Linguistics and Human Brain: A Perspective of Computational Neuroscience

A new computational neuroscience framework proposes bridging the gap between abstract linguistic theories and empirical brain data by formalizing language structures into testable neural models, according to a paper by Nizhuan Wang posted on arXiv [1]. The paper argues that computational neuroscience serves as an interdisciplinary cornerstone, enabling a dialogue between linguistic hypotheses and neural mechanisms through modeling, simulation, and data analysis [1]. This approach draws on the established field of neurolinguistics, which studies the neural mechanisms controlling language comprehension, production, and acquisition using methods from neuroscience, linguistics, and cognitive science [3]. Neurolinguists evaluate linguistic and psycholinguistic theories through aphasiology, brain imaging, electrophysiology, and computer modeling [3]. Recent advances in deep learning, particularly large language models (LLMs), have powerfully advanced this pursuit [1]. LLMs are machine learning models with many parameters, trained with self-supervised learning on vast amounts of text for natural language processing tasks such as language generation [11]. The high-dimensional representational spaces of these models provide a novel scale for exploring the neural basis of linguistic processing [1]. The paper highlights the "model-brain alignment" framework as a methodology to evaluate the biological plausibility of language-related theories [1]. This work sits at the intersection of multiple disciplines. Cognitive science, the interdisciplinary study of the mind and its processes, examines faculties including language, perception, memory, and reasoning, borrowing from psychology, artificial intelligence, neuroscience, linguistics, and anthropology [5]. Behavioral neuroscience, a related subfield, focuses on the biological and neural substrates underlying human experiences and behaviors, with subdivisions including cognitive neuroscience, which emphasizes biological processes underlying human cognition [4]. The broader AI landscape provides context for the paper's reliance on deep learning. Artificial intelligence was founded as an academic discipline in 1956, and the field experienced multiple cycles of optimism followed by periods of reduced funding known as AI winters [7]. Funding and interest increased substantially after 2012, when graphics processing units began accelerating neural networks, and deep learning outperformed previous AI techniques [7]. Growth accelerated further after 2017 with the transformer architecture, and the 2020s saw an AI boom coinciding with advances in generative AI [7]. Companies such as DeepSeek have recently demonstrated that LLMs can be trained at significantly lower cost than previously assumed, with its V3 model reportedly trained for US$6 million compared to the US$100 million cost for OpenAI's GPT-4 in 2023 [9]. The paper was submitted to arXiv on February 9, 2026, and revised on June 25, 2026, with the latest version totaling 29,117 KB [1].