Deontic Reasoning Benchmark for Legal Statutes

DeonticBench is a benchmark dataset for evaluating deontic reasoning over legal and regulatory statutes. It contains natural language case descriptions paired with ground-truth labels and reference Prolog programs encoding statutory rules. The dataset was created by author 'gydou' and last updated on April 3, 2026.

Use Cases

• Train models to predict deontic labels from natural language case descriptions.
• Evaluate the logical consistency of language models using the provided Prolog program references.
• Benchmark reasoning performance across five distinct legal and regulatory domains.
• Fine-tune models for statutory interpretation tasks using the case description and label pairs.

Strengths

• Covers five distinct legal and regulatory domains for broad evaluation.
• Provides ground-truth labels and reference Prolog programs for each case.

Limitations

• Unknown sample size and row count prevents assessment of statistical power.
• Specific column names and data structure are not detailed in the provided description.

Provenance

Source

Hugging Face dataset repository by author 'gydou'.

Collection Method

Constructed as a benchmark, pairing case descriptions with labels and logic programs.

Time Range

null

Freshness

null

Geography

null