Predicting Speech: Cortical Oscillations and Dynamics

言语是一种复杂多变的媒介，然而，当我们用母语听的时候，理解似乎是毫不费力和自动的。同时，通过对我们母语的经验，我们对它在声音、单词和句子层面上的结构有了丰富的知识。最近的工作已经开始强调这种知识在实时语言处理中的重要性。一种可能塑造语言理解毫不费力的本质的重要机制是利用对语言结构的先验知识在语言分析的各个层面预测即将到来的信息，从而促进对未来信息的处理。我的五年研究计划是为了确定这一预测是如何在语音感知过程中神经振荡的时间和空间动力学中编码的，语音感知是语言理解的最早阶段。特别是，这项研究建立在最近的发现基础上，这些发现确定了语音感知过程中的一个大脑预测标记。 我的长期研究目标是将语言学的发现与认知神经科学结合起来，以了解言语感知的大脑基础。我的短期研究目标集中在言语感知过程中预测的“何时”、“什么”和“何处”，主要围绕以下三个研究目标： 1.预测的空间局部化：这个预测的大脑标记在大脑的什么地方，它是否与参与语音感知的其他网络重叠，以及它与其他语音识别指数有何关系？ 2.驱动预测的语音线索：语音表征的哪些方面有助于驱动这种预测机制？虽然我之前的工作已经证明了语音发音的对比发音特性的作用，但其他线索是否也在发挥作用，包括语音共现、统计线索和单词结构？ 3.预测的语言特异性：这些预测机制是通用的，还是仅仅依赖于语音的语言属性：音乐与语音、不同的说话者、不同的语速？对不同类型信息的预测是以不同的振荡大脑模式编码的吗？ 这项研究应用于构建生物学上看似合理的电路，模拟语言处理，这是人工智能研究所必需的，通过精确定位准确感知和产生所需的相关言语属性，帮助恢复各种言语语言运动障碍，以及失语症，失语症的缺陷可能部分是由于失语症大脑在与另一个人交谈时无法同步语音线索，后两者在加拿大产生了重大影响，因为我们对儿童的言语运动障碍变得更加敏感，随着人口的老龄化，失语症病例的数量也随之增加。

Speech is a complex and variable medium, and yet when we listen in our native language, comprehension appears effortless and automatic. Simultaneously, through experience with our native language, we have a wealth of knowledge about how it is structured at the level of sounds, words, and sentences. Recent work has begun to emphasize the importance of this knowledge during real-time language processing. One overarching mechanism that could shape the effortless nature of language comprehension is leveraging prior knowledge of our language's structure to predict upcoming information at all levels of linguistic analysis, thereby facilitating the processing of future information. My five-year research program is to identify how this prediction is encoded in the temporal and spatial dynamics of neural oscillations during speech perception: the earliest stages of language comprehension. In particular, this research builds on recent findings which have identified a brain marker of prediction during speech perception. My long-term research goal is to incorporate findings from linguistics with cognitive neuroscience to understand the brain bases of speech perception. My short-term research goals focus on the “when”, “what” and “where” of prediction during speech perception and are centred around the following three research objectives: 1. Spatial Localization of Prediction: Where in the brain is this brain marker for prediction, does it overlap with other networks involved in speech perception, and how does it relate to other indices of speech recognition? 2. Speech Cues that Drive Prediction: What aspects of speech sound representations contribute to driving this predictive mechanism? While my previous work has demonstrated a role for the contrastive articulatory properties of speech sounds, could other cues also be at play, including speech sound co-occurrences, statistical cues, and word structure? 3. Language Specificity of Prediction: Are these predictive mechanisms general or do they rely solely on the linguistic properties of speech sounds: music vs. speech, different speakers, different speaking rates? Is the prediction of different types of information encoded in distinct oscillatory brain patterns? This research has applications for the construction of biologically plausible circuits that model language processing, necessary for Artificial Intelligence research, aid in the rehabilitation of various speech language motor-deficits by pinpointing the relevant speech properties necessary for the accurate perception and production, as well as in aphasias, where the deficit may partially result from an inability of aphasic brains to synchronize speech cues when talking to another individual, the latter two having major impacts in Canada both as we become more sensitive to speech-motor deficits in children and as our population ages, and with it, the number of cases of aphasia.