Navigating AI feedback in translation training: how text type, proficiency, and attitude shape students' acceptance...

Area of Science:

• Computational Linguistics
• Educational Technology
• Translation Studies

Background:

• The integration of Large Language Models (LLMs) into educational settings presents new opportunities and challenges.
• Understanding student interaction with AI-generated feedback is crucial for optimizing learning experiences.

Purpose of the Study:

• To investigate how undergraduate and graduate translation students engage with and evaluate LLM-generated feedback.
• To analyze acceptance rates, influencing factors, rationales, and perceived limitations of AI feedback.

Main Methods:

• Mixed-methods approach involving 78 translation students (undergraduates and postgraduates).
• Students completed translation tasks with ChatGPT-3.5 feedback, making binary accept/reject decisions with rationales.
• Semi-structured interviews were conducted to explore evaluative criteria and feedback deficiencies.

Main Results:

• Students accepted an average of 68.2% of LLM suggestions, showing selective engagement.
• Acceptance varied significantly by text type (technical/news highest, literary/tourism lowest).
• AI attitude, proficiency, and academic level (undergraduate vs. postgraduate) moderated engagement.

Conclusions:

• Student engagement with LLM feedback is influenced by task characteristics, individual differences, and expertise.
• LLM feedback deficiencies include cultural insensitivity, stylistic flattening, and lack of context.
• Translator training needs contextually aware, stylistically adaptive, and dialogic AI feedback systems.