A computational analysis of crosslinguistic regularity in semantic change

Semantic change is attested commonly in the historical development of lexicons across world's languages. Extensive research has sought to characterize regularity semantic change, but existing studies have typically relied on manual approaches or analysis a restricted set We present large-scale computational explore regular patterns word meaning shared many focus two levels analysis: (1) directionality, which we by inferring direction between source and target meaning; (2) source-target mapping, given meaning. work with DatSemShift, largest public database that records thousands changes from over hundreds For directionality inference, find concreteness explains more than 70% cases strongest predictor among alternatives including frequency valence. parallelogram-style analogy model based contextual embeddings predicts mappings substantially better chance similarity-based models. Clustering pairs reveals shiftings domains, such as body parts geological formations. Our study provides an automated approach evidence for multifaceted