A unified cognitive network model of language

目前大多数理解认知过程的神经基础的方法都严重局限于 两个方面。首先，最常用的方法不具有时间(例如，fMRI)或空间(例如， 脑磁图(MEG/EEG)分辨率，以捕捉相关动态。第二，时空高度均匀的方法 分辨率(颅内EEG-icEEG)通常以一种零碎的、非... 综合方式。例如，语言通常被研究为独立的子系统的集合体： 感知、模式识别、分类、语义/句法上适当的响应选择 跨模式集成、运动控制和传感器运动集成。目前的建议旨在补救 使用ICEEG来研究模型系统，阅读/语音/语言，从一个综合的和 统一的视角。我们专注于阅读，这是一项复杂的任务，涉及视觉模式识别、视觉- 听觉和视觉-运动整合，语义、句法和语音通达，以及(在朗读中)- 响应选择和马达排序。阅读允许简单但在生态上有效的操作 语言系统中的认知负荷。我们建议测试的神经计算框架是 计算不是通过信息通过一系列离散的处理阶段实现的 单个模块，但通过分布式网络的状态转换。我们将招募一大批80人 我们将量化患者在单词阅读过程中的局部和跨区域皮质动力学- 从早期的初级视觉感知，到选择，再到文字输出。我们将利用我们已建立的 绕过稀疏的分组ICEEG数据的精确共定位和分析技术 人类ICEEG实验固有的采样问题。硬膜下格栅电极的联合应用 立体脑电深度电极将不仅能够研究经典的侧脑室周围区 区域，但也有深沟(和区域，如颞平面)。然后我们将描述 在HIGH中使用幅度协方差的线性和非线性测量的动态网络交互 频率，在我们之前开发的分析之后。关键节点和关键过渡 然后，网络状态将通过闭合环路活动触发的直接皮质刺激而受到干扰。至 为了实现这些目标，我们已经在德克萨斯州综合癫痫项目之间建立了合作 和约翰·霍普金斯医学中心--这两个中心都有用icEEG学习语言的可靠记录。 我们的团队在语言、阅读、ICEEG信号分析、人口水平网络等各个方面都有专业知识 根据颅内记录进行建模；以及神经网络。这项工作将大大提高我们的 理解语言系统，测试并开发一种模拟神经计算的新方法 一般说来。

Most current approaches to understanding the neural basis of cognitive processes are severely limited in two respects. First, most commonly used methods do not have the temporal (e.g., fMRI) or spatial (e.g., MEG/ EEG) resolution to capture the relevant dynamics. Second, even methods with high spatio-temporal resolution (intracranial EEG - icEEG) typically approach target cognitive processes in a fragmentary, un- integrated way. For instance, language is typically studied as a conglomeration of separate subsystems: perception, pattern recognition, categorization, semantically/syntactically appropriate response selection, cross-modal integration, motor control and sensorimotor integration. The present proposal aims to remedy both limitations by using icEEG to study a model system, reading/speech/language, from an integrative and unified perspective. We focus on reading, a complex task that involves visual pattern recognition, visual- auditory and visuo-motor integration, semantic, syntactic and phonological access, and (in reading aloud) - response selection and motor sequencing. Reading allows for easy, yet ecologically valid manipulations of cognitive load in the language system. The neuro-computational framework we propose to test is that computation is achieved not by information passing through a sequence of discrete processing stages in individual modules but via state transitions of a distributed network. We will recruit a large cohort of 80 patients in whom we will quantify both local as well as inter-regional cortical dynamics during word reading - from early primary visual perception, through selection, to word output. We will leverage our established techniques for precise co-localization and analysis of grouped icEEG data, circumventing the sparse sampling problem inherent to human icEEG experiments. The combined use of sub-dural grid electrodes and stereo-electroencephalographic depth electrodes will enable the study of not only classic peri-sylvian regions, but also of deep sulci (and regions such as the planum temporale). We will then characterize dynamic network interactions using linear and non linear measures of amplitude covariance in high frequencies, following analyses we have developed previously. Critical nodes and critical transitions in network states will then be perturbed using closed-loop activity-triggered direct cortical stimulation. To achieve these goals we have set up a collaboration between the Texas Comprehensive Epilepsy Program and Johns Hopkins Medical Center - both centers have a proven record of studying language with icEEG. Our team has expertise in all aspects language, reading, icEEG signal analysis, population level network modeling from intracranial recordings; and neural networks. This work will dramatically improve our understanding of language systems and test and develop a new way to model neural computation generally.