BERT-assisted behavioral profiling of polysemy: contrastive analysis of HONG in Chinese and RED in English

Abstract This study conducts a corpus-based contrastive analysis of the basic color term “red” in Chinese (HONG) and English (RED), innovatively employing the AI model BERT to triangulate, validate, and enrich the findings of the behavioral profile (BP) analysis. Results illustrate HONG’s more variational semasiological structure than RED, with both exhibiting unique usage patterns among similarities. The triangulation with BERT embeddings visualizes the performance of each subjectively annotated BP variable, and further demonstrates the semantic extending process in both HONG and RED. The underlying cognitive and socio-cultural factors are lastly discussed. Theoretically, this study bridges frequency-based AI models and usage-based Cognitive Linguistics by simulating human cognitive processes to abstract linguistic knowledge from extensive data, advancing our understanding of language through AI technology. Methodologically, this study is the first to integrate AI models in assisting BP analysis, with BERT embeddings offering objective corroborations and deeper insights for conclusions drawn from subjectively labeled BP data.