Guillaume Zambrano

Université de Nîmes · CHROME Laboratory (EA 7352) · Law, Economics and Management
My research applies natural language processing and machine learning to large corpora of French judicial decisions, with the goal of understanding and modelling how individual judges make decisions.

Research

My work sits at the intersection of legal informatics, natural language processing, and applied machine learning. I study whether individual judges exhibit stable, learnable decision-making patterns that diverge measurably from aggregate court behaviour. This research programme has two pillars: developing reliable methods for extracting structured data from unstructured judicial text, and training specialist predictive models that capture judge-specific decisional profiles.

My most recent study, The Judge Variable (2025), uses 18,937 living arrangements rulings from 10,306 child custody cases in French appellate courts. Specialist models trained on individual judges' past rulings achieve F1 scores as high as 92.85%, compared to 82.63% for a generalist model trained on 20 to 100 times more data. In-domain and cross-domain validity tests provide empirical support for the legal realist hypothesis that judicial identity plays a measurable role in case outcomes.

Legal NLP Judicial Prediction Legal Realism LLM Data Extraction Prompt Engineering Feature Engineering Open Data (Judilibre) French Case Law

Publications

The Judge Variable: Challenging Judge-Agnostic Legal Judgment Prediction
G. Zambrano · arXiv preprint, 2025
arXiv PDF
Case Law as Data: Prompt Engineering Strategies for Case Outcome Extraction with Large Language Models in a Zero-Shot Setting
G. Zambrano · Law, Technology and Humans, 6(3), pp. 80–101, 2024
DOI HAL
Article 700 Identification in Judicial Judgments: Comparing Transformers and Machine Learning Models
S.A. Mahmoudi, C. Condevaux, G. Zambrano, S. Mussard · Stats (MDPI), 7(4), pp. 1421–1436, 2024
DOI HAL
Scribe: A Specialized Collaborative Tool for Legal Judgment Annotation
S.A. Mahmoudi, G. Zambrano, C. Condevaux, S. Mussard · JURIX 2022, pp. 290–293
HAL
NER sur décisions judiciaires françaises : CamemBERT Judiciaire ou méthode ensembliste ?
S.A. Mahmoudi, C. Condevaux, B. Mathis, G. Zambrano, S. Mussard · EGC 2022, pp. 281–288
HAL
Linking Appellate Judgments to Tribunal Judgments: Benchmarking Different ML Techniques
C. Condevaux, B. Mathis, S.A. Mahmoudi, S. Mussard, G. Zambrano · JURIX 2022, pp. 33–42
DOI
Identification of Judicial Outcomes in Judgments: A Generalized Gini-PLS Approach
G. Tagny-Ngompé, S. Mussard, G. Zambrano, S. Harispe, J. Montmain · Stats (MDPI), 3(4), pp. 427–443, 2020
DOI HAL
Generalized Gini Linear and Quadratic Discriminant Analyses
G. Zambrano, C. Condevaux, S. Mussard, T. Ouraga · METRON, 78(2), 2020
DOI
Weakly Supervised One-shot Classification using Recurrent Neural Networks with Attention: Application to Claim Acceptance Detection
C. Condevaux, S. Harispe, S. Mussard, G. Zambrano · JURIX 2019, Madrid, pp. 23–32
HAL
Detecting Sections and Entities in Court Decisions Using HMM and CRF Graphical Models
G. Tagny Ngompé, S. Harispe, G. Zambrano, J. Montmain · EGC 2018 (Best of Volume), pp. 61–86
HAL

Tools & Data

LILA: Litigation Data Extraction with LLMs

Datasets and code for the prompt engineering study on zero-shot legal data extraction from French court decisions.

OSF Repository

Papers with Code & Data

Replication packages for published studies will be made available here as they are prepared for open access.

Contact

guillaume.zambrano [at] unimes.fr

arXiv · HAL · ResearchGate · dblp · Academia.edu · Portail Universitaire du Droit