摘要
法律文本特别是刑法的解释,原则上以词语的“通常含义”为边界。通常含义指的是大多数普通人在特定语境下对法律文本的理解,若偏离通常含义对法律文本进行解释,则要进行充分的说理论证。判断通常含义时,主要借助语言直觉、词典、法律语料库三种辅助工具。在大模型技术赋能法律语料库的背景下,大模型法律语料库通过数据驱动范式,将法律解释从依赖主观直觉转向可验证的客观分析。这一转向不仅增强了解释过程的透明度,提升了解释结论的客观性,更通过技术约束机制倒逼解释者履行论证义务。作为一种辅助解释工具,大模型法律语料库既不试图阻止解释者进行主观判断,亦不会限制或者放纵解释者的解释权。针对大模型法律语料库建构阶段的技术内生操纵风险,应公开其基础数据来源和技术处理全流程信息,并确保辅助法律解释的过程和结论可重复、可推广。针对使用阶段的人为外生操纵风险,应构建技术与制度协同治理框架,确保解释者主观选择的技术可见性与可辩驳性。
In principle,the interpretation of legal texts should be bounded by the"ordinary meaning"of words,which refers to the understanding of legal texts by most ordinary people in a specific context.Deviation from this"ordinary meaning"requires sufficient reasoning and argumentation.When determining the"ordinary meaning",three interpretive tools including linguistic intuition,dictionaries,and legal corpora can be leveraged.Against thebackground of large model technology empowering legal corpora,large model legal corpora leverage a data-driven paradigm to shift legal interpretation away from reliance on subjective intuition and toward verifiable,objective analysis.This shift not only enhances the transparency of the interpretive process and the objectivity of interpretive conclusions but also,through technological constraints,compels interpreters to fulfill their duty of justification.As an auxiliary interpretive tool,the large model legal corpus neither attempts to prevent interpreters from relying on subjective judgments nor restricts or indulges their interpretive discretion.Regarding the inherent technological manipulation risks during the construction phase of large model legal corpora,it is essential to publicly disclose their foundational data sources and comprehensive information about the technological processing workflow.Additionally,analytical processes and conclusions supporting legal interpretation should be replicable and generalizable.To address exogenous human manipulation risks during the usage phase,a collaborative governance framework integrating technology and institutional safeguards should be established to ensure the technological visibility and contestability of subjective choices of the interpreters.
出处
《法学研究》
北大核心
2025年第6期186-203,共18页
Chinese Journal of Law
基金
山东省“泰山学者工程专项经费资助”(tsqn202306038)的阶段性成果
山东大学人文社会科学“全面依法治国战略实施中的数据运用与数据治理”创新团队项目建设经费资助。
关键词
法律文本
大模型
语料库
法律解释
通常含义
legal texts
large language models(LLMs)
corpus
legal interpretation
ordinary meaning
