Poster B34, Tuesday, August 20, 2019, 3:15 – 5:00 pm, Restaurant Hall
Neural Correlates of Semantic Role Processing in Naturalistic Language Comprehension: an fMRI study
Shulin Zhang1, Jixing Li2, Wen-Ming Luh3, John Hale1;1University of Georgia, 2New York University Abu Dhabi, 3National Institute on Aging
Introduction: A central goal of human language comprehension is determining who-did-what-to-whom in the event described by a sentence. The assignment of semantic roles during comprehension has been extensively studied in both clinical and experimental settings. It has often been suggested that Agent is easier to assign than the Patient (see Bornkessel-Schlesewsky and Schlesewsky, 2016, section 30.4.3 among others). The current study examines this suggestion using neuroimaging during ecologically-natural comprehension of a literary audiobook. Methods: 35 Chinese participants (30 female, mean age=21.3) listened to a Chinese audiobook version of "The Little Prince" for about 100 minutes. BOLD functional scans were acquired using a multi-echo planar imaging (ME-EPI) sequence with online reconstruction (TR=2000 ms; TE's=12.8, 27.5, 43 ms; FA=77; matrix size=72x72; FOV=240.0x240.0 mm; 2 x image acceleration; 33 axial slices, voxel size=3.75x3.75x3.8 mm). Preprocessing was carried out with AFNI and ME-ICA (Kundu et al., 2012). The text of the book has been annotated by Li et al 2016 with Abstract Meaning Representations. We considered 1876 words that were marked as Agents and 1401 marked as Patients. Aligning these role-bearing words in time with their utterance in the audiobook, we defined binary regressors in a GLM analysis that compares brain activity across roles. Following Brennan et al 2012 we also included a phrase-structural node-count regressor to model syntactic processing difficulty. This co-regressor helps rule out a syntactic interpretation for any effects that we might observe in response to semantic roles. Other control regressors include the word rate at the offset of each word, the unigram frequency of each word based on the Google Book ngrams, and RMS intensity at every 10 ms of the audio. A whole brain GLM analysis was carried out with SPM12. The predictors were convolved using SPM's canonical HRF. Results: Agent words give rise to 4 clusters of activation: the bilateral STGs, the right Precuneus and the right frontal lobe. Patient words are correlated with a left-lateralized network including the left MTG, ITG, MFG, IFG, SFG, Parietal Lobe, and the right Cerebellum. The Patient-Agent contrast yields bilateral activation in the Cingulate Gyrus, Precuneus and Insula, and right hemisphere activation in the AG, MFG and SFG. Conclusion: We find that the neural signature for Agents is more focused within the superior temporal lobe than for Patients. This asymmetry in activation could offer a neural basis for the suggestion that Agenthood is easier to recognize than Patienthood. References and a Figure are available at https://bit.ly/2GvkMmk
Themes: Meaning: Combinatorial Semantics, Computational Approaches
Method: Functional Imaging