Elamur Research⌕
☰
home / papers / 2026 / An LLM-Based Assistant for Debt Waiver Court Procedures

An LLM-Based Assistant for Debt Waiver Court Procedures

Summary

Since Spain introduced its "fresh start" debt waiver procedure (RLD 1/2020), courts have been flooded with applications. Annual cases rose from about 12,000 in 2020 to over 57,000 in 2024, and each one requires staff to read through a large, inconsistent bundle of documents. The authors built a prototype assistant within the EU project JuLIA to ease this load.

The back-end works in four stages:

  • adaptive PDF-to-text conversion, combining text extraction and OCR depending on document type;
  • rule-based classification of the filed documents;
  • few-shot information extraction with self-hosted open-source LLMs (commercial models were excluded for privacy reasons);
  • post-processing that merges duplicated or partial records.

A front-end then shows judges a structured case overview with links back to the source files and can produce draft rulings for them to approve or edit.

On 185 real, anonymized Catalan proceedings, classification exceeded 92% F1 on test data and personal-data extraction reached about 91% F1. Debt and asset lists proved harder. The work shows that locally run LLMs can realistically support routine judicial tasks while the judge still makes the final decision.

Key points

  • Dataset: 185 completed proceedings, each with 10–50 PDF documents, supplied by the Catalan Department of Justice and anonymized externally. They were split into 10 applications for few-shot examples, 82 for development and 93 for testing, all with hand-built gold JSON annotations.
  • Document classifier: F1 of 95.3 on development and 92.1 on test, with precision 90.3 and recall 94.0 on test.
  • Extraction F1 on test (best few-shot configuration): personal data 91.1, tax returns 81.4, creditor list 77.3, asset list 75.0, average 79.0 (85.5 on development).
  • Application-level results: 12.9% of test applications were extracted perfectly and 60% scored at least 80% F1, compared with 79.2% on development. The authors attribute the drop partly to classifier errors reducing recall and to noisier test data.
  • PDF tools: unstructured (plain and OCR modes), pdftotext and PyMuPDF4llm.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/57sfssjpk6di

@inproceedings{padro2026llm,
  title = {{An LLM-Based Assistant for Debt Waiver Court Procedures}},
  author = {Padro, Lluis and Ferrés, Daniel and Saurí, Roser and Artigot, Mireia},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/57sfssjpk6di},
  url = {https://doi.org/10.63317/57sfssjpk6di}
}

References

20 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdullah Al Zubaer, Michael Granitzer, and Jelena Mitrović. 2023. Performance analysis of large language models in the domain of legal argument mining. Frontiers in Artificial Intelligence, Volume 6 - 2023. find
  2. Farid Ariai and Gianluca Demartini. 2024. Natural language processing for the legal domain: A survey of tasks, datasets, models, and challenges. find
  3. Valerio Bellandi, Christian Bernasconi, Fausto Lodi, Matteo Palmonari, Riccardo Pozzi, Marco Ripamonti, and Stefano Siccardi. 2024. An entity-centric approach to manage court judgments based on natural language processing. Computer Law & Security Review, 52:105904. find
  4. Odysseas Chlapanis, Dimitris Galanis, and Ion Androutsopoulos. 2024. Lar-echr: A new legal argument reasoning task and dataset for cases of the european court of human rights. pages 267–279. find
  5. Consejo General del Poder Judicial. 2025. Efecto de la crisis en los órganos judiciales. https://www.poderjudicial.es/cgpj/es/ ↗
  6. Junyun Cui, Xiaoyu Shen, Feiping Nie, Zheng Wang, Jinglong Wang, and Yulong Chen. 2022. A survey on legal judgment prediction: Datasets, metrics, models and challenges. find
  7. EU Council. 2019. Directive (EU) 2019/1023 of the European Parliamment and of the Council. https://eur-lex.europa.eu/eli/dir/2019/1023/oj/eng. ↗
  8. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, and 557 other authors. 2024. The llama 3 herd of models. https://arxiv.org/abs/2407.21783. ↗
  9. Paula L. Hannaford-Agor. 2023. Use of natural language processing (nlp) in civil case management: A report on three proof of concept projects. find
  10. Albert Q. Jiang, Alexandre Sablayrolles, Arthur Mensch, and 14 other authors. 2023. Mistral 7b. https://arxiv.org/abs/2310.06825. ↗
  11. H. W. Kuhn. 1955. The hungarian method for the assignment problem. Naval Research Logistics Quarterly, 2(1-2):83–97. find
  12. Elena Leitner, Georg Rehm, and Julian Moreno-Schneider. 2020. A dataset of German legal documents for named entity recognition. In Proceedings of the Twelfth Language Resources and Evaluation Conference, pages 4478–4485, Marseille, France. European Language Resources Association. find
  13. Drish Mali, Rubash Mali, and Claire Barale. 2024. Information extraction for planning court cases. In Proceedings of the Natural Legal Language Processing Workshop 2024, pages 97–114, Miami, FL, USA. Association for Computational Linguistics. find
  14. Bruno Mathis. 2022. Extracting proceedings data from court cases with machine learning. Stats, 5(4):1305–1320. find
  15. Ramesh Nallapati and Christopher D. Manning. 2008. Legal docket-entry classification: where machine learning stumbles. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, EMNLP’08, page 438–446. Association for Computational Linguistics. find
  16. Joel Niklaus, Robin Mamié, Matthias Stürmer, Daniel Brunner, and Marcel Gygli. 2023. Automatic anonymization of Swiss federal Supreme Court rulings. In Proceedings of the Natural Legal Language Processing Workshop 2023, pages 159–165, Singapore. Association for Computational Linguistics. find
  17. Giulio Salierno, Rosamaria Bertè, Luca Attias, Carla Morrone, Dario Pettazzoni, and Daniela Battisti. 2024. Giusberto: A legal language model for personal data de-identification in italian court of auditors decisions. find
  18. Holli Sargeant, Ahmed Izzidien, and Felix Steffek. 2025. Topic classification of case law using a large language model and a new taxonomy for uk law: Ai insights into summary judgment. Artificial Intelligence and Law. find
  19. Spanish Government. 2015. Ley 25/2015, de 28 de julio, de mecanismo de segunda oportunidad, reducción de la carga financiera y otras medidas de orden social [Law 25/2015, on the second chance mechanism, reduction of financial burden and other social measures]. Official State Gazette (BOE) No. 180 of… find
  20. Spanish Government. 2020. Real Decreto Legislativo 1/2020, de 5 de mayo, por el que se aprueba el texto refundido de la Ley Concursal [Royal Legislative Decree 1/2020 of 5 May 2020 approving the Consolidated Text of the Insolvency Law]. Official State Gazette (BOE) No. 127 of 7th of May 2020. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/padro-2026-llm-based-assistant-debt-waiver/