aiqbench analysis
This study investigates the reliability of AI models in detecting implicit legal citations within French court decisions, revealing that expert disagreement serves as a critical signal for model failure rather than just noise. By analyzing 1,015 annotated pairs, the authors demonstrate that standard aggregate metrics obscure uncertainty, but reframing the task as top-$k$ ranking with multi-model consensus significantly improves precision without supervision. The work highlights fundamental challenges in applying computational methods to legal reasoning at scale, specifically regarding the separation of genuine reasoning from surface similarity.
Where Experts Disagree, Models Fail: Detecting Implicit Legal Citations in French Court Decisions
arXiv:2603.22973v3 Announce Type: replace Abstract: Applying computational methods to law at scale requires separating genuine legal reasoning from surface similarity. We study this through a concrete task: detecting implicit citations of the French Civil Code, where a court applies a statutory rule without naming it (a post-hoc question about the reasoning a court actually used). We release a benchmark of 1,015 passage-article pairs annotated by three legal experts. Our central finding is that their disagreement is itself informative: the third of cases the experts dispute are where models fail. Our best ensemble reaches an F1 score of 0.70 overall. Yet, two-thirds of its false positives fall on those disputed cases, a concentration that holds across all ten models we evaluate. Expert disagreement thus signals uncertainty in the task and in its gold label, which aggregate metrics hide. This should not block useful tools, however: reframed as top-$k$ ranking with multi-model consensus, the same signals reach 76% precision for the top-200 candidates of the benchmark without supervision.
Abstract:Applying computational methods to law at scale requires separating genuine legal reasoning from surface similarity. We study this through a concrete task: detecting implicit citations of the French Civil Code, where a court applies a statutory rule without naming it (a post-hoc question about the reasoning a court actually used). We release a benchmark of 1,015 passage-article pairs annotated by three legal experts. Our central finding is that their disagreement is itself informative: the third of cases the experts dispute are where models fail. Our best ensemble reaches an F1 score of 0.70 overall. Yet, two-thirds of its false positives fall on those disputed cases, a concentration that holds across all ten models we evaluate. Expert disagreement thus signals uncertainty in the task and in its gold label, which aggregate metrics hide. This should not block useful tools, however: reframed as top-$k$ ranking with multi-model consensus, the same signals reach 76% precision for the top-200 candidates of the benchmark without supervision.
Artificial Intelligence (cs.AI)
arXiv:2603.22973 [cs.AI]
(or arXiv:2603.22973v3 [cs.AI] for this version)
https://doi.org/10.48550/arXiv.2603.22973 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 253 words
