Extracting speaker-specific functional expressions from political speeches using random forests in order to investigate speakers’ individual political styles

In this study we extracted speaker-specific functional expressions from political speeches using random forests in order to investigate speakers’ individual political styles. Along with methodological development, stylistics has expanded its scope into new areas of application such as authorship profiling and sentiment analysis in addition to conventional areas such as authorship attribution and genre-based text classification. Among these, computational sociolinguistics, which aims at providing a systematic and solid basis for sociolinguistic analysis using machine learning and linguistically-motivated features, is a potentially important area. In this study we show the effectiveness of the random forests classifier for such tasks by applying it to Japanese prime ministers’ Diet speeches. The results demonstrate that our method successfully extracted the speaker-specific expressions of two Japanese prime ministers, and enabled us to investigate their individual political styles in a systematic manner. The method can be applied to sociolinguistic analysis of various other types of texts, and in this way, this study will contribute to developing the area of computational sociolinguistics in the field of stylistics.