Exploring preparatory reading in bidirectional sight and written translation through clustering analysis of eye-tracking data

Preparatory reading—the phase between a translator’s initial reading of the source text and the production of the first word of the target text—remains underexplored despite its crucial role in both sight (SiT) and written translation (WT). This study examined preparatory reading patterns of 32 student translators, focusing on the effects of translation mode (SiT vs. WT) and direction (L1-to-L2 vs. L2-to-L1). Translators’ attention allocation, cognitive effort, and reading speed were measured using preparatory reading duration, average fixation duration, and fixation rate (i.e., number of fixations per second) as key indicators. Using linear mixed-effect models, we quantified the effects of translation mode and direction on each measure. We then synthesized all three measures by employing unsupervised machine learning algorithms (i.e. k-means cluster analysis) to identify distinct reading patterns. These quantitative findings were further complemented by expert qualitative categorization of reading patterns, achieved through subjective coding of scanpaths and Translation Process Graphs. We found that translation mode played a primary role in shaping preparatory reading styles, with SiT consistently requiring greater cognitive effort, more attention, and slower reading speeds than WT—particularly under the L1-to-L2 condition. Translation direction further modulated these effects in nuanced ways. Specifically, L2-to-L1 was associated with increased attention allocation in WT, but with less cognitive effort and faster reading in SiT. Additionally, we found three distinct preparatory reading patterns: Fast Surface-level Preparatory Reading, Systematic Deep-level Preparatory Reading, and Extended Iterative Preparatory Reading, each reflecting a distinct combination of cognitive investment and reading speed. These findings could advance our understanding of translators’ preparatory reading behaviors and underscore the need to equip them with adaptable, task-sensitive reading strategies that align with the cognitive demands of different translation modes and directions.