arxivcs.AI2026-07-05
Shortcut Learning in Legal Judgment Prediction: Empirical Evidence from the UK Employment Tribunal
Joe Watson, Joana Ribeiro de Faria, Marcus Tomalin, Måns Magnusson, Huiyuan Xie, Hao Tian Yeung, et al.
Current Legal Judgment Prediction (LJP) is constrained by its reliance on post-hoc judicial materials, increasing the likelihood that models perform retrospective classification rather than true forecasting. This paper empirically investigates shortcut learning in this context by…