ECALP: Empirical Computational Argumentation in Legal Proceedings

该项目旨在为以下面向方法论的领域的法律诉讼计算论证的基础作出贡献：(1)在具有挑战性的“小数据”情景和复杂法律领域的背景下进行强有力的论证挖掘；(2)需要专家知识的可扩展的数据获取方法。为此，我们解决了以下三个研究问题(RQ)：RQ1：基础。如何在法律程序中注释、建模和挖掘法律论点和论证？RQ2：语境化。如何在法律诉讼中模拟和识别类似的辩论模式？RQ3：洞察力。人权法庭上口头和书面辩论的作用是什么？这些研究问题分三个连续的阶段来解决。第一阶段是数据获取和注释，第二阶段是建模和实证分析，第三阶段是研究论元在不同法律体系中的重要性和论元迁移。在计算论证领域，我们确定了两个关键的方法论方向，在这些方向上，自然语言处理的进展将使法律程序的实证研究成为可能--即可以有效部署和转移的神经方法，以及为专家评估提供适当模型置信度的贝叶斯方法。我们还设想对经验性法律研究(例如将法律论证模型的理论与经验证据进行对比)做出贡献，并通过共同的任务扩展到更广泛的自然语言处理社区。

This project aims to contribute to the foundations of computational argumentation in legal proceedings in these methodology-oriented areas: (1) robust argument mining in the challenging context of a "small-data" scenario and complex legal domain, and (2) scalable methodologies for data acquisition that require expert knowledge. To do so, we tackle the following three research questions (RQ): RQ1: Foundations. How to annotate, model, and mine legal arguments and argumentation in legal proceedings? RQ2: Contextualization. How to model and recognize similar argumentation patterns in legal proceedings? RQ3: Insights. What is the role of oral and written arguments at human rights courts? These research questions are tackled in three consecutive phases. The first phase is devoted to data acquisition and annotation, the second phase to modeling and empirical analysis, and the third phase to research on argument importance and argument transfer across different legal systems. Within the field of computational argumentation, we identify two key methodological directions where progress in NLP will empower empirical research on legal proceedings - namely, neural approaches that can be deployed and transferred efficiently and Bayesian approaches that provide proper model confidences for expert assessment. We also envision contributions to empirical legal research (such as contrasting theories of legal argumentation models with empirical evidence) and an outreach to the wider natural language processing community through a shared task.