Translation Studies has long theorized ideologically motivated rewriting (Baker, 2006; Hermans, 1985; Venuti, 1995). Critical-discourse approaches have analyzed manipulation in translated political discourse including recontextualization in institutional news translation (Kang, 2007; Schäffner, 2004) and a model applying van Dijk's ideological square to translation (Daghigh, Sanatifar, & Awang, 2018), while translation quality frameworks catalogue errors without addressing ideology (House, 1997; Lommel et al., 2014). Building on these foundations, this paper takes a step further: it uses ideological manipulation as a reproducible annotation scheme that can be applied consistently across a large parallel corpus, rendering insights coder-checkable and corpus-scalable through a scheme of fifteen computationally tractable tags.
The scheme is designed for ideologically loaded content and not claimed to generalize to neutral or technical translation. Each tag has a theoretical provenance with a distinct operational role: Van Leeuwen's (2008) recontextualization model yields the content operations (EXCIS, INSERT, LEXI); van Dijk's (1998) ideological square supplies the directional coding (+US/−US/+THEM/−THEM) and the calibration tags (INTENSIFY, SOFTEN, POLARITY); Bourdieu's habitus, via Simeoni (1998), grounds the habitus-conversion tags (HAB-REG, HAB-SCHEMA, HAB-VOICE, MODAL); Chilton's (2004) deixis underpins the provenance-erasure tags (ERASE-DEIC, ERASE-ATTR); and Baker's (2006) narrative framing informs the macro-level diagnostics (PSEUDO-ORIG, HABIT-CONFORM, TROJAN).
At the same time, this papers reflects on the human-AI synergy in the spirit of human-in-the-loop NLP design (Wang et al., 2021): The author specified the frameworks and the discursive elements; a large language model assisted in mapping each element and served as annotation assistant, and all outputs were author-verified. The scheme yielded 617 tagged operations and a consistent directional asymmetry across authors and genres in a 5,042 aligned English-Greek pairs from Golden Dawn's translations of far-right texts.
The paper offers both an instrument and a candid account of where AI assistance ends and expert human judgement must begin.
Valentini Kalfadopoulou is a doctoral candidate in the Department of Foreign Languages, Translation and Interpreting of the Ionian University, where she has recently submitted her thesis on ideology, translation, and register across the American far right and Greece's Golden Dawn. Her research focuses on translation strategies, the manipulation of translated political discourse, hate speech, terminology, and the use of AI in Translation and Terminology. She lectures on the intersection of Translation, Political science, and Computational approaches. She is an associate member of the Center for Language and Politics at the Ionian University. She is also a long-standing contributor and speaker at the TermNet’s Terminology Summer School and an Organizing Committee member of the international conference Translating and the Computer (TC). She is serving her second term as General Secretary of the Hellenic Society for Translation Studies (EEM) and is a national expert for Greece on the ISO/TC 37 committee.
