A Novel Approach to Measuring Neural Tuning to Written Words

点击翻译按钮获取中文摘要

Project Description The written word is not represented in a uniform manner in the brain. Instead, different features of the written word are thought to be differentially represented in the same or potentially neighboring, but distinct cortical regions. For instance, cortical areas predominantly in, but not limited to left ventral occipital-temporal cortex (vOTC) are thought to represent different features such as entire word units (e.g. [MINT]), bigrams ([MI], [IN], [NT], or the sub-lexical mappings of letters to sounds ([M]-/m/, [I]-/ɪ/, [N]-/n/, [T]-/t/). In order to develop a more mechanistic view of what aspects of the written word are effectively processed in skilled and impaired readers, it is important to be able to parse apart experience-dependent neural tuning of these different features of printed words. For instance, individuals may have poorly tuned letter-sound units, yet well- tuned bigram units or lexical units. This is a central question in the study of developmental dyslexia and reading impairments. The Specific Aims of this project are to address this question by systematically investigating the multivariate nature of representation in cortex and the tuning of cortical responses to these features concurrently across individuals with a range of reading skill. In doing so we will address a) how features of word forms are distributed across the “reading network”; b) how different neuronal populations become attuned to these different features of the written word in an experience-dependent manner; and c) how the tuning of these different orthographic features predicts reading performance across individuals. To accomplish this, we employ a set of innovative methods for quantifying neural response heterogeneity across voxels with the assumption, based on sparse coding theory, that highly tuned feature representations have more heterogeneous (i.e., unique) neural responses across voxels. The main focus of Aim 1 will be to validate the use of Representational Similarity Analysis (RSA) in combination with a novel multivariate analytic method in fMRI termed Heterogeneity Regression (Hreg) to form a metric of Tuning Similarity Analysis (TSA). Aim 2 is to determine whether this novel set of metrics is predictive of reading performance across the range of typically achieving readers as well as indicative of poor or impaired reading ability. This would confirm that this heterogeneity based approach can index the content and quality of representations in skilled word reading as well as the potential source of reading failure. If successful, the Broader Impacts of this approach could be used to determine representational integrity not only to difficulties in visual word recognition, but also for spoken language processing and other domains of multidimensional perceptual learning complementing behavioral diagnostics.