Process visualization feedback in news translation: a controlled intervention study of human-AI interaction using multimodal process data

Introduction This study examines process visualization feedback in news translation. Methods Framed within Krings' three-dimensional model, a pretest-posttest control group design was employed with 24 MTI students. The final analytic sample comprised 22 participants (11 experimental, 11 control). The experimental group ( n = 11) received personalized process visualizations and structured reflection; the control group ( n = 11) received conventional product-oriented feedback. Translation scores were available for both groups at T1 and T2, and complete process data (Inputlog, screen recording) were available for the experimental group at both time points. Results Within the experimental group, within-subject comparisons showed changes in all three dimensions: in the temporal dimension, increased pause time proportion indicated heightened metacognitive engagement ( p = 0.011); in the technical dimension, the meaning-level revision ratio increased significantly from 18.1% to 28.4% ( p = 0.002), indicating a shift from surface-level correction to deeper substantive refinement; in the cognitive dimension, participants made more strategic use of external sources, with significant increases in search engine queries ( p = 0.02) and knowledge-based searches ( p = 0.01). Rich-points analysis of problem-triggered pauses revealed a shift from linguistic toward textual and extralinguistic problems ( p = 0.038, p = 0.015). A mixed-design ANOVA revealed a significant Group × Time interaction ( p = 0.038, η p 2 = 0.198). Thematic analysis of reflective journals ( n = 22) and interviews ( n = 6) revealed two overarching themes: metacognitive awareness (54 references) of inefficient process patterns and competency gaps, and strategic regulation (60 references) in external source use and text-level restructuring. Discussion These findings suggest that process visualization feedback was associated with enhanced metacognitive awareness and strategic regulation, and may support learners in becoming more reflective, strategic, and autonomous in human-AI collaborative news translation.