Textual reviewing in ChatGPT times

Abstract

This article aims to present the results of the research that evaluated ChatGPT's ability to revise texts, not only to identify and correct grammatical errors, but to perceive broader aspects relating to discursive sphere to which the texts in focus were integrated. It was used an exploratory, bibliographic, and qualitative research as conceived by Gil (2017). We proposed to the algorithm three authentic texts containing deviations of different natures,  grammatical and discursive, to be reviewed, corrected, enhanced, and evaluated, separately. Thus, based on an interactionist and socio-discursive approach (Volochinov, 2017; Sobral and Barbosa, 2019; Rodrigues, 2015; Salgado, 2013), we assessed ChatGPT's capacity to work with texts. The results indicated that it demonstrated notable abilities in intervening and improving texts, in the sphere of gramatical surface, but he also made mistakes. This shows us that textual review does not dispense with human oversight, especially when dealing with more complex discursive issues of  a discursive nature. It is concluded that ChatGPT can be a useful tool to improve textual review, but its performance is not capable of considering the author’s style and intention, as well as the nature of the speech in question.