A Tentative Study on How to Calculate the Proportions of Parts of Speech in Japanese Poetry Collections

An Emphasis on the Sampling Methods for Quantitative Lexicology

Abstract

There has been much research on the proportions of parts of speech in the field of quantitative lexicology. However, the issue of how much variation can occur when applying different calculation methods has not been considered, even though the proportions of parts of speech depend on them. Using waka poetry data, this study empirically examined the differences in proportion for each calculation method in terms of three issues: word unit, variant texts, and statistical sampling. The examination showed that the proportions determined by two different word units had considerable differences, while the proportions calculated using two variant texts of the same waka collection did not change substantially. Moreover, compared to the proportions in the complete survey, the proportions in the sample survey identified statistical errors, but the differences did not change the overall conclusion. Furthermore, when the proportions for the entire waka collection were estimated from the sample proportions, it was revealed that the sampling pieces of waka using cluster sampling yielded unexpectedly better results in terms of the statistical precision than the sampling words using simple random sampling.