Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Greater fidelity of neural patterns during reading is associated with lower frequency and concreteness ratings, and higher orthographic distance
Poster Session B, Friday, September 12, 4:30 - 6:00 pm, Field House
Cory McCabe1, David Rothlein2, Isabella Walsh1, Noah Nixon1, Lucia Zepeda Rivera3, Michaela Brooks3, William Graves1, Donald Bolger3, Jeremy Purcell3; 1Rutgers University - Newark, 2VA Boston Health Care, 3University of Maryland
Lexical and semantic features are represented across a set of language cortical regions. Previous work has used representational similarity analyses (RSA) to determine that specific regions encode lexical and semantic features in a set of individuals. However, such work does not address whether the representational patterns are more or less likely to be reproduced across individuals in relation to these psycholinguistic variables (e.g., orthographic neighborhood). To address this, here we used fMRI and an RSA approach to quantify the representational fidelity (RF), which is the degree to which stimulus-evoked multivoxel patterns are reliably structured across individuals. Although traditionally used to define the noise-ceiling in RSA, here we used it to examine the representational signal-to-noise ratio for individual words in an fMRI study of reading. We hypothesize that there will be a relationship between RF and both lexical and semantic features within brain regions associated with language processing. 32 participants underwent fMRI scanning while reading 160 words presented 4 times. For each participant, an RSM was generated by cross-correlating the activation patterns within a region of interest (ROI) elicited from each word stimulus. RF was then calculated for each word by correlating each participant's RSM with the group average RSM while leaving that participant out (e.g., the correlation between RSM1 and the mean of RSM2-RSM32). This results in an RF value for each word, or 160 RF values total for each ROI. 38 individual ROIs representing left-hemisphere regions previously associated with lexical and semantic processing were generated using the Yeo 17 – Schaefer 400 parcellation atlas. Word frequency (log-transformed), length, concreteness, and orthographic distance (OLD20) measures were also obtained for each word. Multiple regressions with RF as the dependent variable were then performed for each ROI using word frequency, length, concreteness, and orthographic neighborhood as covariates of interest, and corrected for multiple comparisons using FDR (q = 0.05). Additional whole-brain analyses were also conducted (voxel thresholded at p < .005, cluster corrected to p < .05). ROI analyses revealed negative associations between word frequency and RF in parcels corresponding to the inferior frontal gyrus (IFG), supramarginal gyrus (SMG), the ventral occipito-temporal cortex (vOTC), and posterior portions of the middle temporal gyrus (pMTG). Additionally, there were negative associations between concreteness and RF in parcels corresponding to the angular gyrus, anterior temporal lobe, posterior cingulate cortex, and pMTG. That is, words with lower frequency or concreteness values correspond to higher RF within regions associated with language and semantic processing. In contrast, there was a positive association between orthographic distance and RF in parcels corresponding to the inferior frontal and supramarginal gyri, vOTC, and pMTG. That is, words with higher orthographic distance had higher RF within language regions. Whole-brain analyses revealed similar results. This suggests that words with a sparser orthographic neighborhood, lower concreteness, lower frequency, and longer words are associated with higher representational SNR. Overall, this body of work provides a novel approach for examining the neural components of the reading system based on the fidelity of neural patterns.
Topic Areas: Reading, Language Production