Uma análise do feedback corretivo provido por inteligência artificial em atividades de role-play no ELSA Speak

Abstract

This study focuses on the analysis of Corrective Feedback samples provided by the ELSA Speak app during the oral interaction of a group of participants with the app's role-play activity, in light of the state of the art in speaking instruction. This research is characterized as qualitative, descriptive, and bibliographic, as it aims to identify the types of Corrective Feedback provided by the application, as well as its characteristics and the underlying theories of language and learning. The study is theoretically grounded in Larsen-Freeman and Anderson (2011), Richards and Rodgers (2014), Oliveira (2014), Wang et al. (2024), Ellis (2009), Sheen e Ellis (2011), Nassaji and Kartchava (2021), among others. The research was conducted at the Federal Institute of Amapá (IFAP) - Macapá Campus. The participants were undergraduate students from the Portuguese/English Letters program, selected on a voluntary basis. Data analysis was conducted following Bardin’s (2011) Content Analysis perspective. Based on the analysis of the full assessment provided by the app, three categories of feedback are proposed: Positive Feedback, Suggestive Feedback, and Corrective Feedback. Within the latter category, two types of CF are proposed: Explicit correction and Explicit correction with metalinguistic explanation. Although differences were observed in the provision of the full assessment, specifically regarding the presence of fewer analytical dimensions for beginner and basic level speakers, no methodological difference was detected in the provision of the CF. The CF, isolated, presents a structuralist view of language, but it plays a fundamental role within the full assessment, which demonstrates a functionalist view. The corrections are based on American English norms and thus do not account for the view of English as a Lingua Franca. In terms of learning, the CF indicates a cognitive perspective, as it requires conscious attention to the linguistic forms present in the correction input. This study offers a significant contribution to the academic community investigating the interaction between English learning and technology, as well as the teaching of oral production through language learning apps, addressing a gap in knowledge regarding AI-generated CF for the development of speaking skills. Furthermore, it may assist teachers, researchers, and other education professionals in selecting technological resources to support English language teaching.