Using syntactic features to predict author personality from text

The style in which a text is written re ects an array of meta-information concerning the text (e.g., topic, register, genre) and its author (e.g., gender, region, age, personality). The eld of stylometry addresses these aspects of style. A successful methodology, borrowed from text categorisation research, takes a two-stage approach which (i) achieves automatic selection of features with high predictive value for the categories to be learned, and (ii) uses machine learning algorithms to learn to categorize new documents by using the selected features (Sebastiani, 2002). To allow the selection of linguistic features rather than (n-grams of) terms, robust and accurate text analysis tools are necessary. Recently, language technology has progressed to a state of the art in which the systematic study of the variation of these linguistic properties in texts by different authors, time periods, regiolects, genres, registers, or even genders has become feasible.