Halverson (2017, 2024) proposes a cognitive explanation for source and target language influences in translation: translators are attracted to choices whose mental network connections are particularly salient (entrenched in memory). However, examining translation choices in a standard parallel corpus entails numerous confounding variables, since each translation originates from a different context, translator, and task condition. By contrast, MUST (Granger & Lefer 2020) contains multiple translations of the same source text, so that results differ only in the translators – about whom the corpus also provides rich metadata.
I take advantage of this opportunity to study the variability of multiple translations, i.e. why some converge on one solution while others diverge widely (cf. Malmkjær 1998). Following Halverson’s logic, I hypothesize that networks with many connections and evenly distributed entrenchment will produce greater translation variability.
Based on 11 MUST texts in the specialized domain of sustainable finance, I have analyzed 122 English noun sequences with 5,084 French translations (e.g. research firm → entreprise de recherche, entreprise dédiée à la recherche, agence de notation). These items pose formal and semantic difficulties (Lefer & De Clerck 2021).
After illustrating how these difficulties contribute to high variability, I discuss how I operationalize network entrenchment using both frequency data from mono- and bilingual reference corpora (for non-translated language), and reaction times from elicitation and judgment experiments. Results to date indicate that noun sequences with higher English frequencies and shorter experimental RTs give rise to significantly less variable translations.
(References)
Prinzie, T. (2026, September 11). The effect of entrenchment on translation variability. Multilingual Student Translation Corpus (MUST) Workshop, Aachen. https://hdl.handle.net/2078.5/280367