Neuro-Symbolic Legal Reasoning: Reasoning by PROLEG, Rule and Fact Extraction by LLM

The rapid development of generative AI has created new opportunities for applying artificial intelligence to the legal domain. Large language models (LLMs) can process legal documents and answer legal questions in natural language. However, their tendency to produce hallucinations raises serious concerns about the reliability of legal reasoning. This lecture introduces a neuro-symbolic approach that combines LLMs with symbolic legal reasoning to provide reliable conclusions and understandable explanations.

First, I introduce PROLEG, a legal reasoning system based on the Japanese Presupposed Ultimate Fact Theory. This theory provides a framework for legal reasoning under incomplete information by distinguishing general rules from exceptions. PROLEG implements this framework using logic programming and can derive legal conclusions while generating explanations of the reasoning process. More than 2,500 rules concerning Japanese contract law have been formalized in PROLEG.

Next, I present our approach to integrating LLMs with PROLEG. One challenge in using symbolic legal reasoning systems is that legal rules and case facts must be represented in logical formulas, which are difficult for legal professionals to construct manually. To address this problem, we investigate two complementary techniques: extracting structured facts from natural-language descriptions of legal cases and generating PROLEG rules from legal provisions. For rule generation, we introduce an example-driven approach in which an LLM generates examples from normative texts and generalizes them into formal rules. Experimental results show improvements in both rule completeness and behavioral accuracy compared with direct LLM-based formalization.

Finally, I demonstrate how the resulting system can perform legal reasoning and visualize its explanations, using examples involving contract law and data protection regulations. This combination of generative AI and symbolic reasoning offers a promising direction for developing more reliable, explainable, and accessible AI systems for legal applications.

Ken Satoh has been studying the logical foundations of artificial intelligence for more than 35 years. In 2009, seeking practical applications of AI research in the legal domain, he graduated from the Law School of the The University of Tokyo, one of the leading law schools in Japan, and passed the Japanese bar examination in 2017. He also founded the JURISIN workshop series in 2007 and COLIEE workshop series in 2014. As a major contribution to AI and Law research, he proposed PROLEG, a legal reasoning system based on logic programming through which he formalized more than 2,500 rules from the Japanese Civil Code. He is widely recognized as one of the leading researchers in the field of AI and Law in Japan.

Moderator: Marianne von Blomberg, Global Academic Fellow, The University of Hong Kong Faculty of Law

This is an in-person event but presentation will be conducted online.

To register, please go to https://hkuems1.hku.hk/hkuems/ec_regform.aspx?guest=Y&UEID=110341.

For inquiries, please contact Ms. Grace Chan at mcgrace@hku.hk / 3917 4727.