Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Reduced engagement of Theory of Mind regions during lower-proficiency language comprehension
Poster Session C, Thursday, October 1, 10:45 am - 12:45 pm, Wangari Maathai
Saima Malik-Moraleda1, Megha Vemuri2, Evelina Fedorenko1,2; 1Harvard University, 2MIT
The language and theory of mind (ToM) networks are anatomically adjacent yet functionally distinct systems: the language network selectively tracks linguistic input, whereas the ToM network preferentially tracks mental-state content (Paunov et al., 2022). During narrative comprehension, ToM regions may become engaged when linguistic input contains information about the beliefs, intentions, or emotions of characters (Blank & Fedorenko, 2023; Paunov et al., 2022). In the present study, we examined multilingual speakers (n = 25) using an fMRI language localizer paradigm based on Alice in Wonderland (Malik-Moraleda, Ayyash, et al., 2022), administered in both participants’ dominant and lower-proficiency languages. Participants also completed a standard verbal theory of mind localizer contrasting false-belief and false-photograph stories (Saxe & Kanwisher, 2003). Outside the scanner, a subset of participants (n=18) completed a comprehension task in which they described what they understood of the story in the lower-proficiency language using their dominant language (English in all cases). Comprehension was quantified using multilingual transformer-based embeddings (Rei et al., 2020). Within the language network, both the dominant and lower-proficiency language conditions elicited reliable responses above baseline. Critically, the dominant language elicited stronger neural responses than the lower-proficiency language (dominant: 2.23% BOLD change, SE = 0.173; lower-proficiency: 1.91% BOLD change, SE = 0.143; LME: β = 0.32, p = .05). Similarly, passages presented in the dominant language elicited significantly stronger responses in ToM regions than passages presented in the lower-proficiency language (dominant: 0.31, SE = 0.15; lower-proficiency: 0.11, SE = 0.11; LME: β = 0.27, p < .001). This effect was particularly pronounced in the bilateral temporoparietal junction (TPJ), where dominant-language passages elicited significant above-baseline responses (mean = 0.75, SE = 0.27, p = .013), whereas lower-proficiency language passages did not (mean = 0.20, SE = 0.17, p = .206). TPJ responses were also significantly stronger for the dominant language than for the lower-proficiency language (LME: β = 0.55, p = .003). Furthermore, in the subset of participants with behavioral data, comprehension scores positively correlated with neural responses in both the language network (r = .58, p = .01) and the ToM network (r = .67, p < .01). These findings suggest that successful comprehension of socially rich content in lower-proficiency languages may be constrained by reduced engagement of both language and ToM systems. More broadly, the results point to dynamic interactions between language and ToM networks during multilingual comprehension, with successful engagement of ToM regions potentially depending on the quality of linguistic representations computed by the language system.
Topic Areas: Multilingualism,