<p>Text marking has been extensively studied in reading research, but less is known about its effects in sight translation, a task requiring simultaneous visual input processing and oral output production, particularly across proficiency levels. We tested 64 Chinese L1 translation majors (28 undergraduates, 36 graduates) using a 2 (Proficiency: undergraduate vs graduate, between) × 2 (Marking: marked vs unmarked, within) mixed design with controlled rolling text display (eliminating re-reading confounds). Undergraduates benefited substantially from marking (average gain: + 9.18 points, <i>d</i> = 0.72–0.89), while graduates showed minimal gains (average: + 5.43 points, <i>d</i> = 0.23–0.41), yielding a significant proficiency × marking interaction (<i>F</i>(1,60) = 4.82, <i>p</i> = .032, <i>ηp</i>^<i>2</i> = .07). Behaviorally, undergraduates reduced both pause and revision rates substantially with marking, whereas graduates maintained efficient processing across conditions with minimal marking effect. Pause-revision coupling was stronger for undergraduates (<i>r</i> = .76) than graduates (<i>r</i> = .43), consistent with more selective resource allocation at higher proficiency levels. Retrospective reports confirmed this pattern: undergraduates described marking as “essential for segmentation,” while graduates reported “selective use” or “occasional distraction.” These findings align with expertise reversal theory and inform adaptive pedagogy. Marking serves as critical scaffolding for novices but may become optional or interfering as proficiency develops.</p>