Impact of Text Structures on Student Language in High- and Low-Quality Summaries

This study explores the impact of text structures of source texts on student language in high- and low-quality summaries written by adult learners in AutoTutor ARC (Adult Reading Comprehension). We used the Coh-Metrix-ENA (Epistemic Network Analysis) approach to visualize the structure of language connections across five Coh-Metrix dimensions: non-narrativity, word abstractness, syntactic complexity, referential cohesion, and deep cohesion. The ENA networks detected differences in language connection structures between high- and low-quality summaries when source texts were causation rather than comparison texts. Specifically, in causation texts, the connections between word abstractness and referential cohesion, and between word abstractness and deep cohesion were stronger in the high-quality summary networks compared to the low-quality summary networks. Moreover, the connections between syntactic complexity and non-narrativity, and between syntactic complexity and deep cohesion were weaker in the high-quality summary networks compared to the low-quality summary networks. Implications are discussed for the design of summary writing interventions and instructions for academic writing.