MSc. Jason Khoury

Infants frequently touch their own bodies from the earliest months of life, raising questions about whether these self-directed actions reflect active body exploration. We hypothesize that infants’ self-touch actions are not random but rather occur in distinct latent states that differ in exploratory value, with some states yielding higher information gain about body structure and sensorimotor contingencies. To test this hypothesis, we conducted a longitudinal observational study involving four healthy full-term infants (two females, two males) aged between 6 and 37 weeks. Infants were video-recorded in a naturalistic home environment while lying supine, with minimal external distractions. Self-touch actions were manually coded to capture discrete parameters including touch location, duration, and inter-touch intervals. These events were then discretized into a high-dimensional state space and analyzed using Hidden Markov Models (HMMs) to infer latent behavioural states. The entropy of the emission probability matrices from the HMMs provided a measure of state variability, and information gain was quantified using the Kullback-Leibler divergence between successive probability distributions over touch locations. Our analyses revealed two distinct states differing in emission entropy, with the higher-entropy state associated with significantly greater information gain about the body. These findings suggest that early self-touch may serve as an active learning mechanism, enabling infants to gather crucial information about their own bodies.

Previous influential accounts have hypothesized that early self-directed movements may play a fundamental role in the formation of a bodily-self representation. By contacting the body, newborns may experience the unique intermodal perception between proprioception and touch which only pertains to their own body. Although this behaviour has been anecdotally observed in foetuses, it remains partially unexplored in early postnatal life. Here, we aim to provide quantitative evidence of the newborns’ propensity to engage in self-directed movements. We extracted hand-related kinematics from videos recorded in fourteen newborns (12-57 hour-old), in the ecological setting of General Movements’ clinical evaluation. Our results show that newborns engage significantly more in self- than outer-directed movements. Furthermore, self-directed movements often culminate in a body contact (i.e., self-touch), and the trunk is the most frequently targeted body area. We discuss these findings by endorsing self-directed behaviour as a crucial context fostering the emergence of an early bodily-self representation in typical development. In addition, the manually scored dataset presented here is currently being used as a reference for the development of automatic algorithms, which will allow classifying neonatal movements of a larger dataset acquired in these highly ecological conditions (i.e., a simple video recording). Such an innovative tool could help develop kinematic markers of typical development, thus laying the groundwork for investigating its possible alteration in atypical development.

Children are not born tabula rasa. However, interacting with the environment through their body movements in the first months after birth is critical to building the models or representations that are the foundation for everything that follows. We present longitudinal data on spontaneous behavior of three infants observed between about 8 and 25 weeks of age in supine position. We combined manual scoring of video recordings with an automatic extraction of motion data in order to study infants’ behavioral patterns and developmental progression such as: (i) spatial distribution of self-touches on the body, (ii) spatial patterns and regularities of hand movements, (iii) midline crossing, (iv) preferential use of one arm, and (v) dynamic patterns of movements indicative of goal-directedness. From the patterns observed in this pilot data set, we can speculate on the development of first body and peripersonal space representations. Several methods of extracting 3D kinematics from videos have recently been made available by the computer vision community. We applied one of these methods on infant videos and provide guidelines on its possibilities and limitations—a methodological contribution to automating the analysis of infant videos. In the future, we plan to use the patterns we extracted from the recordings as inputs to embodied computational models of learning of body representations in infancy.