Predictive processing during simultaneous interpreting: Insights from the visual-world paradigm, interpreting performance, and retrospective self-reports

Despite growing interests in predictive processing during simultaneous interpreting (SI), the real-time processing mechanisms supporting it have only received limited investigation, leaving a gap in our understanding of how interpreters dynamically utilise predictive processing in practice. This study systematically investigates predictive processing during SI through the triangulation of the visual-world paradigm, interpreting performance, and retrospective self-reports. A total of twenty-two professional interpreters and forty-four interpreting students were recruited to perform SI tasks involving multi-sentence paragraphs while viewing visual displays containing a target object, two semantic competitor objects, and one distractor objects. The collected data were analysed to uncover the presence, mechanism, and effects of predictive processing during SI.

First, this study examined whether professional and student interpreters predict semantic information about upcoming content in their second language during SI. Both groups demonstrated predictive eye movements toward the target objects before hearing the corresponding words in contextually constraining sentences, indicating the presence of predictive processing under demanding conditions. Second, by integrating eye-tracking, performance, and retrospective data, this study explored the mechanisms underpinning predictive processing during SI. Two distinct mechanisms were identified: prediction-by-production, characterised by top-down simulation of upcoming content, and prediction-by-association, driven by automatic semantic-thematic and lexical activation. These mechanisms appeared to operate in parallel, with interpreters switching between them depending on contextual and cognitive demands.

Thirdly, this study explored the relationship between predictive processing and interpreting performance. While no conclusive relationship was identified, high-performing interpreters, whether defined by short production latencies, high output quality, or self-reported anticipatory strategies, tended to exhibit earlier and more robust predictive fixations. The convergence across eye movements patterns, SI performance, and subjective attitudes points to a critical role of metacognitive control in shaping predictive processing during SI. Finally, this study investigated expertise-related differences in the cognitive rhythms and strategic control of predictive processing during SI. The professionals displayed more pronounced and temporally dynamic predictive fixations, more consistent SI performance across conditions, and greater metacognitive awareness, suggesting deeper engagement in top-down, production-based prediction. In contrast, students demonstrated more static gaze patterns, prioritised input-output synchrony, and often adopted a streamlined processing strategy, reflecting reduced cognitive flexibility and heavier reliance on reactive, association-based prediction.

Together, these findings highlight predictive processing as a dynamic, individualised, and strategically modulated component of SI, shaped by both automatic cognitive mechanisms and metacognitive control. Implications are discussed for interpreter training and for future research into cognitive adaptation in high-demand language tasks.