teacher vs. machine correction: comparing assessments of students’ reading comprehension and writing skills

This article presents research that compared two correction techniques applied to a pisa text summary question written by 30 spanish students aged 14-16, one by automatic correction software (g-rubric) and the other by 30 spanish language teachers varying in age, sex, and classroom experience. the methodology was a parametric approach based on latent class analysis using latent gold 4.5 software, and correspondence analysis. in the results, the euclidean distances between each individual and the system were measured as low, medium or high dissimilarity, based how close the teachers’ assessment was to that of the correction software. the results showed a first cluster, comprised of teachers whose correction scores exhibited a significant correlation with the tool, represented the quartile of younger and less experienced teachers. this stands in contrast to a second cluster, characterized by high dissimilarity, which consisted of older and more experienced teachers whose corrections deviated notably from the system, yielding scores lower than those produced by the tool.