Learning Legalese Through AI: A Directional NLP and Learning-Analytics Study of Formulaic Legal Register in English-Arabic Student Translation
Description
This article investigates how artificial-intelligence-mediated translation tasks support, shape, and sometimes distort student acquisition of formulaic legal register across English-Arabic translation directions. The study analyzes a legal subcorpus of 37 classroom task records extracted from a broader collection of 65 student submissions: 26 Arabic-to-English legal tasks and 11 English-to-Arabic legal tasks. The records include direct translation, post-editing assignments, embedded process metrics, learner reflections, and, in several cases, prompt traces. The article reframes the dataset as a small but information-rich learning-analytics corpus and proposes a computationally tractable workflow for studying AI-mediated register learning. The method combines directional task profiling, target/source length-ratio analysis, formulaic-marker extraction, reflection coding, and close qualitative interpretation of legal phraseology. Results indicate a strong directional asymmetry. Arabic-to-English tasks tend to encourage phraseological elevation into Anglo-legal drafting through expressions such as “sets forth,” “undertakes,” “therefrom,” “shall constitute,” and “governed by and construed in accordance with.” English-to-Arabic tasks, by contrast, tend to stabilize legal force through compact Arabic normative markers and contractual collocations such as “يلتزم,” “لا يجوز,” “بعناية ومهارة واجتهاد,” and “يُعد باطلًا ولاغيًا.” The article argues that AI in this corpus functions as a phraseological tutor: it offers candidate formulae, accelerates register exposure, and gives students material for metalinguistic reflection. Yet this tutoring is double-edged, because legal-sounding output can mask weak control over legal function, deontic force, and genre appropriateness. The article contributes a direction-sensitive AI register-learning framework for legal translation pedagogy and outlines how future NLP-supported systems could detect formulaic uptake, prompt dependence, simplification, and over-legalization in student work.