Readability Assessment of Academic Texts at Different Degree Levels

Developing machine learning tools to aid students in the process of writing a thesis document is of great interest to students, universities, supervisors and evaluation committees. This article presents the construction and evaluation of readability comparators based in Spanish-written thesis documents of four different academic levels: Advanced College Level Technician (ACT), Undergraduate, Master and Doctoral. Specifically, we provide comparators that can evaluate, between two academic texts which one is more readable than the other. In particular, we focus on three thesis sections: Problem Statement, Justification, and Results. The successful completion of these different comparators, as shown in results, allowed to test readability evaluators for new undergrad writings, determining whether correspond to its academic level. Some guidelines to determine comparators and academic level archetypes for readability assessment are proposed, based on our results.