A Grammar of French Civil Cassation: A Ground-Level Dataset and Tool
This paper introduces a ground-level dataset of 121,536 ground–reasons pairs extracted from 86,464 French civil cassation rulings spanning 2016–2025. Using a deterministic, rule-based pipeline, the authors pair each legal complaint with the Court’s response and classify the pair into established doctrinal families, producing a descriptive grammar of civil cassation. The resource enables quantitative analysis of which grounds succeed, which legal texts are cited, and how both evolve over time, alongside an interactive consultation tool.
The paper makes a solid empirical contribution by lowering the unit of analysis from the ruling to the individual ground, exposing intra-decision variation that case-level labels hide. The rule-based pipeline is carefully validated against gold-standard legal annotations, and the resulting descriptive findings—such as the steady rise in partial cassations and summary rejections—are well-supported by the data. However, the work remains fundamentally descriptive: it does not advance predictive models or causal claims, and its utility depends heavily on the stability of French drafting conventions.
The dataset construction is methodically executed and transparent. The authors leverage Judilibre zoning to pair grounds with reasons, achieving 99.5% agreement on structural stages and strong inter-annotator reliability for doctrinal families (Cohen's $\kappa$ up to $0.96$). The decision to use deterministic rules rather than black-box models is well-motivated for the legal domain, yielding interpretable labels linked to Boré's canonical taxonomy. The decade-scale trends they surface, such as the twenty-seven-point rise in summary rejections, are genuinely novel and doctrinally significant.
The main limitation is the brittleness of the rule-based taxonomy: families are inferred from canonical wording, so unusual phrasing escapes detection and specific counts are lower bounds. The cascade architecture—where violation of the law serves as the residual category—risks systematically misclassifying rare families into the dominant bucket. Additionally, the paper is almost purely descriptive; while it frames the tool as helping lawyers choose grounds, there is no empirical evaluation of whether practitioners actually benefit from the interactive tool or the ground-level statistics. The scope is also narrow, excluding the criminal chamber and any rulings before 2016, which limits historical comparability.
The evidence largely supports the claims because the methodology is validated with blind, legally trained double annotation on 400 grounds. Comparisons to related work are fair: the authors correctly position themselves against case-level prediction work and rhetorical-role labeling, while acknowledging that their task is a form of relational argument mining. The baseline comparison against JuriBERT and CamemBERTv2 shows that fine-tuned models inherit the silver labels' blind spots, reinforcing the value of interpretable rules, though the zero-shot LLM results are underwhelming and the fine-tuning details are somewhat sparse.
Reproducibility is strong: the dataset is released on Zenodo, the code is on GitHub, and an interactive web tool is provided. The source decisions are open data via Judilibre and PISTE. The paper documents the regular-expression cascades (149 patterns for cassation families, 110 for rejection) and the annotation protocol in detail. However, the exact regular expressions are not shown in the paper, and hyperparameters for the fine-tuned baselines are only briefly noted (learning rate $2 \times 10^{-5}$, effective batch 32, four epochs). Independent reproduction would require consulting the released code to recover the full pattern grids and train-test splits.
This paper presents a dataset of 121,536 ground-reasons pairs across 86,464 rulings of the five civil chambers of the French Court of Cassation (2016-2025). An appeal on points of law cannot itself be appealed and often decides a case for good. Understanding how the Court rules is thus crucial and can help lawyers who draft those appeals. Inside a ruling, every ground of appeal (a distinct legal complaint against the appealed decision) has its own answer from the Court. This answer is called the reasons and it belongs to one of a few doctrinal families. Using a deterministic pipeline, we pair each ground with its reasons and attach the pair to its family, according to the established French taxonomy. We obtain a descriptive grammar of civil cassation that shows which grounds succeed, which legal texts they refer to, and how both evolve over the decade. We release the dataset, the code and an interactive consultation tool.
Pick a starting point or write your own. Challenges run in the background, so you can keep reading while the AI investigates.
No challenges yet. Disagree with the review? Ask the AI to revisit a specific claim.