A study titled “Where meanings converge and diverge: Revisiting linguistic relativity across 81 languages” was published on October 24, 2026, on the arXiv preprint server (ID 2610.30778v1), focusing on the central question of linguistic relativity: Do speakers of different languages organize concepts in similar ways? The research did not rely on human‑generated semantic judgments; instead, it employed two multilingual word‑embedding models to quantify the semantic relationship structures among 196 basic concepts across 81 languages.
The analysis reveals systematic differences in the consistency of cross-linguistic conceptual relationships: the degree of correspondence for socio-cultural meanings (such as “family,” “law,” and “religion”) is significantly higher than that for natural categories (such as “water” and “fire”) and motion-related concepts (such as “roll” and “slide”). This contrast holds consistently across both models and multiple conceptual classification schemes. Furthermore, within the socio-cultural domain, interlinguistic differences in conceptual relations grow stronger with increasing geographic distance, and the more distantly related the languages are phylogenetically, the more pronounced these differences become across all semantic domains.
Researchers note that these findings neither support the strong relativism thesis—that language entirely determines thought—nor the absolute universalism thesis—that conceptual structures are wholly universal. Instead, they suggest that universality and relativity coexist in an intertwined manner across different semantic domains. The theoretical explanation emphasizes the ongoing interplay between shared cognitive resources, linguistic specificity, and local experience, and accordingly predicts that humans will exhibit stronger cross-linguistic consistency in semantic judgments of sociocultural concepts than in those of natural and motion concepts. It should be noted that these conclusions are based solely on language-use patterns reflected in word embeddings and do not draw on empirical human‑behavior experiments or neuroscientific evidence.