Toward automatic support for leading court debates: a novel task proposal & effective approach of judicial question generation

Changzhen Ji, Yating Zhang, Xiaozhong Liu, Adam Jatowt, Sourav S. Bhowmick, Changlong Sun, Conghui Zhu, Tiejun Zhao

Abstract: Court debate, with multiple parties (e.g., judge, plaintiff, defendant), is an essential component in a civil trial where the judge leads the conversation and the litigants respond in turn following the judge’s question. Unlike other types of dialogues, the judge’s leading role can be critical with respect to the goal of case investigation, and it is non-trivial to examine the case logic considering also the need for specialized domain knowledge. Judge question generation in court debate is a novel but significant task to assist/train the junior judges to raise effective questions in a legal context as well as help the litigants prepare a court debate in advance. We propose an innovative end-to-end model called ’Judicial Questioning Aid’ which is capable of proactively leading the court debate by asking useful questions to a certain litigant given previous context. Unlike prior efforts in Natural Language Generation (NLG), the proposed model encodes the contextual utterance information with respect to global legal knowledge and local case judicial factors, as well as simulates the intention switch across different conversation turns. Extensive experiments based on a large civil trial dataset show that the proposed model can generate more accurate and readable questions against several alternatives in the multi-party court debate scene.