Abstract
Whether relationships between word form and affect reflect shared human constraints or language-specific conventions remains largely unresolved. Addressing this, here we tested whether affective mappings learned in one language zero-shot predict estimations for the emotional dimensions of valence and arousal in another using 85,375 lexical entries from ten languages diverse in origin and culture across eight (sub)lexical representations spanning orthographic, phonological, articulatory, and distributional semantic information. Results revealed a strong semantic transfer, greater for valence than arousal, while form-based transfer was smaller but consistently above zero across all seven representations for arousal, with the strongest effects for articulatory features. Language-distance analyses further revealed that valence-related orthographic and phoneme-bigram transfer was stronger between more closely related languages, particularly genealogically and the modest correspondence between fitted form coefficients indicated that transferable prediction coexisted with differences in feature weighting across languages. These findings characterize affective iconicity as graded and representation-dependent: valence generalizes most strongly through distributional meaning, whereas arousal leaves a more consistent cross-linguistic trace in word form. This pattern motivates an account in which culturally and historically shaped evaluative mappings coexist with activation-related mappings that may draw more broadly on shared embodied constraints, connecting linguistic form with the conceptual organization of affective experience.