Beyond what was said: Neural computations underlying pragmatic reasoning in referential communication

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Linguistics

Background:

• Pragmatic inference is crucial for communication, enabling understanding of intended meaning beyond literal words.
• Neurocomputational mechanisms underlying pragmatic inference and individual variability remain largely unknown.

Purpose of the Study:

• To investigate the neurocomputational basis of pragmatic inference using a reference game.
• To determine if individual-level or population-level models better predict pragmatic reasoning performance.
• To identify brain regions and functional connectivity patterns associated with Bayesian belief updating in pragmatic inference.

Main Methods:

• Employed a reference game paradigm with functional magnetic resonance imaging (fMRI).
• Utilized model-based fMRI to correlate brain activity with Bayesian inference computations.
• Assessed individual differences in higher-order reasoning abilities.

Main Results:

• An individual-level pragmatic inference model significantly outperformed a population-level model in predicting listener performance.
• Bayesian posterior probability correlated positively with activity in the ventromedial prefrontal cortex (vmPFC) and ventral striatum.
• Negative correlations were observed in dorsomedial PFC, anterior insula (AI), and inferior frontal gyrus (IFG); higher-order reasoning correlated with IFG/AI activation and vmPFC-AI connectivity.

Conclusions:

• The study provides a preliminary neurocomputational model for how the brain represents Bayesian belief inferences during pragmatic reasoning.
• Individual differences in higher-order reasoning influence pragmatic inference through specific neural pathways, particularly involving the vmPFC and AI.
• Findings highlight the neural basis of heterogeneity in pragmatic inference capabilities.