JudgeMeNot: Hebrew Court Rulings and QA Pairs for Judicial AI

Israeli court decisions in Modern Hebrew, consisting of sole-judge rulings. The dataset provides materials organized by judge, including question-answer pairs and raw ruling text segments. It serves as replication data for the ACL 2026 paper 'JudgeMeNot: Personalizing Large Language Models to Emulate Judicial Reasoning in Hebrew' and was uploaded to Harvard Dataverse on 2026-04-20.

Use Cases

• Train instruction-tuned language models based on the provided legal question-answer pairs.
• Evaluate LLM performance on judicial reasoning tasks using the defined train/evaluation/test splits.
• Perform causal language modeling of a judge's writing style based on the raw ruling text segments.

Strengths

• Data is structured by individual judge, enabling personalized modeling.
• Provides two complementary formats: QA pairs for instruction tuning and raw text for continuation modeling.
• Includes explicit train, evaluation, and test split labels for standardized benchmarking.

Limitations

• Row count, file formats, and column-level documentation are unknown, requiring manual inspection after download.
• Data scope is limited to Israeli sole-judge rulings, which may not represent multi-judge panels or other legal systems.
• License information is unavailable, which may restrict usage.

Provenance

Source

Harvard Dataverse, uploaded by RAZUMENKO, ITAY.

Collection Method

Extracted from Israeli court decisions where a single judge decided the case.

Time Range

null

Freshness

Last updated 2026-04-20 14:47:29.

Geography

Israel