An Agent-Based System to Assess Legibility and Cognitive Depth of Scientific Texts

Abstract

Knowledge transmitted through writing is suitable to be refined by understanding, criticizing, reflecting upon, and using it. Although several types of writings, from diffusion to highly specialized texts, fulfill this purpose, they differ considerably in syntax, word selection and phrases length. It is widely accepted that proper scientific writings deploy facts with detail, rigor and legibility, for which scientists acquire writing skills through experience, by following guidelines, by obtaining feedback from fellow scientists or through a combination of those approaches. We question whether scientific texts possess common characteristics that can be determined through quantitative metrics. A positive answer is confirmed by the fact that such writings in both languages, Spanish and English, display a normal probability distribution for a metric called $\mu$ legibility. Moreover, by analyzing texts through a new proposed metric called cognitive depth, scientific writings in Spanish display that analysis is the dominant Bloom’s cognitive level. These preliminary findings suggest that it is possible to evaluate and classify new manuscripts through an agent-based human-computer interactive system that informs writers if the ongoing text lies into the ranges discovered for published texts, and what is the prevalent cognitive level. By having this feedback, writers can modify their manuscripts to make them display good metrics.