ScholarWorks@성균관대학교

초록

This study analyzes the polysemy of the English verb break by integrating corpus-based analysis with the contextualized language model BERT. Using 620 annotated tokens extracted from COCA, we examine the syntactic and semantic distribution of break senses and semantic–pragmatic factors that influence the sense distribution. PCA visualizations and statistical tests show that across break’s polysemous senses, causer identifiabilityemerged as the strongest predictor of sense distribution, with causer intentionality also playing a significant—though somewhat weaker—role. Integrating contextualized neural embeddings with manually annotated features further reveals previously unobserved break’s usages and polysemy patterns. Methodologically, this study applies an extended distributional semantics framework combining verb meaning, syntax, and argument structure in BERT, demonstrating the potential of contextualized language models for fine-grained analysis of verb polysemy.

키워드