Title: Teacher and AI-assistant: potential and problems for correction tasks

Name of presenter: Jutta Ransmayr

Affiliation and academic title: University of Vienna (German Department and Centre for Teacher Education), Assoc. Prof. Dr.

Email address: jutta.ransmayr@univie.ac.at

​

Keywords: teachers, correcting and marking learner text, AI

​

Abstract:

Artificial intelligence (AI) has emerged as a societal phenomenon that increasingly shapes educational contexts and school-based learning. Educational institutions are currently engaged in processes of adaptation to AI, ranging from proactive integration into classroom practices to fundamental changes in task design, such as the reduction or elimination of written homework due to students’ widespread and largely unguided use of AI tools. While much of the current discourse focuses on learners and their interaction with AI, the present pilot study shifts the perspective to teachers and examines the potential of AI to support instructional work. Specifically, the study investigates the extent to which AI can assist teachers in the correction and feedback of learner texts in language-related subjects. The paper presents initial findings on the use of AI as a correction assistant. Results indicate that, at present, AI can only be employed in a limited capacity for text correction; however, when guided by carefully designed prompts and highly granular task specifications, AI is able to provide targeted support to teachers. At the same time, the findings underscore that human oversight remains indispensable (“human in the loop”). Methodologically, the study adopts a corpus-analytical approach: Two learner texts (written in German) – one produced by an upper-secondary school student and one by a master’s-level university student – were evaluated by both AI systems and human teachers. The analysis focuses on differences between human and AI-generated feedback and examines how the precision of AI feedback varies depending on prompt design.