Influence of language on perception and concept formation in a brain-constrained deep neural network model

Abstract

A neurobiologically constrained model of semantic learning in the human brain was used to simulate the acquisition of concrete and abstract concepts, either with or without verbal labels. Concept acquisition and semantic learning were simulated using Hebbian learning mechanisms. We measured the network's category learning performance, defined as the extent to which it successfully (i) grouped partly overlapping perceptual instances into a single (abstract or concrete) conceptual representation, while (ii) still distinguishing representations for distinct concepts. Co-presence of linguistic labels with perceptual instances of a given concept generally improved the network's learning of categories, with a significantly larger beneficial effect for abstract than concrete concepts. These results offer a neurobiological explanation for causal effects of language structure on concept formation and on perceptuo-motor processing of instances of these concepts: supplying a verbal label during concept acquisition improves the cortical mechanisms by which experiences with objects and actions along with the learning of words lead to the formation of neuronal ensembles for specific concepts and meanings. Furthermore, the present results make a novel prediction, namely, that such ‘Whorfian’ effects should be modulated by the concreteness/abstractness of the semantic categories being acquired, with language labels supporting the learning of abstract concepts more than that of concrete ones. This article is part of the theme issue ‘Concepts in interaction: social engagement and inner experiences’.

Concepts in interaction: social engagement and inner experiences

“…Other papers focus more explicitly on language and its relationship with concepts, showing the importance of inner language, verbal labels and word associations for concrete and abstract concepts, and highlighting the role of language in enhancing cognition. Across the various sections, the theme issue offers many insights into the differences between kinds of concepts, from the significant distinction between concrete and abstract ones [ 2 , 8 , 9 , 56 , 57 ], to specific concepts like the religious [ 58 ], the social [ 59 , 60 ], the olfactory [ 61 ] and the emotional ones [ 38 , 42 ].…”

“…Some examples are simulations of the emergence of categories in individuals and populations, and new computational models (e.g. [ 57 , 62 ]), new ecological methods (e.g. [ 64 ]), new sophisticated data analysis techniques, including cross-linguistic analyses (e.g.…”

Causal influence of linguistic learning on perceptual and conceptual processing: A brain-constrained deep neural network study of proper names and category terms

“…Secondly, only a small number of conceptual features were realized, and a small set of shared features determined concept membership. This situation may hold for some concrete terms but not for others and certainly not for abstract concepts ( Henningsen-Schomers et al, 2022 ). Furthermore, PN and CT were acquired by different networks to allow straightforward separation and evaluation of the mechanistic side of different label types—although label types are normally copresent in the same mind and brain.…”

“… For details and a more elaborate discussion of the corresponding equations as well as their mathematical implementations, please see Henningsen-Schomers et al (2022) . …”

“…Previous research found that several of the extrasylvian areas targeted by the deep neural model (including, for example, *V1 and *AT) are important for processing instance- and concept-related information ( Binder et al, 2005 ; Martin, 2007 ; Ralph et al, 2017 ; Henningsen-Schomers et al, 2022 ). Therefore, the current data analyses and statistical testing focused on the extrasylvian region of the deep neural network.…”