Tracing Lexical Semantic Change with Distributional Semantics: Change and Stability

Abstract

Recent studies suggest an increasing interest in detecting lexical semantic changes in the context of distributional semantics. However, most proposals have been implemented with English datasets but not much with Chinese data. This paper thus presents an exploratory study using the popular Skip-gram models and post-processing operations to obtain historical word embeddings, testing whether methods in fashion could capture lexical semantic change in Chinese historical texts. Our results demonstrate a positive answer to this question by suggesting interesting cases which may have undergone the process of meaning generalization and shown competence among homographs. Additionally, our analysis also indicates that social contexts play an important role in lexical semantic change.