Towards A Computational Assessment of Freewriting Quality

This study examines the linguistic features of freewrites and how those features relate to human scores of freewrite quality. Freewriting is a common prewriting strategy that has received little attention by researchers, particularly in terms of the linguistic features of good and poor freewrites. To address this issue, we developed a scoring rubric to assess the qualities of freewrites and how they are correlated with linguistic features. The results showed that many linguistic features positively correlated with human scores (e.g., referential cohesion, syntactic complexity, lexical difficulty), but the only significant predictors in a regression analysis were, number of words and noun overlap. Better freewrites are longer ones with lexical overlap between sentences. While these results fail to conclusively exclude other potentially important features of higher quality freewrites, this study is a first step toward computationally defining freewrite quality.

• Type: Conference
• Year: 2010
• Where: FLAIRS
• Authors: Jennifer L. Weston, Scott A. Crossley, Danielle S. McNamara
• Keywords: Artificial Intelligence, FLAIRS 2010, Freewrite Quality, Human Scores, Linguistic Features