Activist discourse resists standardization. As Neural Machine Translation and Large Language Models increasingly mediate cross-lingual political communication, identity-affirming language is routinely absorbed into normative form. This paper examines how prominent AI systems process LGBTQ+ terminology, including pronominal, orthographic, and lexical markers of gender identity, in political and advocacy texts translated from French and German into Greek. The study draws on a purpose-built corpus of roughly 200,000 words, compiled from official websites of private and public organizations and from guides issued by various Francophone and German-speaking states, then processed through several widely used NMT and LLM systems and annotated for terminological, morphological, and stylistic shift.
French and German mark gender inclusivity through distinct morphological strategies, notably French écriture inclusive and German geschlechtergerechte Sprache, none with a stable Greek equivalent. The corpus also tracks core identity terms: transgender, cisgender, and non-binary. These already circulate in Greek activist usage but carry unequal visibility: transgender is a recognized, marked category, while cisgender names an unmarked default that systems frequently omit or translate away.
A structural linguistic analysis identifies three recurring shifts. Neo-pronouns are rendered as standard third-person forms, a case of terminological displacement. Typographic and morphological markers of inclusivity are lost through stylistic flattening. Most tellingly, cisgender is disproportionately dropped, paraphrased, or replaced with a binary opposite of transgender, reinstating cis identity as the unmarked norm. These shifts reveal systems that default to binary-coded structures when the source is explicitly non-binary, altering the text's political framing.
Beyond documentation, the paper asks what faithful translation can mean when the target language lacks the resources the source mobilizes, or when a system erases a category rather than mistranslating it, and what this demands of translators in AI-assisted pipelines. It closes with implications for translator training and community-informed terminology development in Greek LGBTQ+ advocacy contexts.
Stavroula Paraskevi Vraila is an Assistant Professor in the Department of Foreign Languages, Translation and Interpreting at the Ionian University, specializing in Economic, Legal and Political Translation (Greek–German). She holds a PhD from the National and Kapodistrian University of Athens and a Master’s in Translation Studies from the Ionian University. Her research focuses on the translation of political and legal texts, and the applications of AI in translation, exploring issues of political ethics and digital bias. She has published widely in Greek and international journals and presented at global conferences. At the university, she teaches undergraduate and postgraduate courses and actively participates in departmental committees (Erasmus, Evaluation, Outreach). Since 2007, she has also worked as a professional translator with publishing houses and digital companies, specializing in institutional and communicative texts with an emphasis on accuracy and functional effectiveness.
