Simplify or preserve? Audience design, compression, and information loss in student AI-mediated translation
Description
This article examines simplification as a deliberate translational strategy in student AI-mediated translation. Post-editing research often treats simplification as a feature of machine-generated or post-edited language; this study instead shows how students use AI assistance and prompt design to compress, soften and repurpose source texts for imagined readers. The analysis draws on 65 task records extracted from 47 student submission files, representing 44 normalised student identities. Strong simplification was identified through a combined metric-and-reflection procedure: records were included when the output/source length ratio fell below 0.85 or when reflective commentary explicitly stated that the student had simplified, shortened or made the translation easier to understand. Thirteen focal records met these criteria. Twelve were labelled Translate rather than Post-edit MT, and eleven occurred in Arabic-to-English tasks, suggesting that audience redesign was concentrated in student-authored forward translation. Close reading, metric comparison and thematic coding show that simplification operated through four recurrent moves: lexical downgrading, syntactic shortening, discourse pruning and register flattening. These moves often reflected genuine audience awareness, especially when students imagined beginners, general readers or readers with limited English. However, they also produced information loss, genre drift and legal downtoning when technical or contractual source texts were recast as plain-language summaries. The article therefore distinguishes accessibility-oriented adaptation from under-translation and argues that AI-mediated translation pedagogy must make audience and purpose explicit if readability is to be evaluated fairly alongside fidelity.