Abstract:

Cognitive control is a dynamic regulatory mechanism that enables individuals to achieve goal-directed behaviors. Although traditional research has largely emphasized comparisons between discrete, isolated control states (e.g., high versus low conflict in the Stroop task), real-world control-driven behaviors are inherently sequential. They require the preservation of goal consistency, flexible updating across multiple steps, and integration over extended temporal windows. To address this gap, this review systematically investigates behavioral, computational, and neural evidence to pioneer a shift from discrete-state comparisons to a sequential processing framework. We highlight a profound convergence between studies of cognitive control and serial memory, proposing that the sequential nature of cognitive control represents a critical interdisciplinary frontier.
A core contribution of this review is that it outlines an evolving trend in the literature toward studying long sequence of abstract control. Traditionally, research on cognitive control has focused on short-sequence effects, such as congruency sequence effects, post-error slowing, and switch costs. These effects are primarily driven by local, reactive adjustments such as conflict monitoring, associative learning, and the active inhibition of recently executed or irrelevant task sets (e.g., the N-2 repetition cost). However, a growing body of work has turned the interest to the long-term effects of cognitive control and the long control sequences. Executing long sequences requires the establishment of long-term higher-order representations, supported by proactive mechanisms including hierarchical chunking, temporal predictive coding, and abstract sequence compression. Concurrently, in contrast to the more concrete sequences (e.g., word or number sequences) emphasized in the past, the literature on sequence memory has also shifted its interest toward abstract sequences (e.g., rule series) and sequence abstraction (e.g., learning from concrete sensory inputs to construct abstract relational transition graphs). The convergence of these two fields highlights a promising interdisciplinary area: the study of cognitive control sequences.
Critically, the seemingly distinct fields of cognitive control and sequence memory exhibit substantial cross-talk. For instance, both domains share an interest in suppression effects (i.e., suppressing completed items/tasks to avoid interference) and commonly employ sequence working memory paradigms to investigate their core questions. Moreover, their representative neural regions demonstrate strong functional interaction: (1) The prefrontal-hippocampal loop binds transient control states into episodic, context-rich sequence representations. During offline or resting states, the hippocampus generates rapid sequence replay that drives prefrontal activation, supporting memory consolidation and structural generalization. (2) The cortico-basal ganglia-thalamic circuits translate abstract rules into executable action flows, with the striatum forming sequence chunk boundaries, the globus pallidus regulating execution speed, and the thalamus gating the execution of learned subroutines back to the prefrontal cortex. Together, these findings indicate that the integration of cognitive control and sequence memory is functionally plausible. To formalize this theoretical integration, we explicitly introduce the concept of the Representation of Cognitive Control Sequence, defined as the brain's memory representation of structured sequences composed of multiple abstract cognitive control states arranged in a specific temporal order.
Methodologically, we advocate for paradigm innovations such as the Abstract Cognitive Task Sequence (ACTS) and conflict sequence designs, combined with representational geometry analyses, to quantify the abstract dynamics of these control states. Furthermore, the translational implications of this sequential perspective are substantial. In clinical psychopathology, re-evaluating psychiatric disorders through a sequence-level lens offers novel diagnostic and interventional pathways. In artificial intelligence, our framework provides biologically grounded constraints (such as hierarchical chunking and replay-based sequence consolidation) that address the "goal-maintenance" deficit in current models, offering a neurobiologically plausible blueprint for autonomous agents capable of adaptive, multi-step planning.
In conclusion, by establishing cognitive control sequence as a core construct, we bridge the historical divide between cognitive control and sequence memory research, yielding transformative insights for both clinical intervention and brain-inspired computing.

Key words: cognitive control, sequential processing, representation, abstract task, memory