What do infants and young children tend to see in their everyday lives? Relatively little work has examined the categories and objects that tend to be in the infant view during everyday experience, despite the fact that this knowledge is central to theories of category learning. Here, we analyzed the prevalence of the categories (e.g., people, animals, food) in the infant view in a longitudinal dataset of egocentric infant visual experience. Overall, we found a surprising amount of consistency in the broad characteristics of children's visual environment across individuals and across developmental time, in contrast to prior work examining the changing nature of the social signals in the infant view. In addition, we analyzed the distribution and identity of the categories that children tended touch and interact with in this dataset, generalizing previous findings that these objects tended to be distributed in a Zipfian manner. Taken together, these findings take a first step towards characterizing infants' changing visual environment, and call for future work to examine the generalizability of these results and to link them to learning outcomes.

Childhood is marked by the rapid accumulation of knowledge and the prolific production of drawings. We conducted a systematic study of how children create and recognize line drawings of visual concepts. We recruited 2-10-year-olds to draw 48 categories via a kiosk at a childrens museum, resulting in >37K drawings. We analyze changes in the category-diagnostic information in these drawings using vision algorithms and annotations of object parts. We find developmental gains in childrens inclusion of category-diagnostic information that are not reducible to variation in visuomotor control or effort. Moreover, even unrecognizable drawings contain information about the animacy and size of the category children tried to draw. Using guessing games at the same kiosk, we find that children improve across childhood at recognizing each others line drawings. This work leverages vision algorithms to characterize developmental changes in childrens drawings and suggests that these changes reflect refinements in childrens internal representations.

How do children’s representations of object categories changeas they grow older? As they learn about the world aroundthem, they also express what they know in the drawings theymake. Here, we examine drawings as a window into how chil-dren represent familiar object categories, and how this changesacross childhood. We asked children (age 3-10 years) to drawfamiliar object categories on an iPad. First, we analyzed theirsemantic content, finding large and consistent gains in howwell children could produce drawings that are recognizable toadults. Second, we quantified their perceptual similarity toadult drawings using a pre-trained deep convolutional neuralnetwork, allowing us to visualize the representational layoutof object categories across age groups using a common featurebasis. We found that the organization of object categories inolder children’s drawings were more similar to that of adultsthan younger children’s drawings. This correspondence wasstrong in the final layers of the neural network, showing thatolder children’s drawings tend to capture the perceptual fea-tures critical for adult recognition. We hypothesize that thisimprovement reflects increasing convergence between chil-dren’s representations of object categories and that of adults;future work will examine how these age-related changes re-late to children’s developing perceptual and motor capacities.Broadly, these findings point to drawing as a rich source ofinsight into how children represent object concepts.

The ability to process social information is a critical compo-nent of children’s early language and cognitive development.However, as children reach their first birthday, they begin tolocomote themselves, dramatically affecting their visual ac-cess to this information. How do these postural and locomotorchanges affect children’s access to the social information rele-vant for word-learning? Here, we explore this question by us-ing head-mounted cameras to record 36 infants’ (8-16 monthsof age) egocentric visual perspective and use computer visionalgorithms to estimate the proportion of faces and hands in in-fants’ environments. We find that infants’ posture and orienta-tion to their caregiver modulates their access to social informa-tion, confirming previous work that suggests motoric develop-ments play a significant role in the emergence of children’s lin-guistic and social capacities. We suggest that the combined useof head-mounted cameras and the application of new computervision techniques is a promising avenue for understanding thestatistics of infants’ visual and linguistic experience.

The faces and hands of caregivers and other social partners offer a rich source of social and causal information thatmay be critical for infants cognitive and linguistic development. Previous work using manual annotation strategies andcross-sectional data has found systematic changes in the proportion of faces and hands in the egocentric perspective ofyoung infants. Here, we examine the prevalence of faces and hands in a longitudinal collection of nearly 1700 headcamvideos collected from three children along a span of 6 to 32 months of agethe SAYCam dataset (Sullivan, Mei, Perfors,Wojcik, & Frank, under review). To analyze these naturalistic infant egocentric videos, we first validated the use of amodern convolutional neural network of pose detection (OpenPose) for the detection of faces and hands. We then appliedthis model to the entire dataset, and found a higher proportion of hands in view than previous reported and a moderatedecrease the proportion of faces in childrens view across age. In addition, we found variability in the proportion offaces/hands viewed by different children in different locations (e.g., living room vs. kitchen), suggesting that individualactivity contexts may shape the social information that infants experience.

How do children’s visual concepts change across childhood,and how might these changes be reflected in their drawings?Here we investigate developmental changes in children’s abil-ity to emphasize the relevant visual distinctions between objectcategories in their drawings. We collected over 13K drawingsfrom children aged 2-10 years via a free-standing drawing sta-tion in a children’s museum. We hypothesized that older chil-dren would produce more recognizable drawings, and that thisgain in recognizability would not be entirely explained by con-current development in visuomotor control. To measure recog-nizability, we applied a pretrained deep convolutional neuralnetwork model to extract a high-level feature representation ofall drawings, and then trained a multi-way linear classifier onthese features. To measure visuomotor control, we developedan automated procedure to measure their ability to accuratelytrace complex shapes. We found consistent gains in the recog-nizability of drawings across ages that were not fully explainedby children’s ability to accurately trace complex shapes. Fur-thermore, these gains were accompanied by an increase in howdistinct different object categories were in feature space. Over-all, these results demonstrate that children’s drawings includemore distinctive visual features as they grow older.

Children produce increasingly more recognizable drawings of object concepts throughout childhood. What drives this improvement? Here we explore the role of children's ability to include relevant parts of those objects in their drawings. We crowdsourced part tags for every pen stroke in 2,160 drawings of 16 common object categories that had been produced by children between 4 and 8 years old. These part decompositions revealed both substantial variation in the number and kind of parts that children emphasized, as well as a non-monotonic relationship between the number of parts that children drew and how recognizable their drawing was. We plan to publicly release these data to catalyze further investigation of how children's drawings change across development.

What structures children's early language environment? Large corpora of child-centered naturalistic recordings provide an important window into this question, but most available data centers on young children within the home or in lab contexts interacting primarily with a single caregiver. Here, we characterize children's language experience in a very different kind of environment: the preschool classroom. Children ages 3 – 5 years (N = 26) wore a head-mounted camera in their preschool class, yielding a naturalistic, egocentric view of children's everyday experience across many classroom activity contexts (e.g., sand play, snack time), with >30 hours of video data. Using semi-automatic transcriptions (227,624 words), we find that activity contexts in the preschool classroom vary in both the quality and quantity of the language that children both hear and produce. Together, these findings reinforce prior theories emphasizing the contribution of activity contexts in structuring the variability in children's early learning environments.

Curiosity is a fundamental driver of human behavior, and yet because of its open-ended nature and the wide variety of behaviors it inspires in different contexts, it is remarkably difficult to study in a laboratory context. A promising approach to developing and testing theories of curiosity is to instantiate them in artificial agents that are able to act and explore in a simulated environment, and then compare the behavior of these agents to humans exploring the same stimuli. Here we propose a new experimental paradigm for examining children's -- and AI agents' -- curiosity about objects' physical interactions. We let them choose which object to drop another object onto in order to create the most interesting effect. We compared adults' (N=155) and children's choices (N=66; 3-7 year-olds) and found that both children and adults show a strong preference for choosing target objects that could potentially contain the dropped object. Adults alone also make choices consistent with achieving support relations. We contextualize our results using heuristic computational models based on 3D physical simulations of the same scenarios judged by participants.