About this Event

What does robot judege/ Legal Judgment Prediction (LJP) model actually predict from? I will talk about that neural LJP models without expert-informed adjustments can be vulnerable to shallow, distracting surface signals. Our experiments on neural LJP predicting European Court of Human Rights cases shows adversarial training can help better align models with experts, and in many cases can even achieve better prediction performance.

Shanshan Xu is a PhD student of Legal Tech in the Department of Informatics at the Technical University of Munich. Before transferring her PhD to TUM, she was a PhD student at L3S Research Center. Previous to that, she worked as a speech scientist at Nuance Communications/Cerence. She did her master in Cultural and Cognitive Linguistics at LMU Munich. Her research interest is NLP in LegalTech, Computational Social Science and Computational Linguistics.