Political Translation in the AI Era: Journalists as Dual Agents and Systemic Vulnerabilities in Cross-Linguistic Political Communication
Keywords:
political translation, journalistic translation, neural machine translation, large language models, prompt design, cross-cultural communicationAbstract
This study examines how German political statements change when rendered for Greek audiences and what AI systems return for the same material. A corpus of sixteen documented cases from 2010 to 2025, thirteen in which the Greek departs from the German and three in which it holds, yields a taxonomy of five patterns: context isolation, systematic tone modification, narrative-driven translation, weaponization under polarization, and semantic manipulation. Three of the cases were submitted to GPT-4o and GPT-5.3 Instant under five prompt conditions, three runs per cell, and once each to DeepL and Google Translate, giving 96 coded outputs. Under the two conditions that ask only for a translation, seventeen of eighteen language model outputs in each carry none of the five patterns in the author's coding, as do five of the six machine translation outputs. An instruction to translate for a Greek news website returns adaptations in fifteen of eighteen outputs, against one of eighteen under a bare instruction, and no headlines.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2026 Stavroula Paraskevi Vraila

This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License.