Mentalizing regions represent distributed, continuous, and abstract dimensions of others' beliefs

Abstract

The human capacity to reason about others' minds includes making causal inferences about intentions, beliefs, values, and goals. Previous fMRI research has suggested that a network of brain regions, including bilateral temporo-parietal junction (TPJ), superior temporal sulcus (STS), and medial prefrontal-cortex (MPFC), are reliably recruited for mental state reasoning. Here, in two fMRI experiments, we investigate the representational content of these regions. Building on existing computational and neural evidence, we hypothesized that social brain regions contain at least two functionally and spatially distinct components: one that represents information related to others' motivations and values, and another that represents information about others' beliefs and knowledge. Using multi-voxel pattern analysis, we find evidence that motivational versus epistemic features are independently represented by theory of mind (ToM) regions: RTPJ contains information about the justification of the belief, bilateral TPJ represents the modality of the source of knowledge, and VMPFC represents the valence of the resulting emotion. These representations are found only in regions implicated in social cognition and predict behavioral responses at the level of single items. We argue that cortical regions implicated in mental state inference contain complementary, but distinct, representations of epistemic and motivational features of others' beliefs, and that, mirroring the processes observed in sensory systems, social stimuli are represented in distinct and distributed formats across the human brain.

Keywords: Multi-voxel pattern analysis (MVPA); Theory of mind; fMRI.

Fig. 1

Example stimuli from Exp. 1. In the fMRI task, participants listened to audio recordings of 40 target stories and 10 physical control stories. Stories lasted from 22 to 36 s. Individual participants saw each target story in one of the four epistemic conditions, forming a matched and counterbalanced design; every target story occurred in all four conditions across participants. After each story, participants heard the question “Happy or sad?” and indicated whether the main character in the story felt happy or sad using a button press (left/right).

Fig. 2

(a) ROIs (b) Classification accuracy in Exp. 1 for Mental: mental vs. physical stories (see SM), Modality: visual vs. auditory modality (S,H), Justification: strong vs. weak evidence (SS, SW), Directness: first person vs. hearsay (HF, HS), and Valence: positive vs. negative emotion, (c) Classification accuracy in the replication study for Justification: strong vs. weak evidence. (d) LTPJ: density plot of classification accuracies by condition. Modality is not a continuous feature and thus not compared to behavior; rather we observe striking similarity in the classification accuracy in LTPJ between seeing conditions (red: SS and SW) and hearing conditions (purple: HF HS). (e) RTPJ, RSTS, VMPFC: Correlation between item-wise classification scores and behavioral ratings. In RTPJ and RMSTS there were significant correlations between justification classification scores (how many times an item was scored as “strong”, across participants) and behavioral judgments (how good is the character’s evidence?). In VMPFC, there was a significant correlation between valence classification score (how many times an item was scores as “positive”) and behavioral judgments (how happy does the character feel?). (*p < 0.05).

Fig. 3

Example stimuli from Exp. 2. Here, beliefs were unjustified not because of obscured and ambiguous evidence as in Exp. 1, but because of missing or misleading evidence (see SM). After each story, participants responded to the question “How morally blameworthy is [the agent] for [performing the action]?” on a 4-point scale (1-not at all, 4-very much), using a button press.