Bayesian Linear Mixed Model に基づく漢字親密度推定の試みー言語資源としての統計モデルの検討

要旨

This study reports on a method for estimating kanji familiarity based on data constructed for use in a course at Tokyo University of Foreign Studies. The course aimed to teach empirical research methods in linguistics, including experimental design and statistical analysis techniques. As exercise data for statistical analysis, evaluation information was collected for 2,136 commonly used kanji characters through Yahoo! Crowdsourcing. For each kanji, 50 participants (a total of 1,258 respondents) provided self-assessments on fi ve aspects: "knowing the character," "being able to write it," "knowing its On'yomi (Chinese reading)," "knowing its Kun'yomi (Japanese reading)," and "knowing its radical." Based on these evaluations, a Bayesian linear mixed model was applied to experimentally estimate kanji familiarity.

主題