Algorithmic Legal Hegemony In Ai Translation: Greek Criminal Law Between Continental And Common Law Ontologies
Auteurs : Stavroula Paraskevi Vraila .
Résumé
The deployment of Large Language Models (LLMs) in legal translation raises a structural paradox: systems trained predominantly on English-language data are increasingly tasked with mediating between legal traditions that share no ontological common ground with Anglo-American law. This study investigates algorithmic bias in LLM-mediated translation of criminal law concepts between Greek and German, two continental legal systems rooted in a shared Roman law heritage, and introduces the concept of algorithmic legal hegemony to account for the systematic dominance of common law frameworks in model outputs. Drawing on a purpose-built corpus of Greek Penal Code concepts translated by three state-of-the-art LLMs (GPT-4, Claude 4.5 Opus, and Gemini 3.1 Pro) under three prompting conditions, and combining qualitative doctrinal evaluation with quantitative Zipfian lexical richness metrics, the study identifies three recurring patterns of bias: anglicization, conceptual simplification, and framework substitution. These patterns persist even under expert continental law prompting conditions, indicating that the biases reflect architectural properties of the models rather than correctable surface habits. The findings carry direct normative implications. Systematic doctrinal distortion of the kind documented here is incompatible with the quality standards established by EU Directive 2010/64/EU and with the fair trial guarantees of Article 6 ECHR and raises accountability questions under the EU AI Act (2024) and the CEPEJ 2025 Guidelines on Generative AI in Judicial Systems. The study argues that the concept of algorithmic legal hegemony, designating a structural rather than incidental form of translational bias, is necessary for both the theoretical analysis and the institutional governance of AI-mediated legal translation.
Mots clés
Algorithmic bias ; Large Language Models ; Legal translation ; Translation equity ; AI-mediated translation ; Criminal law terminology
