Past work has demonstrated a link between semantic memory and verbal creativity. Yet, few studies have considered this relationship through the lens of the controlled semantic cognition account, which anticipates that multimodal concepts in long-term memory interact with semantic control processes to generate goal and context-appropriate patterns of retrieval. In particular, while the creativity literature has distinguished divergent and convergent aspects of creativity, little is known about their relationship with separable aspects of semantic control, or the semantic intrinsic functional architecture of the brain. We investigated whether tasks with greater reliance on controlled semantic retrieval (assessed through weak association) versus semantic selection (assessed through semantic feature matching) were differentially linked to divergent creativity (assessed with the unusual uses task; UUT) and convergent creativity (assessed with the remote associates task; RAT). Better performance on the RAT was linked to semantic selection, while stronger performance on UUT was linked to more efficient retrieval of weak associations. We also examined individual differences in the intrinsic functional architecture of the semantic system using resting-state fMRI. Greater coupling between the anterior temporal lobe (multimodal semantic store) and left inferior frontal gyrus (LIFG) (in the semantic control network) was linked to stronger convergent creativity. This pathway also correlated with semantic feature matching performance, but not the retrieval of weak associations. In contrast, better divergent creativity was linked to greater coupling between LIFG and language-related auditory-motor regions, and decoupling from the default mode and frontoparietal networks. These connections correlated with the retrieval of weak associations. Interestingly, while decoupling of LIFG with default mode and frontoparietal networks correlated with the retrieval of weak associations, coupling of LIFG with these networks correlated with semantic feature matching. These behavioural and neurocognitive dissociations show that semantic control and creativity are highly related yet multifaceted constructs that depend on the underlying intrinsic architecture of key sites related to semantic cognition.

Control processes are critical for the context-appropriate use of meaningful stimuli. Similar definitions have been adopted in two distinct literatures focusing on identifying the neural correlates of “semantic control” and of executive control across domains (the “multiple demand network”). Surprisingly, despite their proposed functions varying only in relation to domain-specificity, these networks appear to differ anatomically. However, prior comparisons are confounded by variations in task design. To what extent might varying task requirements drive differences in activation patterns that are typically attributed to stimulus domain? Here, for the first time, we use functional MRI to disentangle the effects of task process and stimulus domain during cognitively demanding tasks. Participants performed an odd-one-out task requiring rule-switching, inhibition and selection processes, and an n -back working memory task, each with meaningful semantic and non-semantic stimuli, in a factorial design. Both stimulus domain and task process affected the control regions activated, indicating that task process is indeed a key factor confounding prior studies. However, core semantic control regions (left inferior frontal gyrus, left posterior temporal cortex) also showed a preference for semantic stimuli even with matched task processes, while more peripheral semantic control regions, overlapping the multiple demand network (dorsomedial prefrontal cortex, right inferior frontal gyrus), showed little preference across task or stimulus. Conversely, most multiple demand network regions were preferentially engaged for non-semantic stimuli. These results highlight the mutual importance of stimulus domain and task process in driving variation in control region engagement, both across and between semantic control and multiple demand networks.