Does Kabashima’s Law Hold in Contemporary Japanese?

A Statistical Validation Using Large-Scale Corpora

Abstract

This study aimed to clarify the quantitative lexical structure of contemporary Japanese by empirically examining the validity of Kabashima’s Law, which posits that the ratio of nouns explains the proportions of other parts of speech and that these proportions correspond to text genres. To this end, part-of-speech proportions were calculated following Kabashima’s framework, based on data collected from diverse text genres in large-scale corpora, and simple linear regression analyses were conducted. The results showed that, although the overall shapes of the graphs suggest that the proportions of other parts of speech can be predicted from the ratio of nouns, the comparisons of residuals revealed that predictions are inaccurate when the noun ratio is low, indicating that the original predictive formulas require coefficient revision. Therefore, this study proposed new regression models derived from corpus data. Furthermore, although genre distinctions were not visually apparent in scatterplots, statistical testing demonstrated that significant differences exist among genres in terms of part-of-speech proportions.