Answers are grounded in this paper and cite their sources.
Multisensory integration is traditionally conceptualized as transient, stimulus-driven feature binding. Here, we propose a hierarchical framework that reconceptualizes multisensory processing as a continuum of representational emergence across three distinct levels: level I (multimodal binding), where features are integrated via probabilistic causal inference; level II (supramodal representation), where recurrent co-activation and value-guided selection establish consolidated, perceptual-semantic engrams that act as emergent modalities; and level III (amodal concepts), where meaning becomes increasingly detached from sensory form and is represented within transmodal semantic networks. Drawing on recent mammalian evidence for cross-modal associative memory neurons and experience-dependent network reorganization, this framework proposes how local synaptic selection may scale into increasingly autonomous cognitive modules. Finally, we delineate an empirical roadmap and outline implications for supramodal AI, toward architectures capable of more autonomous semantic representation and generation.
No takes yet. Share an insight, caveat, or question.
Paraskevopoulos et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: