Classifying and Characterizing Poets for Manyo Tanka via Sparse Canonical Discriminant Analysis

Abstract

Manyo tanka is a short Japanese poem included in Manyoshu which is the oldest collection of Japanese poems. Since the short poems are composed by several poets, each poem has characteristics for each poet. Until now, the characteristics have been subjectively studied or have been investigated based on a single sound. In this paper, we use a statistical method to study the characteristics based on multiple sounds in Manyo tanka. In particular, we analyze the Manyo tanka dataset using sparse canonical discriminant analysis. This analysis uncovers inherent properties of poets for Manyo tanka.