The influence of structural priming in German-to-Dutch translation: a multi-methods study.

Structural priming—the tendency to repeat recently processed syntactic structures—has been widely studied in psycholinguistics (Pickering & Ferreira, 2008; Gries & Kootstra, 2017) and, more recently, has also been extended to translation studies (Maier et al., 2017; De Sutter et al., 2023; Jacob et al., 2024). However, research on structural priming in translation remains in its early stages, and key methodological and theoretical issues persist. Corpus studies are criticised for their inability to conclusively prove priming effects (Branigan et al., 1995), while controlled experiments lack ecological validity (Gries, 2005), underscoring the need for methodological integration. Additionally, conflicting results suggest trained translators may resist priming more than untrained bilinguals, but no study directly compares them.

This research investigates whether and to what extent structural priming occurs in German-to-Dutch translation, focusing on voice alternation. It examines whether translators are influenced by source-text voice structure and whether translation training modulates this effect. Since agentless passives constitute most passives and their direct alternants are generalized actives—active sentences with nonspecific subjects (Weiner & Labov, 1983)— our investigation focuses on these constructions. To bridge the methodological divide, we combine experimental and corpus-based approaches. Additionally, we address the gap in previous research by comparing bilinguals without translation training to translation students, examining whether they differ in susceptibility to structural priming.

The corpus study analyzes InterCorp (V16UD), focusing on a German-to-Dutch sample of 700 instances of agentless passives and passivizable generalized actives. These instances are obtained by automatically querying and extracting relevant sentences from the corpus, followed by a manual validation process to ensure precision. Each instance is annotated for Dutch voice, German voice, and additional predictors (e.g., genre, VP complexity, syntactic weight, person, animacy, and definiteness of the constituents). By including the 'German voice' predictor, we assess whether the source-text structure systematically influences the target-text structure—an essential step in identifying potential priming effects. To further investigate this influence, we compare translated and non-translated Dutch by analysing an original Dutch sample of 300 instances of agentless passives and passivizable generalized actives. We expect linguistic constraints to function similarly in both datasets but hypothesize that the distribution of voice constructions will differ, with translated Dutch more closely aligning with German source structures, reflecting priming effects.

To complement the corpus study, we conduct an experiment capturing real-time translation behaviour (since corpus data reflect only the final translation and cannot definitively establish priming effects). The experiment aims to involve 100 participants—50 German-Dutch bilinguals without translation training and 50 translation students—matched in age (20–30 years), native language (Dutch), and German proficiency (B2–C1). Participants do not receive a specific translation brief beyond basic instructions, to avoid influencing their translation strategies. During the experiment, each source sentence appears on screen for a brief, controlled duration before disappearing, ensuring translation choices reflect participants’ mental representations and providing clearer evidence of priming. All data are recorded via Pavlovia. Building on the need for direct comparisons between bilinguals with and without translation training, we expect both groups to show priming effects, but trained translators to rely less on source-text structures due to their expertise. 

To allow direct comparison between real-time and published translations, experimental stimuli are authentic corpus sentences drawn from fictional texts in InterCorp and part of the aforementioned corpus study, controlled for length (six words), genre, syntactic type (main clauses), and basic word order patterns. Lexical frequency and VP complexity are considered in statistical analyses, conducted in R. This methodological integration strengthens the study by capturing priming effects in controlled conditions while also evaluating ecological validity through the use of authentic stimuli, advancing our understanding of translation processes.