Generative causal testing to bridge data-driven models and scientific theories in language neuroscience

The article introduces generative causal testing (GCT), a novel framework that utilizes large language models (LLMs) to generate and test explanations for BOLD fMRI responses to language stimuli. GCT effectively elucidates language selectivity in both individual voxels and cortical regions, including newly identified microROIs in the prefrontal cortex, and highlights the correlation between explanatory accuracy and the predictive stability of the underlying models. This approach is significant for practitioners as it enhances the interpretability of LLMs in neuroscience, potentially bridging the gap between empirical data and theoretical frameworks.