摘要侵入式语言脑机接口（BCI）可绕过发声通路，直接将大脑语言相关皮层的活动转换为文字或语音输出，为重度语言障碍患者重建沟通能力。目前该系统已在临床受试者中初步实现较高精度的语言解码，但仍面临依赖长期和大规模侵入式记录、临床推广困难等挑战。本文概述语言BCI的技术框架，并与运动BCI对比，分析其在可量化神经表征与可迁移解码模型方面的不足；在此基础上，提出未来研发应围绕"可量化的语言神经表征-可对齐的低维特征空间-具迁移能力的预训练解码模型"，结合普通话音节与声调结构及中文语言模型，构建面向本土临床需求、具有推广潜力的技术体系。

abstractsInvasive language brain-computer interface (BCI) can bypass the vocalization pathway and directly convert the language-related cortical activity in the brain into text or speech output, thereby restoring communication abilities for patients with severe language disorders. Currently, this system has achieved relatively high-precision language decoding in clinical trial participants, but it still faces challenges such as reliance on long-term and large-scale invasive recordings and difficulties in clinical promotion. This article summarizes the technical framework of language BCI and compares it with motor BCI, analyzing their deficiencies in quantifiable neural representations and transferable decoding models. On this basis, it is proposed that future research and development should focus on "quantifiable language neural representation-alignable low-dimensional feature space-transferable pre-trained decoding model", and establish a technical system meeting the local clinical needs and with promotion potentials, by combining the structures of Mandarin syllables and tones with Chinese language models.