Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Characterizing Empathy Processing in Autism: The Intertwined Roles of Language and Social Brain Networks
Poster Session B, Wednesday, September 30, 4:30 - 6:30 pm, Wangari Maathai
Brea Chouinard1, James McKinnon1, Cory Efird1, Alona Fyshe; 1University of Alberta
Historically, the medical model sought empathy “deficits” in autism, ignoring that story-based empathy tasks required the dual engagement of language and social brain networks. In contrast, a neurodiversity-affirming lens instead investigates how the role of both networks during empathy processing. To advance beyond historical limitations, we must explicitly account for the role of language when evaluating social cognition in autistic and non-autistic adults. We investigated how well the semantic meaning of different stories could be traced in the fMRI data of 15 autistic and 15 non-autistic adults with intact language and reading abilities. We used word embeddings derived from a large language model to create single semantic representations for each TR, which were then used as features in encoding analyses. In an encoding analysis, we predict what the neuroimaging data of left out TRs should look like after training the model on the other TRs. Participants read four different stories via rapid serial visual presentation: an empathy-evoking story; a grocery store story with the potential to be highly anxiety-inducing for autistic individuals (‘anxiety’); a grocery store story intended to be calming to autistic individuals (‘calming’); and a control story. All stories were equivalent for factors known to influence readability, such as word frequency and difficulty. Notably, the empathy story had more emotional content than the other stories, as measured using a validated, National Research Council database; and the anxiety story was more likely to lead to anxiety in an autistic person than other stories, as judged by a paid autistic advisory panel. We trained/tested within each story and obtained encoding performance scores by finding the correlation between the predicted and actual fMRI data. We evaluated encoding performance separately in Neurosynth-defined language and social brain networks. We hypothesized no group differences in encoding performance for the empathy and control stories, but that encoding performance would differ between the groups for the anxiety and calming stories. There were no group differences for the empathy story in either network (ps > .50). However, for the control story, there was stronger encoding performance in autistic than non-autistic adults in the language network (p = .025) and no difference in the social network (p = .50). This unexpected finding might reflect the fact that the control story was written by the autistic advisors and thus contained structure or framing that more closely resembled autistic language processing. For the anxiety and calming stories, there was no group difference in either network for the calming story (ps > .420), and greater encoding performance for non-autistic compared to autistic in the language (p = .004) but not social (p = .250) networks for the anxiety story. This may suggest that autistic groups were using internalized language to moderate internal states (e.g., navigate anxiety), competing for linguistic processing resources. Equivalent social network engagement refutes the “empathy deficit” stereotype. Instead, task differences lie in the language network, revealing that autistic narrative tracking is a dynamic process shaped by authorship alignment and internal emotional regulation.
Topic Areas: Meaning: Discourse and Pragmatics, Disorders: Developmental