Distributed Phonetic Representations in the Brain

描述（由申请人提供）：神经影像学研究通过提供关于语音在大脑中如何处理的见解，提高了我们对人类语言和语言处理的理解。目前大多数fMRI研究缺乏统计和描述能力，无法解析以分布式活动模式编码的复杂信息，这使得它们难以适应现实世界中语言感知的复杂性。相比之下，最近的研究使用新颖的多变量方法进行fMRI分析，揭示了神经活动的分级、分布模式，有望提供大脑感知分类的详细、定量描述。范畴知觉需要增强不同类别刺激之间的对比，增强同一类别刺激之间的相似性。本建议的第一个具体目标是发现与这一过程有关的活动模式。当受试者被动地听许多独特的、自然地重新合成的音节时，将收集高分辨率的、与事件相关的fMRI数据。可用于识别刺激类别的活动模式将被识别，并使用多维尺度分析来探索它们定义的感知相似性空间。同样类型的分析将被应用于行为数据，从而产生一种探索大脑-行为关系的新方法。第二个具体目标是探索语音分类的多变量分析提出的方法学问题，特别是：识别包含刺激身份信息的神经活动模式的最佳方法是什么？一个神经网络分类器将被训练来根据对刺激的神经反应来确定在每次试验中出现了哪个音节。然后将进行一系列测试，以确定该方法是否优于标准单变量技术，或者是否可以将基于分类器和基于单变量的方法的互补优势结合起来。关于这些分析的许多技术细节将被详细探讨，以便得出一套“最佳实践”，这将使这项技术能够最佳地整合到一个更大的研究项目中，包括多模态神经成像和行为研究。如果成功，本研究将为研究语音感知的神经基础提供一套新的分析工具。这些技术将适用于成人加工、典型发展和一系列沟通障碍的研究，包括阅读障碍和特殊语言障碍，其中言语感知缺陷可能起着核心作用。具体来说，它将提供一种方法来更好地表征言语类别表征中的个体差异，并探索有关不同障碍中缺陷本质的更详细的假设，而不是目前主流技术所能做到的。

DESCRIPTION (provided by applicant): Neuroimaging research has advanced our understanding of human speech and language processing by providing insights about how speech sounds are processed in the brain. Most current fMRI studies lack the statistical and descriptive power to resolve complex information encoded in distributed patterns of activity, making them an awkward fit to the complexities of speech perception in the real world. In contrast, recent studies using novel multivariate approaches to fMRI analysis have revealed graded, distributed patterns of neural activity that promise to provide detailed, quantitative descriptions of perceptual categorization in the brain. Categorical perception requires enhancing contrast between stimuli of different categories and enhancing similarity between stimuli from the same category. The first specific aim of this proposal is to discover patterns of activity related to this process. High-resolution, event-related fMRI data will be collected while subjects passively listen to many unique, naturalistically resynthesized syllables. Patterns of activity that can be used to identify stimulus categories will be identified, and analyzed using multidimensional scaling analyses to explore the perceptual similarity space they define. The same type of analysis will then be applied to behavioral data, resulting in a novel means of exploring brain-behavior relationships. The second specific aim is focused on exploring methodological issues presented by multivariate analysis of speech categorization, in particular: What is the best way to identify patterns of neural activity that contain information about stimulus identity? A neural network classifier will be trained to determine which syllable was presented on each trial based on the neural response to that stimulus. A series of tests will then be conducted to determine whether this approach provides advantages over standard univariate techniques, or whether the complementary strengths of classifier- and univariate-based methods can be combined. A number of technical details regarding these analyses will be explored in detail, in order to arrive at a set of "best practices" that will permit this technique to be optimally integrated into a larger program of research including multi-modality neuroimaging and behavioral studies. PUBLIC HEALTH RELEVANCE If successful, the proposed research will provide a new set of analytic tools for the study of the neural basis of speech perception. These techniques will be applicable to research on adult processing, typical development, and a range of communication disorders -- including dyslexia and specific language impairment -- in which speech perception deficits may play a central role. Specifically, it will provide a means to better characterize individual differences in the representation of speech categories, and to explore more detailed hypotheses about the nature of deficits in different disorders than is possible with currently predominant techniques.