Cross-linguistic semantic stability as a driving force of diachronic semantic evolution across lexica

Word meanings change over time, yet the extent of change varies systematically within each lexicon. Although frequency, age of acquisition, and concreteness have been identified as within-language predictors, a language-general account of why some concepts show greater resistance to diachronic semantic change across languages remains elusive. Here, we propose that cross-linguistic semantic stability - the degree to which a concept's semantic neighborhood is conserved across languages - as a concept-level predictor of diachronic semantic change. We operationalized stability as neighborhood overlap between a target concept's nearest neighbors in English and the back-translated neighbors of its translation equivalents across 10 to 41 languages, and quantified diachronic change as neighborhood turnover in historical embeddings spanning up to 100 years in English, French, German, and Chinese. Higher cross-linguistic stability predicted less diachronic semantic change in each language and in an aggregated analysis of concepts shared across the four languages. This effect dominated all other predictors for shared concepts; remained significant after established psycholinguistic predictors were controlled; was represented as a direct conditional edge in partial-correlation networks; remained robust as the cross-linguistically shared concept set was expanded; and was replicated using historically derived stability estimates. Together, these findings reveal the topological consistency of a concept's semantic neighborhood across languages is a temporally durable, concept-level predictor of resistance to semantic change, linking synchronic cross-linguistic conceptual structure with diachronic lexical evolution across lexica.