From Neurons to Symbolic Representation and Rapid Learning in Mind and Brain: A Brain-Constrained Neural Model

02.02.2024, 10:15 - 11:15  –  room 0.108, building 28
SFB-Kolloquium

Rosario Tomasello, FU Berlin

To grasp the meaning of words and their relationship to the outside world, higher cognitive processes unique to the human brain are at work. However, despite decades of research on the neural substrates of conceptual processing, a consensus about the functions and components of the semantic system has not been reached among neuroscientists. Additionally, the unique striking ability of humans to instantaneously map a novel word together with its related referent, known as "fast mapping", poses several further challenges to research in the neurobiology of language and acquisition. A fruitful approach to examining the brain mechanisms underlying conceptual representation and learning is the use of computational neural networks that resemble critical functional and physiological features of the human brain and open a unique avenue to investigate how symbolic internal representations are rapidly built at the neural, synaptic and cortical levels.  
Following this strategy, a biologically constrained cortex model that mimics the anatomical and physiological features of the frontotemporal-occipital regions was built to simulate the learning of new words in the context of object perception and action execution, mimicking typical language acquisition scenarios. The simulations demonstrate that meaningful linguistic units in the brain are represented in the form of cell assemblies that have spontaneously arisen through the mutual interaction of a single set of biological mechanisms operating within specific neuroanatomical structures. Crucially, when implementing the model with a two-step learning process, equipping the networks with a pre-existing repertoire of referential and phonological representations before word meaning mapping takes place, it enables ultra-rapid associative semantic learning, as documented in empirical studies on fast mapping. Overall, I will demonstrate that through the mutual effect of learning, cortical semantic areas and connectivity structure are sufficient to provide a straightforward explanation for conceptual representations in mind and brain and offer an account for the rapid semantic interlinking of phonological and conceptual circuits.

Introducing "What is...?" lecture for young scientists from 9:15 to 10 am in room 1.22, building 9.