pkheria/indian-courtroom-objections-100k

Indian Courtroom Objections Dataset (100K)

A synthetic dataset of 100,000 Indian courtroom objection scenarios for training AI judge models. Each example contains a case premise, opposing lawyer's question, an objection with legal reasoning, and the judge's ruling (SUSTAINED or OVERRULED) with a one-line legal explanation.

Dataset Structure

Splits

• —train: 95,000 examples
• —test: 5,000 examples

Format (Conversational)

Each example follows the TRL SFTTrainer messages format:

{
  "messages": [
    {
      "role": "system",
      "content": "You are an experienced Indian courtroom judge..."
    },
    {
      "role": "user",
      "content": "Premise: The State of Maharashtra v. Ramesh Kumar...\n\nOpposing Lawyer: \"...\"\n\nObjection: ...\nReason: ..."
    },
    {
      "role": "assistant",
      "content": "SUSTAINED. The question is irrelevant and does not relate to any fact in issue."
    }
  ]
}

Objection Taxonomy (13 Types)

1. 1.Relevance - Questions unrelated to facts in issue
2. 2.Leading Question - Questions suggesting the answer
3. 3.Hearsay - Out-of-court statements offered for truth
4. 4.Speculation - Calls for conjecture beyond personal knowledge
5. 5.Compound Question - Double-barreled questions
6. 6.Badgering - Argumentative and harassing questions
7. 7.Privilege - Protected communications (lawyer-client, spousal, official)
8. 8.Character Evidence - Using bad character to prove conduct
9. 9.Self-Incrimination - Compelling witness to confess crime
10. 10.Opinion - Non-expert opinion evidence
11. 11.Improper Impeachment - Attacking credibility without foundation
12. 12.Asked and Answered - Repetitive questioning
13. 13.Assumes Facts Not in Evidence - Questions with false premises

Legal Framework

All examples are grounded in:

• —Indian Evidence Act, 1872 (Sections 5, 8, 14, 45, 52-55, 60, 122, 124, 126, 141, 145, 146, 154, 159)
• —Indian Penal Code, 1860 (Sections 147-148, 302-304, 323-326, 376, 379-381, 392, 406-409, 420, 463-465, 499-500, 120B, 124A)
• —Code of Criminal Procedure, 1973
• —Constitution of India (Article 20(3))

Dataset Distribution

• —SUSTAINED: ~61,079 (61.1%)
• —OVERRULED: ~33,921 (38.9%)

Usage with TRL SFTTrainer

from datasets import load_dataset
from trl import SFTTrainer, SFTConfig

dataset = load_dataset("pkheria/indian-courtroom-objections-100k")

trainer = SFTTrainer(
    model="unsloth/llama-3-8b-Instruct-bnb-4bit",
    train_dataset=dataset["train"],
    args=SFTConfig(max_length=2048),
)
trainer.train()

Full Metadata

The train_full.jsonl and test_full.jsonl files contain additional fields:

• —id: Example ID
• —obj_type: Objection category
• —ipc_context: Criminal offence context
• —ruling: SUSTAINED or OVERRULED
• —premise, question, objection, reason, explanation

License

This dataset is released for research and educational purposes.

<!-- ml-intern-provenance -->

Generated by ML Intern

This dataset repository was generated by ML Intern, an agent for machine learning research and development on the Hugging Face Hub.

• —Try ML Intern: https://smolagents-ml-intern.hf.space
• —Source code: https://github.com/huggingface/ml-intern