of supportive, yet appropriately-challenging learning environment that Yeh (2018) recommends. As students use IXL, the software adapts to provide skills at the right level of difficulty. Within each skill, students earn a dynamic proficiency score that can range from 0-100. Students receive frequent, nonevaluative feedback, including detailed explanations of questions they answer incorrectly. Because literacy and language skills are fundamental for success in all academic areas, in this study we examined IXL English language arts (ELA) usage and its impact on ELA achievement. As part of a research partnership with a large, majority-Hispanic district (72% of students) in Arizona, we examined assessment, demographic, and IXL usage data from Hispanic students in grades 3-8 who had used IXL ELA during the 2018-19 school year (N=569; 55% female; 8.3% English language learners; 10% IEP students). This study used a pretest-posttest design, with students’ scores on the state’s end-of-year assessments in 2018 and 2019 serving as the pretest and posttest measures, respectively. Students’ IXL usage was obtained from IXL’s database. IXL’s usage recommendations highlight the importance of reaching skill proficiency (i.e., a score of 80 or above); thus, we examined students’ mean number of skills proficient per week (SP/week).  Results. Controlling for students’ pretest performance, grade, gender, IEP status, and ELL status, we found that mean SP/week was a significant predictor of posttest assessment performance over and above the other predictors, b=11.92, p=.02. That is, for each additional IXL ELA skill proficient per week, a student could expect their assessment score to increase by about 12 points (.43 SD). Discussion. Reaching proficiency in one additional IXL ELA skill per week was associated with a significant (.43 SD) increase in state assessment scores. In the context of this assessment, these results have enormous practical significance: this increase would be sufficient for many students to reach the state’s next-highest proficiency category (e.g., moving from “not proficient” to “proficient”). These results provide empirical support for Yeh’s (2018) theory suggesting individualized task difficulty may help improve marginalized students’ achievement and show that IXL could significantly reduce the opportunity gap in the United States.

Cassondra Eng¹, Erik Thiessen¹, Anna Fisher¹¹Carnegie Mellon University

Environmental enrichment is critical in shaping the development of executive function (EF)-cognitive processes that subserve goal-directed behavior-through increased quantity and quality of multimodal input to the prefrontal cortex (PFC; Werchan & Amso, 2017). Fostering EF is important because EF longitudinally predicts academic achievement and learning-related classroom behaviors controlling for IQ and socioeconomic status (Diamond, 2011). One way to create an enriched environment is through contexts that stimulate cognitive and motor functions simultaneously. While there are well-documented beneficial effects of combining computerized cognitive training with exercise on EF in elderly populations, research on children is scarce. Studies with multiple EF assessments and complimentary neuroimaging with children are needed to clarify the unique efficacy combined training has on EF and elucidate the mechanisms underlying changes. This preregistered study filled these gaps by 1) assessing the efficacy of combined training on EF by comparing combined training to cognitive training, exercise training, and an age/sex/classroom-matched control group 2) including multiple EF tasks and teacher reports of EF to assess whether the skills children learn transfer to real-world classroom contexts and 3) investigating the neural mechanisms of training-induced changes with fNIRS. 120 children (ages 4 to 5) were recruited and randomly assigned to 1 of 4 conditions: Combined, Exercise, Cognitive, or Control. Each condition was projected onto a wall with a game step mat with one left arrow and one right arrow. The Combined condition (high exercise & high cognitive activity) was modeled on the Flanker Task, but instead of pressing left and right arrows on a computer, children responded by stepping left or right on the physical game mat’s arrows depending on the direction that the central target was facing. The Cognitive condition was identical to the Combined condition, except participants sat on the mat and pressed left or right with their hands (low exercise & high cognitive activity). The Exercise condition was identical to the Combined Condition, except the central target was not surrounded by distractors (high exercise & low cognitive activity). Neural connectivity, performance on EF tasks (Flanker Task, Day-Night Task) and teacher EF ratings (BRIEF-P) were assessed at pretest and posttest. Children in each condition trained for 2 consecutive days for 20 mins/day. Children in the Control group continued their typical activities and just participated in the pre- and posttest assessments. Combined training improved children’s performance on the behavioral EF tasks, EF-related behaviors in the classroom, and PFC functional connectivity compared to the other conditions. The Combined condition exhibited significantly stronger changes in connectivity strength within the PFC from pre- to posttest compared to all other conditions (Figure 1A). Only children in the Combined condition exhibited improvement on the transfer Day-Night EF Task and teacher ratings of EF (Figure 1B). This area is particularly timely and important because parents and educators across the world are attempting to find ways to keep children physically active and engaged in the wake of stay-at-home orders. Combined contexts can easily be implemented in children’s lifestyles and have high potential to improve essential EF skills that are crucial for academic success.

Joseph Colantonio¹, Igor Bascandziev², Maria Theobald³, Garvin Brod³, Elizabeth Bonawitz²¹Rutgers University – Newark, ²Harvard University, ³The DIPF | Leibniz Institute for Educational Research and Educational Information

Young children rely on “intuitive theories” when reasoning about the world (Gopnik & Wellman, 2012; Carey, 2004; Keil, 2003). The mechanisms underlying belief revision remain important to study. In particular, to comprehensively understand the process of belief revision, three challenges need to be met. First, for any domain, a representation of the learner’s current beliefs must be described. Second, we require a model for how these beliefs are updated moment-to-moment as new evidence is encountered, alongside an understanding of how different response modalities affect learning (such as predicting vs evaluating). Third, while Bayesian models have provided a framework with which to capture learning, additional factors remain less understood – how might emotions (e.g., surprise) relate to these models? This is critical to address, as affective states, like surprise, may be vital for finding errors and revising theories (Brod, Hasselhorn & Bunge, 2018). Here, we present work that computationally investigates these three questions with data on elementary school children’s understanding and revision of intuitive beliefs of water displacement – as affected by whether they predict or post hoc evaluate (Brod & Theobald, 2021). We first describe computational representations that capture initial representations of individual children’s beliefs of water displacement. This provided a domain to measure initial belief and learning where children often have misconceptions (Burbules & Linn, 1988). We found that children’s behavior was best described by different generative models that represent possible beliefs that they may hold. Second, we looked to capture learning as it unfolds continuously and to understand the role of prediction in engaging model-based reasoning by using theory-based Bayesian models. We compare three models that vary on whether they account for individual differences in children’s prior beliefs or the ability to consider multiple competing theories. Model comparison entailed generating predictions of the choices made by children at each trial by the three models. Across 3000 trials, we found that a Bayesian model that accounts for both factors outperforms competitors that ignore either factor (vs no-Prior Beliefs, z = 18.36, p ~ 0; vs Single Theory, z = 18.90, p ~ 0). Third, we describe quantified, “real-time” surprise responses. We generated likelihoods of a surprise response based on prior beliefs, observed evidence, and children’s behavior. Then, we compare this “model surprise” with a physiological measure of surprise: pupil dilation. We found that model surprise significantly correlates with children’s surprise at the individual level, trial-by-trial (r(2898) = 0.04, p < 0.05). Further comparing between conditions, we find significant correlations for children making predictions (r(1436) = 0.05, p < 0.052), but not for those post hoc evaluating (r(1460) = 0.037, p = 0.15). This fits with past research, highlighting that the process of making a prediction may prompt a stronger surprise-pupillary response compared to others who did not engage prior theories. This work shows the importance of computational models of learning. We describe some first steps toward understanding belief revision as potentially affected by emotions. Future work entails incorporating the measured surprise responses into our models’ algorithms, potentially uncovering the effects of affect on learning.