Presentation

Search Abstracts | Symposia | Slide Sessions | Poster Sessions | Poster Slams

The cortical representation of lexical semantics is shared across English and Chinese

Poster E18 in Poster Session E, Saturday, October 8, 3:15 - 5:00 pm EDT, Millennium Hall
Also presenting in Poster Slam E, Saturday, October 8, 3:00 - 3:15 pm EDT, Regency Ballroom

Catherine Chen^1*, Lily Gong^2*, Christine Tseng², Daniel Klein¹, Jack Gallant², Fatma Deniz^2,3; ¹Department of Electrical Engineering and Computer Sciences, University of California, Berkeley, CA 94720, USA, ²Helen Wills Neuroscience Institute, University of California, Berkeley, CA 94720, USA, ³Institute of Software Engineering and Theoretical Computer Science, Technische Universität Berlin, Berlin, Germany

During language comprehension semantic information from spoken and written English is represented in functional networks that are distributed broadly across the human cerebral cortex (Huth et al. 2016, Deniz et al. 2019, Nakai et al. 2021). However, it is unclear whether these cortical representations are similar or different across languages. To address this question, we compared cortical semantic selectivity between English and Chinese language comprehension. In our study we used functional magnetic resonance imaging (fMRI) to record brain activity from fluent Chinese-English bilingual participants. Each participant read over two hours of natural narratives in both English and Chinese. The same narratives were presented in both languages. For each language, the words of each narrative were presented one-by-one using rapid serial visual presentation (RSVP). Each language was presented in its native script. We then used voxelwise modeling (VM) to estimate the semantic selectivity of each voxel in each language. In the VM framework, features of interest are first extracted from the stimuli and then linearized regression is used to determine how each feature is represented in each voxel (Wu et al. 2006, Naselaris et al. 2011). In this study, stimulus features were constructed by projecting each word onto a 300-dimensional embedding space that reflects the lexical semantics of the word (Bojanowski and Grave et al. 2017, Joulin et al. 2018). A separate voxelwise encoding model was estimated for each voxel, participant, and language. Separate datasets were used for model estimation and evaluation in order to estimate prediction accuracy. Prediction accuracy was quantified by computing the Pearson correlation coefficient (r) between predicted and recorded BOLD responses. The estimated voxelwise encoding models reveal the semantic selectivity of voxels located across the entire cerebral cortex. We found that voxels distributed across temporal, prefrontal, and parietal cortices were well predicted by the estimated encoding models in each language. Moreover, models estimated in one language produced highly accurate predictions of brain responses to the other language. These findings show that the same cortical regions are activated for both languages, and furthermore that within these regions semantic selectivity is shared between English and Chinese. We suggest that in higher-level cortical areas lexical semantic information is encoded largely independently from the stimulus language.

Topic Areas: Meaning: Lexical Semantics, Reading