Assessing Readability Formulas in the Wild

Abstract

Recent advances have facilitated major improvements in developing intelligent and purpose-oriented readability formulas to predict the overall difficulty of a text in terms of text comprehension and processing. Such readability formulas are mediating technologies that help match appropriate reading texts with students, thus enabling the development of smart learning environments that adapt learning resources to learner skills. Newer readability formulas include linguistic features that are more predictive of human judgments of text readability than traditional readability formulas, such as Flesch-Kincaid Grade Level. However, in many cases, these formulas have not been tested beyond their ability to predict reading scores. The purpose of this study is to examine the validity of newer readability models along with more traditional readability formulas using behavioral data and text comprehension scores. The results indicate that readability models employing linguistic features more theoretically related to text processing and comprehension outperform readability models that do not employ similar features. The findings support the long-term growth of readability formulas that are continuously improved to increase the wellbeing of learners.