Model-based semantic distance reveals adaptive coordination of distinct cognitive systems in flexible knowledge retrieval

Flexible cognition requires the adaptive retrieval of conceptual knowledge, spanning a continuum from proximal to distal semantic associations. However, the neural dynamics that facilitate this flexibility remain poorly understood. Here, combining computational linguistics with functional magnetic resonance imaging (fMRI) and machine learning methods, we derive a whole-brain signature that captures graded variations in semantic distance. This domain-specific neural model revealed three distinct large-scale cognitive systems whose interactions coordinate semantic retrieval: a left-lateralised frontotemporal language and a bilateral frontoparietal control network, both recruited for distant associations, and a medial default mode memory network, facilitating access to proximal relations. Importantly, we show that adaptive retrieval across the continuum of semantic distance is facilitated by a dynamic coordination mechanism. As semantic distance increases, representational patterns converge across the three cognitive systems. These findings provide a unifying model of the neural architecture underlying semantic processing, revealing how dynamic interactions between competing cognitive systems enable flexible knowledge retrieval. Combining AI language models with brain imaging, researchers identify a neural signature that tracks how easily the mind retrieves related concepts, revealing three brain networks that cooperate more closely as retrieval becomes more demanding.