Global and local aspects of temporal and lexical predictions for speech processing

人脑的特点是不断保持对即将发生的事件的感官预测，从而优化其性能。这种预测机制主要在视觉上进行了研究，在某种程度上，也在听神经科学中进行了研究。对于语音理解，特别是对于单段处理，对预测机制的系统描述在某种程度上被忽视了。为此，拟议的研究项目旨在系统地检查语音中与语音片段、单词和意义相关的预测机制。重要的是，关于语音事件的时间的预测(时间预测)将与关于语音事件的词汇特征的预测(特征预测)区分开来。这种区别涉及语音事件之间的同步在多大程度上有利于处理的问题，另一方面涉及诸如语音出现频率之类的语言特定属性在多大程度上影响预测处理的问题。虽然之前的研究集中在时间预测或特征预测上，但本项目中的7个实验将系统地评估时间预测和特征预测的交互作用，并区分由单词上下文生成的预测和由句子上下文生成的预测。在方法上，所有实验都将基于事件相关脑电位(ERPs)，即通过脑电(EEG)记录获得的。这保证了对大脑活动的既定测量，具有非常好的时间分辨率。此外，以前的研究表明，预测效果发生在早期的ERP潜伏期，因此该方法对于本项目的目的是最佳的。所有拟议实验的主要问题如下：(A)语音处理中的预测机制与一般听觉处理中的预测机制有多大程度的不同？(B)具有交互处理水平的现有言语感知模型如何将这种机制结合起来，其特征将由本项目加以完善？

The human brain is characterized by constantly maintaining sensory predictions about upcoming events, thereby optimizing its performance. Such predictive mechanisms have been mainly investigated in the visual, and to some degree, in the auditory neurosciences. For speech comprehension, and in particular, for single-segment processing, a systematic account of predictive mechanisms has somewhat been neglected. To that end, the proposed research project aims at systematically examining predictive mechanisms in speech that pertain to the level of speech segments, words, and meanings. Importantly, predictions regarding the timing of speech events (temporal predictions) will be distinguished from predictions regarding the lexical features of speech events (featural predictions). This distinction relates to the question to what degree synchrony between speech events is beneficial for processing and, on the other hand, to what degree language-specific properties such as frequency of speech sound occurrence influences predictive processing. While previous studies focused on either temporal or featural predictions, the 7 experiments in this project will systematically assess the interaction of temporal and featural predictions, and also distinguish between predictions generated by a word context and predictions generated by a sentence context. Methodologically, all experiments will be based on Event-Related brain Potentials (ERPs), obtained by Electroencephalography (EEG) recordings. This guarantees an established measure of brain activity with a very good temporal resolution. Furthermore, previous studies have shown that predictive effects occur at early ERP latencies, such that this method is optimal for the purpose of this project. The overarching questions for all proposed experiments are the following: (a) To what degree do predictive mechanisms in speech processing differ from predictive mechanisms in general auditory processing? (b) How can existing models of speech perception with interactive levels of processing integrate such mechanisms, the characterization of which shall be refined by this project?