The usage and meaning of words may change over time. In this study, we compared two existing BERT-based methods of capturing semantic changes in Japanese words. Each method clusters word tokens obtained from BERT in different ways. One method uses example sentences from a dictionary, whereas the other method uses centroids from k-means clustering are used. We calculated the usage ratio for each period (25 years) according to the clustering results. The clustering-based method outperformed the dictionary-based method in almost all the cases. We also found that the dictionary-based method was sensitive to definitions and the specific choices of example sentences.
CITATION STYLE
Kobayashi, K., Aida, T., Oka, T., & Komachi, M. (2023). Analysis of Semantic Changes in Japanese Words Using BERT. Journal of Natural Language Processing, 30(2), 713–747. https://doi.org/10.5715/jnlp.30.713
Mendeley helps you to discover research relevant for your work.