A Measure of Text Formality as a Human Construct

Formality has long been of interest in the study of language and discourse, and different measures of formality have been developed to predict genre variation. However, it is unclear to what extent these formality metrics are similar to the formality construct perceived by humans. This study first investigated what linguistic features predicted the text formality as humans constructed, then developed a weighted formality model, and finally tested this measure in different approaches. The corpus of this study consisted of 390 excerpts in TASA corpus with three genres: language arts, social studies and science. The five Coh-Metrix dimensions were used to develop the weighted formality model. Results showed the weighted model perceived by humans was constructed by five dimensions as theories constructed, but each dimension contributed differently to formality construct. This formality model was evaluated through comparisons between human construct of formality, Flesch-Kincaid Grade Level, and genres. All results showed the weighted formality scores had much higher correlations with human judgments of formality than non-weighted formality scores.