Navigating risks in multi-stage translations for sustainability communication in the age of Artificial Intelligence
Navigating risks in multi-stage translations for sustainability communication in the age of Artificial Intelligence

Summary/Abstract: Aim: In teaching and science, texts are translated from different languages. In this context, thepresent study investigates the potential distortions and systemic risks that arise when a sourcetext on energy transition and sustainability is translated multiple times across differentlanguages and by different agents, including professional translators and AI-based translationmodels. The research aims to analyze how these translations impact meaning, tone, and factualintegrity, particularly in the context of complex topics like energy transition and relatedsystemic risks. By comparing multiple versions of a text across English, Polish, and German,the study assesses the implications of translation-mediated communication in sustainabilitydiscourse.Design / Methodology: First, an English source text was created, summarizing two scientificarticles on the urgent need for energy transition and related system risk of such a transition.This text was translated into Polish by AI (text A) and by two professional translators (text Band C). The analysis of complexity (using Jasnopis) showed that the Polish texts were morecomplex (7/7, 7/7 and 6/7) for respectively texts A, B and C, than the English original text (O,5/7). Text C was selected for translation into German by AI and by two professionaltranslators. For comparison, the English source text was translated into German. The complexity of these translations was compared to the source text and the Polish versions.Afterwards, linguistic and semantic comparisons were carried out, evaluating shifts in meaningusing cosine similarity (TF-IDF) and Levenshtein distance (edit distance). Furthermore,changes in emphasis, severity, and emotional tone across translations were analyzed.Findings: This study shows that multi-stage translations in sustainability communicationintroduce significant distortions, affecting meaning, tone, and emphasis. AI translations tendto neutralize urgency and emotional intensity, while human translations introduce biases,either amplifying or softening risk perceptions. Additionally, differences in sentencecomplexity and terminology shift the focus of sustainability discourse. These findingshighlight the risks of translation-mediated miscommunication in critical topics like energytransition and systemic risks.Research limitations: The article presents a case study based on a small sample oftranslations. The results should be the basis for a more detailed research, comparing a largergroup of AI translation and professional translations due to translator’s bias, language-specificissues and the complexity of sustainability related notions.Originality / Value: This research contributes to the sustainability communication discourseby focusing on the risks of multi-stage translations, where small wording changes can lead tosignificant distortions in meaning of notions and key-concepts, where miscommunication canimpede decision-making and stakeholder involvement.

• Details
• Contents