The Discriminativeness of Internal Syntactic Representations in Automatic Genre Classification

ABSTRACT Genre characterizes a document differently from a subject that has been the focus of most document retrieval and classification applications. This work hypothesizes a close interaction between syntactic variation and genre differentiation by introspecting stylistic cues in functional and structural aspects beyond word level. It has engineered 14 syntactic feature sets of internal representations for genre classification through Machine Learning devices. Experiment results show significant superiority of fusing structural and lexical features for genre classification (F∆max. = 9.2%, sig. = 0.001), suggesting the effectiveness of incorporating syntactic cues for genre discrimination. In addition, the PCA analysis reports the noun phrases (NP) as the most principle component (66%) for genre variation and prepositional phrases (PP) the second. Particularly, noun phrases with dominant structures of prepositional complements and pronouns functioning as a subject are most effective for identifying printed texts of high formality, while prepositional phrases are useful for identifying speeches of low formality. Error analysis suggests that the phrasal features are particularly useful for classifying four groups of genre classes, i.e. unscripted speech, fiction, news reports, and academic writing, all distributed with distinct structural characteristics, and they demonstrate an incremental degree of formality in the continuum of language complexity.