Infant is the most powerful learner: He learns in a few months to master language, complex social interactions, etc. Powerful statistical algorithms, simultaneously acting at the different levels of functional hierarchies have been proposed to explain learning. I propose here that two other elements are crucial. The first is the particular human cerebral architecture that constrains statistical computations. The second is the ability to access a rich symbolic system. In 6 work packages using the complementary information offered by non-invasive brain-imaging techniques, I have studied the neural bases of infant statistical computations and symbolic competence during the first months. The goal was to clarify the specificities of a neural functional architecture that is critical for human learning from the onset of cortical circuits.
Through the project, we were able to show that many functional features characteristics of the human neural architecture described in adults are seen from start revealing a strong biological constraint on the neural networks that are nevertheless modulated by the environment. For example, the superior temporal sulcus, which hosts the verbal and non-verbal human communication system, is deeper on the right side in almost all humans, but its depth is modulated by the term at birth, that is probably by the auditory environment. We also found that phonemes, number and musical pitch are part of the initial representations automatically computed (i.e. even during sleep) by the infant brain. These representations are abstract, in the sense that they are not dependent on a single sensory feature but consist in a multi-level integration, even across the auditory and visual modalities in the case of number.
These representations are robust to local variations and allow conditional statistical computations, that is the computation of the probability of an event given the previous event (i.e. transitional probability). We showed that the computation of transitional probabilities between auditory events is already present at 6 months of gestation in preterm neonates. This computation allows the chunking of artificial stream in triplets in sleeping neonates as in attentive adults. Interestingly, the same operation on quadruplets is not possible at both ages revealing a hard limit of 4 items in the verbal short-term memory which appears to be constant across ages, attention and linguistic expertise.
Finally, infants are not limited to automatic processes but can recover explicit representations, that can even be symbolic allowing the understanding of negation (A and nonA), abstract rules and logic already in the first semester at a preverbal age. As adults, the access to a conscious space is limited by a bottleneck whose duration is longer in infants, around 1 second vs ~300 ms.
To summarize, thanks to the use of state of the art brain imaging techniques, our work has revealed the complex operations human infants are already computing during the first semester to structure their environment and the robust similarities in terms of operations between infants and adults. Thus the difference between ages is not in the impossibility of complex operation by infants but more in the slowness of these operations in infants. These similarities between infants and adults, even for very abstract operations involving the use of symbols and logical reasoning, suggest strong genetic bases structuring a specific human brain architecture underlying these high-level functions.
