The Predictive Coding Account of Schizophrenia: Dysfunctional Interaction across Linguistic Levels?

精神分裂症的特点是在语言层次的多个层面上发生明显的语言功能障碍，从较低级别的听觉感知到较高级别的语义处理。迄今为止，精神分裂症语言缺陷的神经病理学仍未得到解决。在这里，我们的目标是在预测编码的框架内提供这些语言缺陷的统一电生理学解释，这将精神分裂症的语言功能障碍解释为预测和传入的感觉输入之间的不平衡。具体来说，从现有的研究封装语言水平内的语言功能障碍的临床神经科学研究出发，我们假设患者在较高抽象语言水平的预测与较低听觉感觉水平的预测误差之间的相互作用中受到损害，并且对这些损害的研究可能为幻听背后的功能缺陷提供现象学解释。为此，我们将采用脑电图（EEG）来比较三组参与者之间的在线语音感知和语言处理：有或没有幻听的精神分裂症患者，以及匹配的健康对照，并具有三个工作包（WP）。在 WP1 中，我们使用经典的奇怪范式和基于句子的范式以音节省略形式检查听觉感知，以测试精神分裂症中统计（奇怪）/语义（句子）预测和听觉感知之间的相互作用是否受损。在 WP2 中，将采用自然主义范式，其中通过最先进的计算建模获得语义预测的逐字索引和音素级预测误差。在 WP3 中，我们通过静息态脑电图检查兴奋和抑制（im）平衡。我们假设功能失调的语言预测的脑电图标记（事件相关电位、窄带振荡）在患者和对照组之间以及有或没有幻听的患者之间是可分离的。此外，WP1-3 的脑电图标记将在 WP4 中使用现代机器学习方法进行进一步分析，以测试电生理学帐户是否可以支持基于脑电图的分类和聚类。总而言之，该项目是第一个通过脑电图解决精神分裂症跨语言水平的语言预测受损和预测错误的项目，不仅将提供对精神分裂症语言缺陷的神经生物学理解，而且还将提供有利于临床实践的翻译知识。

Schizophrenia is characterized by marked language dysfunctions occurring at multiple levels of the linguistic hierarchy, ranging from lower-level auditory perception to higher-level semantic processing. To date, the neuropathology of language deficits in schizophrenia remains unresolved. Here, we aim at providing a unitary electrophysiological account of these linguistic deficits within the framework of predictive coding, which explains language dysfunctions in schizophrenia as an imbalance between prediction and incoming sensory inputs. Specifically, stepping from extant clinical neuroscience studies that investigate language dysfunctions within encapsulated linguistic levels, we hypothesize that patients are impaired in the interaction between prediction from higher abstract linguistic levels and prediction error from lower auditory sensory levels, and that the study of these impairments may provide a phenomenological explanation of the functional deficit underlying auditory hallucinations. To this end, we will employ electroencephalography (EEG) to compare online speech perception and language processing between three groups of participants: patients with schizophrenia with and without auditory hallucinations, and matched healthy controls, with three work packages (WPs). In WP1, we examine auditory perception in the form of syllable omission with a classic oddball paradigm and a sentence-based paradigm, to test if interaction between statistic (oddball) / semantic (sentence) prediction and auditory perception is impaired in schizophrenia. In WP2, a naturalistic paradigm will be employed in which word-by-word indices of semantic prediction and phoneme-level prediction error are obtained through state-of-the-art computational modelling. In WP3, we examine excitation and inhibition (im)balance with resting-state EEG. We hypothesize that EEG markers of dysfunctional linguistic prediction (event related potentials, narrow-band oscillations) are dissociable between patients and controls, and between patients with and without auditory hallucinations. Additionally, the EEG markers from WP1–3 will be further analyzed in WP4 with modern machine learning methods, to test if the electrophysiological account could support EEG-based classification and clustering. To summarize, being the first to address impaired linguistic prediction and prediction error across linguistic levels in schizophrenia with EEG, the project will not only provide a sharpened neurobiological understanding of language deficits in schizophrenia, but also will provide translational knowledge that benefits clinical practice.