What goes on in our brains and minds when we learn concepts and the meaning of symbols? I propose to try it out. By creating networks of neuron-like elements, by connecting these as they are in local cortical networks, areas, and, at a larger scale, in the human connectome, by implementing regulation and control mechanisms and, most importantly, by applying realistic learning procedures. Such “brain-constrained” neural networks can be treated like infants who experience objects with varying degree of similarity and later-on words and larger chunks of language in their context. We can then ask what goes on in the brain-like architectures when they experience the world and when they learn symbols ¹.

One result of this endeavor is (of course) that the answer very much depends on the brain-constraints implemented. Networks with sequential area links, but no within-area connections, build fully distributed dynamic patterns. However, adding reciprocal between-area links, within-area excitatory connections and local inhibition leads to formation of strongly connected neuronal assemblies or circuits. A circuit includes many neurons and may be distributed across different network areas. Importantly, it may ‘ignite’ as a whole and maintain its reverberant activity for some time ². Concept formation may be based on neuronal circuit formation, and symbol learning may be the result of interlinking a conceptual circuit with a symbol circuit implementing the sound and articulatory pattern of a spoken word form.

The talk will report on some recent results from simulating – and possibly explaining at a neurobiological level – spontaneous concept formation, as it is seen in preverbal infants ³. Simulation experiments will target the learning of different symbol types, including concrete and abstract concepts/symbols, along with questions about their structural differences ^4,5. The mechanistic basis of causal effects of symbol learning on perception and cognition will be highlighted in the context of learning specific and general symbols ⁶.

This mechanistic modelling work can be used to found semantic learning and symbol grounding in concrete neurobiological mechanisms. The results can be seen as a best guesses about the neurobiological reality of aspects of cognition and language ⁷. Whether these guesses are right needs to be evaluated with physiological and other experimental measures.

1 Pulvermüller, F., Tomasello, R., Henningsen-Schomers, M. R. & Wennekers, T. Biological constraints on neural network models of cognitive function. Nat Rev Neurosci 22, 488-502, doi:10.1038/s41583-021-00473-5 (2021).

2 Pulvermüller, F., Garagnani, M. & Wennekers, T. Thinking in circuits: Towards neurobiological explanation in cognitive neuroscience. Biol Cybern 108, 573-593, doi:10.1007/s00422-014-0603-9 (2014).

3 Henningsen-Schomers, M. R. & Pulvermüller, F. Modelling concrete and abstract concepts using brain-constrained deep neural networks. Psychol Res 86, 2533-2559, doi:10.1007/s00426-021-01591-6 (2022).

4 Henningsen-Schomers, M. R., Garagnani, M. & Pulvermüller, F. Influence of language on perception and concept formation in a brain-constrained deep neural network model. Philos Trans R Soc Lond B Biol Sci 378, 20210373, doi:10.1098/rstb.2021.0373 (2023).

5 Dobler, F. R., Pulvermüller, F. & Henningsen-Schomers, M. R. Temporal dynamics of concrete and abstract concept and symbol processing: A brain constrained modelling study. Language Learning 74, 258-295, doi:https://doi.org/10.1111/lang.12646 (2024).

6 Nguyen, P. T. U., Henningsen-Schomers, M. R. & Pulvermuller, F. Causal Influence of Linguistic Learning on Perceptual and Conceptual Processing: A Brain-Constrained Deep Neural Network Study of Proper Names and Category Terms. J Neurosci 44, doi:10.1523/JNEUROSCI.1048-23.2023 (2024).

7 Pulvermüller, F. Neurobiological Mechanisms for Language, Symbols and Concepts: Clues From Brain-constrained Deep Neural Networks. Prog Neurobiol in press, 102511, doi:10.1016/j.pneurobio.2023.102511 (2023).