Brain Region Analysis — Humor Classification

This is the strongest humor-predictive region, and it aligns directly with published fMRI studies. The superior frontal gyrus contains part of the medial prefrontal cortex (mPFC), which is one of the most consistently activated regions in humor processing studies. Chan et al. (2013) showed that the left superior frontal gyrus activates during humor comprehension, while the ventromedial prefrontal cortex activates during the "elaboration" stage — experiencing amusement after the joke is understood. Bekinschtein et al. (2011) further showed that jokes activate reward circuits (amygdala, ventral striatum) whose activity correlates with subjective funniness ratings. The mPFC's involvement likely reflects both cognitive processing of incongruity and reward signaling.

The STS is the brain's hub for social perception and social cognition (Deen et al., 2015). It processes biological motion, social interaction, speech prosody, and — critically — the understanding of others' intentions. Much verbal humor depends on theory of mind (understanding the speaker's intended meaning vs. the literal meaning), which is a core STS function. The STS is also involved in pragmatic language processing, which is essential for detecting the incongruity that makes jokes funny.

The precuneus is implicated in perspective-taking and self-referential processing (Cavanna & Trimble, 2006). Nakamura et al. (2018) found that precuneus volume correlates with self-enhancing humor style, though notably only in individuals with high cognitive empathy — suggesting the link is mediated by social cognition rather than humor per se. The superior parietal lobule supports attentional reorienting, relevant for the surprise/incongruity detection phase of humor. The dual-path model of humor processing (Cheng et al., 2017) specifically implicates the precuneus in the resolution of resolvable incongruity.

Part of the temporoparietal junction (TPJ), the supramarginal gyrus is a core theory-of-mind region. It's activated when understanding others' beliefs and intentions, which is exactly what verbal humor requires — the listener must infer the speaker's intended meaning behind the joke setup. The right TPJ in particular is one of the most replicated regions in false-belief tasks and social cognition studies.

These motor and somatosensory regions are less expected for humor. One possible explanation is embodied simulation — humor often involves physical scenarios (slapstick, bodily references) that may engage motor imagery. Another possibility is that TRIBE v2's text-to-speech pipeline engages articulatory motor regions differently for humorous vs. neutral content (e.g., different prosody or speech rhythm).

The MFG contains the dorsolateral prefrontal cortex (DLPFC), the brain's executive control center. It is engaged during analytical reasoning, working memory, and controlled attention (Curtis & D'Esposito, 2003). For neutral factual sentences, more controlled analytical processing would be expected compared to the more automatic, insight-like processing of humor. The DLPFC also supports ambiguity resolution and discourse management (Hertrich et al., 2021).

The AG is a multimodal convergence zone that binds semantic features (Seghier, 2013). It overlaps with the default mode network and is activated during semantic matching, factual knowledge retrieval, and controlled semantic access (Binder et al., 2009). Neutral factual sentences likely engage this semantic retrieval system more than humor, which bypasses conventional semantic processing through incongruity.

Early visual cortex. Research shows that visual cortex activates during concrete language processing through mental imagery. Neutral sentences in our stimuli tend to describe concrete, factual scenarios ("Water boils at 100 degrees"), which may evoke more vivid visual imagery than humor stimuli, which rely on abstract incongruity.

Contains the visual word form area and supports semantic processing at the word level (Binder et al., 2009). More straightforward lexical-semantic processing for neutral content would activate this region more strongly.