Mémotions: A Falsifiable Unit of Affect-Semantic Coupling in Biological and Artificial Neural Systems

Abstract This article advances the concept of mémotion as a testable unit of affect-semantic coupling across biological and artificial neural systems. In contrast to content-only engram models and affect-only accounts, the mémotion is defined as a unified trace linking semantic structure, affective charge, directional salience, and variable inertia. The paper first situates this hypothesis within the biological literature on engrams, arguing that memory consolidation may involve not only encoded content but also action-relevant affective orientation. It then examines sparse autoencoder features in large language models as possible computational proxies for mémotions, provided they satisfy four criteria: trans-contextual stability, directional coherence, paraphrase invariance, and causal sensitivity. A further hypothesis connects mémotion-active features to motivationally aligned attention gradients, thereby linking latent structure and telic orientation. The proposal does not attribute phenomenology to artificial systems; it formulates a functional and falsifiable bridge model between biological memory research and mechanistic interpretability. Related OSF preregistration OSF preregistration: 10.17605/OSF.IO/DPXJ3. Note on preregistration: P-M4 was added on 25 March 2026, subsequent to the initial OSF preregistration, following theoretical analysis of the relationship between cortical excitation/inhibition co-activation and the Fc ⊗ Fi orthogonality postulate. The amendment has been deposited at 10.17605/OSF.IO/DPXJ3. No empirical data were available at the time of addition. Scope This Zenodo record contains the current preprint version of the article.