Responses to illness inferences in the precuneus (PC). Panel A: Percent signal change (PSC) for each condition among the top 5% Illness-Causal > Mechanical-Causal vertices in a left PC search space (Dufour et al., 2013) in individual participants, established via a leave-one-run-out analysis. Panel B: Average PSC in the critical window (marked by dotted lines in Panel A) across participants. The horizontal line within each boxplot indicates the overall mean. Panel C: Whole-cortex results (one-tailed) for Illness-Causal > Mechanical-Causal and Illness-Causal > Noncausal (both versions of noncausal vignettes), corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001. Panel C: Example stimuli. ‘Magical’ catch trials similar in meaning and structure (e.g., “Sadie forgot to wash her face after she ran in the heat. Now she has a cucumber nose”) enabled the use of a semantic ‘magic detection’ task.

Spatial dissociation between univariate responses to illness inferences and mental state inferences in the precuneus (PC). The left medial surface of 6 individual participants were selected for visualization purposes. The locations of the top 10% most responsive vertices to Illness-Causal > Mechanical-Causal in a PC search space (Dufour et al., 2013) are shown in red. The locations of the top 10% most responsive vertices to mentalizing stories > physical stories (mentalizing localizer) in the same PC search space are shown in blue. Overlapping vertices are shown in green.

Individual-subject analysis of language- and logic-responsive vertices. Panel A: percent signal change (PSC) for each condition among the top 5% most language-responsive vertices (language > math) in a temporal language network search space (Fedorenko et al., 2010). Results from a frontal language search space (Fedorenko et al., 2010) can be found in Supplementary Figure 8. Panel B: PSC among the top 5% most logic-responsive vertices (logic > language) in a logic network search space (Liu et al., 2020). Group maps for each contrast of interest (one-tailed) are corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001. Boxplots display average PSC in the critical window (marked by dotted lines) across participants. The horizontal line within each boxplot indicates the overall mean.

Responses to mechanical inferences in anterior parahippocampal regions (anterior PPA). Panel A: Percent signal change (PSC) for each condition among the top 5% Mechanical-Causal > Illness-Causal vertices in a left anterior PPA search space (Hauptman, Elli, et al., 2025) in individual participants, established via a leave-one-run-out analysis. Panel B: Average PSC in the critical window (marked by dotted lines in Panel A) across participants. The horizontal line within each boxplot indicates the overall mean. Panel C: The intersection of two whole-cortex contrasts (one-tailed), Mechanical-Causal > Illness-Causal and Mechanical-Causal > Noncausal that are corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001. Similar to PC responses to illness inferences, anterior PPA is the only region to emerge across both mechanical inference contrasts. The average PPA location from separate study involving perceptual place stimuli (Weiner et al., 2018) is overlaid in black. The average PPA location from separate study involving verbal place stimuli (Hauptman, Elli, et al., 2025) is overlaid in blue.

Functional localization of language, logical reasoning, and mentalizing networks (see Monti et al., 2009; Fedorenko et al., 2010; Dodell-Feder et al., 2011; Liu et al., 2020). Group maps for each contrast of interest (one-tailed) are corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001.

Overlap between left precuneus (PC) responses to illness inferences in the current study and people-related stimuli in a separate study (Fairhall & Caramazza, 2013b). The average location from a separate study comparing people and place concepts (Fairhall & Caramazza, 2013b) is overlaid in blue on the response to illness inferences observed in the current study. The group map (one-tailed) is corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001.

Spatial dissociation between responses to illness inferences and mental state inferences in left precuneus (PC). The left medial surface of all participants (n=20) is shown. The locations of the top 10% most responsive vertices to Illness-Causal > Mechanical-Causal in a PC search space (Dufour et al., 2013) are shown in red. The locations of the top 10% most responsive vertices to mentalizing stories > physical stories (mentalizing localizer) in the same PC search space are shown in blue. Overlapping vertices are shown in green.

Full whole-cortex univariate results. Regions whose activity scales with response time (RT) are displayed under “RT.” Frontal RT regions are outlined in white on the lateral surface for other contrasts where frontal effects are observed. Group maps (two-tailed) are corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001.

Comparison of whole-cortex results for number of people in each vignette (left) and illness inferences (right) from the same GLM. Group maps (two-tailed) are corrected for multiple comparisons (p < .05 FWER, cluster-forming threshold p < .01 uncorrected). Vertices are color coded on a scale from p=0.01 to p=0.00001.

Group overlap in univariate contrasts comparing causal (Illness-Causal, Mechanical-Causal) and noncausal conditions (Noncausal-Illness First + Noncausal-Mechanical First) in the PC. Each vertex in a PC search space (Dufour et al., 2013) was color-coded according to the proportion of participants who showed significant activation (p < .05 uncorrected) at that location.

Group overlap in univariate contrasts comparing causal (Illness-Causal, Mechanical-Causal) and noncausal conditions (Noncausal-Illness First + Noncausal-Mechanical First) in the PC, winner-take-all approach. Each vertex in a PC search space (Dufour et al., 2013) was color-coded according to the proportion of participants who showed a preference for Illness-Causal > Noncausal compared to Mechanical-Causal > Noncausal (red) and vice versa (blue) at that location.

Responses to illness inferences in the fusiform face area (FFA). Percent signal change (PSC) for each condition among the top 5% Illness-Causal > Mechanical-Causal vertices in left and right FFA search spaces (Julian et al., 2012) in individual participants, established via a leave-one-run-out analysis. Average PSC in the critical window (marked by dotted lines in Panel A) across participants is displayed via boxplot. The horizontal line within each boxplot indicates the overall mean.

Searchlight MVPA group maps. Whole-brain searchlight maps were thresholded using a vertex-wise threshold (p < .001 uncorrected) and a cluster size threshold (FWER p < .05, corrected for multiple comparisons across the entire cortical surface). Vertices are color coded on a scale from 55-65% decoding accuracy.

Subject dispersion data for individual-subject MVPA performed in functional ROIs. We tested whether patterns of activity elicited during illness inferences vs. mechanical inferences could be decoded in each fROI: left and right PC, left and right TPJ, language network, logic network. In accordance with our preregistration, 2 types of fROIs were constructed using PC and TPJ search spaces: 1) top 300 most active vertices for mentalizing stories compared to physical stories in the mentalizing/animacy localizer (Mentalizing stories > Physical stories), and 2) top 300 most active vertices for both causal conditions compared to rest (Illness-Causal + Mechanical-Causal > Rest) Chance: 50%. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. Full statistical results are included in Supplementary Table 2.

Subject dispersion data for individual-subject MVPA performed in functional ROIs. 5 tests were performed in each fROI: left PC (LPC), language network, logic network. Chance: 50%. Significance codes: 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. Full statistical results are included in Supplementary Table 3.

Responses to causal inference in the language network. Panel A: Percent signal change (PSC) for each condition among the top 5% most language-responsive vertices (language > math) in a temporal language network search space (Fedorenko et al., 2010). Panel B: The same results in a frontal language search space (Fedorenko et al., 2010). Boxplots display average PSC in the critical window (marked by dotted lines) across participants. The horizontal line within each boxplot indicates the overall mean.

Responses to illness inferences in bilateral PC and TPJ. Percent signal change (PSC) for each condition among the top 5% Illness-Causal > Mechanical-Causal vertices in bilateral PC and TPJ search spaces (Dufour et al., 2013) in individual participants, established via a leave-one-run-out analysis, is shown. We hypothesized that the PC and TPJ would exhibit a preference for illness inferences and report all data for completeness (see preregistration https://osf.io/6pnqg). Significance codes for Illness-Causal > Mechanical-Causal comparison (paired samples t-tests): 0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘’ 1. Subject dispersion data are shown in Supplementary Figure 14.

Subject dispersion data for responses to illness inferences in bilateral PC and TPJ (see Supplementary Figure 13). We hypothesized that the PC and TPJ would exhibit a preference for illness inferences and report all data for completeness (see preregistration https://osf.io/6pnqg). Boxplots display average PSC in the critical window (5-21 s) across participants. The horizontal line within each boxplot indicates the overall mean.

Comparison of mentalizing localizers used in previous work and in the current study, in 3 pilot participants. The mentalizing localizer in the current study used the same mentalizing stories as in previous work (Dodell-Feder et al., 2011) but contained new physical stories that included more vivid physical description and did not refer to animate agents. Individual-subject maps are shown at p < .01 uncorrected.

Illness types present in the stimulus set.

Results of preregistered MVPA for Illness-Causal vs. Mechanical-Causal in individual-subject functional ROIs. Each fROI was created by selecting the top 300 vertices for each contrast (see ‘Contrast’) in each search space. Accuracy refers to classifier performance against chance (50%) for Illness-Causal vs. Mechanical-Causal. Permuted and Bonferroni-corrected (across fROIs) p-values are reported. Ment_vs_phys: mentalizing stories > physical stories (mentalizing localizer). Caus_vs_rest: Illness-Causal + Illness-Mechanical > Rest. Logic_vs_lang: logic > language (language/logic localizer). Lang_vs_math: language > math (language/logic localizer). Visualizations of these results are displayed in Supplementary Figure 10.

MVPA results for all tests in select individual-subject functional ROIs. Each fROI was created by selecting the top 300 vertices for each contrast in each search space: left PC (LPC) = top main experimental conditions > rest, language = top language > math (language/logic localizer), logic = top logic > language (language/logic localizer). Accuracy refers to classifier performance against chance (50%) for each test. Permuted and Bonferroni-corrected (across fROIs) p-values are reported. Visualizations of these results are displayed in Supplementary Figure 11.