Abstract

Reading comprehension in school contexts is influenced not only by the reader’s characteristics but also by the linguistic features of the texts. Within this framework, textual complexity has become a relevant research focus, particularly for improving the criteria used to select educational materials. This study analyzes the relationship between teachers’ judgments regarding the complexity of school science texts and the automated linguistic indices generated by TRUNAJOD, a computational tool designed to assess readability in Spanish. A quantitative, non-experimental, correlational design was used. Reading and writing specialists participated by evaluating a corpus of 80 science texts extracted from secondary school textbooks. The texts were grouped by educational level, and the data were analyzed using hierarchical regressions, Student’s t-tests, and Kendall’s coefficient of concordance. Results indicate that the most predictive indices of perceived textual complexity are lexical (familiarity, frequency, diversity, imageability) and cohesive (verb overlap, causal markers). Furthermore, variation in evaluation criteria was observed by educational level, with broader linguistic attention in lower grades and a narrower focus on lexical elements in higher grades. It is concluded that teachers’ judgments reflect sensitivity to the linguistic properties of texts, although mediated by their disciplinary background. This study provides evidence for the development of hybrid tools for textual evaluation and encourages further research using expanded corpora and evaluators from diverse subject areas. © 2025 Asociacion de Linguistica y Filologia de America Latina. All rights reserved.