NETHERLANDSResearch
arXiv: University of Amsterdam team models Dutch insolvency avoidance rule with neuro-symbolic AI
University of Amsterdam researchers built a neuro-symbolic model of the actio pauliana in Article 42 of the Dutch Bankruptcy Act, which lets insolvency practitioners undo transactions that prejudice creditors. A rules core written in the Catala language encodes the statute, while language models resolve open-textured terms using retrieved Dutch civil judgments. On ten hand-annotated cases it handled most criteria but struggled with the knowledge requirement.
Key points
- Language models turn a case's narrative facts into the true-or-false and date inputs the symbolic solver needs.arXiv
- Segmenting and labeling judgments reached high precision and recall against a manually annotated set of 100 judgments.arXiv
- Retrieving more judgments had no clear quantitative effect, though retrieved cases gave explicit grounding in prior case law.arXiv
- The Catala code is written in a literate style that keeps the statutory text and its formalization aligned.arXiv
- The paper has been accepted to JURIX 2026.arXiv