Beyond the surface: uncovering trainee translators’ reading processes during GenAI translation quality evaluation tasks

Abstract

This study examines the influence of text genre, i.e. literary and tourism texts, and prompting techniques, specifically Chain-of-Thought (CoT) reasoning and Domain-Specific Prompts (DSP), on trainee translators’ reading processes when evaluating GenAI-generated translations. Drawing on eye-tracking and post-task cued retrospection data from 30 trainees, the study found that (a) GenAI-generated translations of literary texts required longer and more frequent fixations than those of tourism texts during evaluation, indicating intricate genre-specific reading behaviours; (b) the use of DSP, when combined with CoT reasoning, reduced cognitive effort in translation quality evaluation than the CoT-only case, as evidenced by shorter fixation durations on the rating scale; (c) trainees demonstrated a general serial genre-independent reading pattern that begins with the source text, followed by the shifts between the source and target texts, and concludes with the rating scale. These findings offer insights into the cognitive mechanisms underlying reading for post-editing purposes across diverse genres in a GenAI era, further highlighting the necessity of integrating metacognitive awareness and tailored prompting strategies into translator training.