Abstract
The presenter has previously introduced a method for analyzing waka anthologies in Japanese classical literature using sentence vectors generated by an embedding model. This time, they propose an advanced approach that treats the entire set of sentence vectors in each anthology as a high-dimensional structure, offering a new way to describe it.As source material, they use the Hachidaishū (the eight imperial anthologies from Kokin Wakashū to Shin Kokin Wakashū), converting all included poems into vectors using an embedding model. Their method involves: Placing all waka sentence vectors in the same space and observing the configuration of their centroid positions. Measuring changes in Euclidean distances between vectors in the high-dimensional space. Conducting principal component analysis (PCA) on the difference vectors between the centroids of each anthology’s sentence vectors to identify the primary dimensions of change. By comprehensively applying these techniques, they have been able to capture significant and characteristic changes in the history of waka anthologies as shifts in vector space. Furthermore, this mathematical representation of literary history opens up new possibilities for AI-driven literary studies.
