Cognitive Computational Neuroscience of Language: Using Computational Models to Investigate Language Processing in the Brain
Name
nol_e_00131.pdf
Description
Published version
Size
194.77 KB
Format
Adobe PDF
Checksum (MD5)
4e81133d4703d170c2aec2d98d55bfc8
Author(s)
Lopopolo, Alessandro
Fedorenko, Evelina
Levy, Roger
Rabovsky, Milena
Date Issued
April 1, 2024
Journal
Neurobiology of Language
Publisher
MIT Press
Citation
Alessandro Lopopolo, Evelina Fedorenko, Roger Levy, Milena Rabovsky; Cognitive Computational Neuroscience of Language: Using Computational Models to Investigate Language Processing in the Brain. Neurobiology of Language 2024; 5 (1): 1–6.
Version
Final published version
Abstract
From the inception of large-scale deep learning models to the development of cognitively inspired artificial neural networks (ANNs), computational modeling has ushered in a new era of exploration into language processing within the human brain. This special issue serves to showcase cutting-edge research in the field, all united by a common interest in employing computational models as a tool for generating and testing theories, including through methodological innovations. These studies not only underscore the remarkable progress achieved by the computational cognitive neuroscientific approach but also shed light on the immense potential that this dynamic discipline holds, promising groundbreaking insights into the mechanisms that underpin our language capacity.
The goal of neurobiology of language is to identify the neural substrates of linguistic computations and representations. This pursuit takes inspiration from a broad range of disciplines (e.g., cognitive science, linguistics, neuroscience, and neurophysiology) and domains (e.g., vision, memory, attention) to generate hypotheses that are tested against neurobiological experimental observations. However, despite considerable efforts, attempts to connect abstract theoretical constructs and the concrete properties of the human brain have encountered significant challenges, partly because it is unclear how to map between the “part list” of cognition and that of neurobiology—the “mapping problem” (Embick & Poeppel, 2015; Poeppel, 2012). The difference of these two domains necessitates a nuanced and interdisciplinary approach that respects their inherent complexities.
Computational modeling approaches can bridge the ontological gap between cognition and neurobiology by providing a means of transforming hypotheses into implemented models and stimuli into numeric descriptors. This advantage holds for not only advanced deep learning models and neural encoding analysis methods, but also more traditional tools, such as parsers that return syntactic trees, language models that provide probabilistic descriptions of stimuli (e.g., surprisal) or make quantitative predictions about processing difficulty, or traditional vector-space models based on distributional semantics. In other words, computational models transform verbal hypotheses into numeric representations or measures that can be mapped onto neural (or behavioral) data. The papers featured in this issue are unified by their common reliance on computational models of language processing in order to address the challenge of linking theoretical constructs to neural data.
Terms of Use
Creative Commons Attribution
Persistent DSpace Link
DOI of Published Version
https://doi.org/10.1162/nol_e_00131
