You are viewing the SNL 2017 Archive Website. For the latest information, see the Current Website.

 
Poster E55, Friday, November 10, 10:00 – 11:15 am, Harborview and Loch Raven Ballrooms

Using a novel Local Heterogeneity Regression method to index orthographic representations in reading.

Jeremy Purcell1, Brenda Rapp1;1Department of Cognitive Science, Johns Hopkins University, USA

Amid the rich literature regarding the neural basis of written language, one prominent theory posits that left ventral occipitotemporal cortex (vOTC) contains orthographic neural representations that are highly selective to well-known written words (Dehaene and Cohen 2011). Initial attempts to index these orthographic representations reported that the mean neural response in the left vOTC is relatively low for high-frequency words as compared to low-frequency words and pseudowords (Kronbichler et al. 2004; Mechelli, Gorno-Tempini, and Price 2003). More recently, it has been proposed that orthographic representations become sparser after learning (i.e. strong activation in a relatively small set of neurons), and that differences in the mean neural response does not capture this relative degree of sparseness (Glezer et al., 2009). Instead more advanced measures that quantify the local neural heterogeneity are better suited to detect changes in the relative sparseness of orthographic neuronal representations due to learning (e.g. Glezer et al. 2015). To build upon this work, here we introduce a novel Local-Heterogeneity Regression (Hreg) Analysis which quantifies the relative heterogeneity of the local neural responses within the context of a specific condition - such as reading words. This approach is based on the premise that sparse neural representations will have heterogeneous local responses across adjoining voxels. We apply this approach to an fMRI reading study which included high frequency (HFW), low frequency words (LFW), and pseudowords (PW). We acquired block design, reading data for the following conditions (N=40): HFW, LFW, PW, and checkerboards. Two analyses were performed: (1) A traditional random-effects univariate analysis. (2) A novel Local-Hreg analysis. The Local-Hreg analysis is a search-light analysis where, for each search-light, a general psychophysiological interaction analysis (gPPI; McLaren et al., 2012) was performed using the center voxel to predict each surrounding voxel in a pair-wise manner. For each pair-wise comparison a condition-specific (e.g. reading HFW) voxel-to-voxel interaction parameter was obtained; the Local-Hreg value is the median of condition-specific pairwise, interaction values within a searchlight. Lower average condition-specific interactions, indicate higher local heterogeneity. For both analyses we performed comparisons for HFW > LFW and HFW>PW. Results reveal: (1) in the univariate analysis: a left hemisphere vOTC cluster that shows lower mean activation for HFW relative to LFW, and (2) in the Hreg analysis: a left vOTC cluster that shows higher local heterogeneity for HFW relative to LFW. A similar finding was observed for HFW>PW for both analyses. In summary, whereas there is a lower BOLD response to HFW, there is higher local heterogeneity for HFW relative to both LFW and PW within the left vOTC. This work provides a novel approach for examining the relative sparseness of orthographic representations, and has applications for probing the neural dynamics of representation and learning.

Topic Area: Perception: Orthographic and Other Visual Processes

Back to Poster Schedule