metadata
license: apache-2.0
language:
- ps
- en
- multilingual
tags:
- pashto
- pashto-ai
- قانوني
- legal
- legal-reasoning
- multilingual-reasoning
- chain-of-thought
- law
size_categories:
- 1K<n<10K
Legal Issues Multilingual Reasoning
Dataset Summary
Legal Issues Multilingual Reasoning is a specialized, high-quality dataset designed for training and evaluating Large Language Models (LLMs) on complex legal analysis, step-by-step logical chain-of-thought (CoT) reasoning, and multilingual legal problem-solving.
The dataset contains structured samples featuring system prompts, user queries, reasoning steps, and comprehensive legal answers adapted for specialized jurisdictions and languages (such as Pashto (ps) and others).
Dataset Structure
Data Fields
Each example in the JSONL files is structured with a conversation-style message format containing the following roles:
messages: A list of message objects consisting of:role("system"): Establishes the persona of an experienced legal analyst and expert skilled in statutory interpretation, regulatory analysis, and complex legal breakdown.role("user"): Presents a specific legal query, scenario, or question regarding laws, legal aid organizations, rights, or procedures.role("assistant"): Contains the model's structured response, which includes:<reasoning>: Step-by-step internal analytical breakdown and evaluation of the legal issue.<answer>: The final, comprehensive response delivered to the user.
Example Instance
{
"messages": [
{
"role": "system",
"content": "تاسو یو تجربه لرونکی حقوقي شنونکی او حقوقي پوه یئ چې د مرحلې په واسطه د قوانینونو تشریح، د قضایي مقرراتو تحلیل، او د سختو حقوقي مسایلو تجزیه کولو کې مهارت لرئ."
},
{
"role": "user",
"content": "په عموم کې د حقوقي مرستې سازمانونو له لارې کوم ډول حقوقي مرستې وړاندیز کیږي؟"
},
{
"role": "assistant",
"content": "<reasoning>\nد دې مؤثره مقابلو لپاره، ما اړتیا درلوده چې دوه اصلي برخې حل کړم...\n</reasoning>\n\n<answer>\nد قانوني مرستې سازمانونه عموماً د مختلفو قانوني مرستې خدمات وړاندې کوي...\n</answer>"
}
]
}
Source and Processing
- Data Cleaning: Unwanted metadata tags (such as raw IDs and specific language tags embedded at the root level) have been stripped to align with standard conversational training frameworks.
- Use Cases: Fine-tuning open-source LLMs (such as Llama, Mistral, or Qwen) for legal domain adaptation, chain-of-thought reasoning alignment, and low-resource multilingual legal assistant deployment.
Citation
If you use this dataset in your research or applications, please cite it appropriately:
@dataset{legal_issues_multilingual_reasoning,
author = {Nassim JP},
title = {Legal Issues Multilingual Reasoning},
year = {2026},
publisher = {Hugging Face},
url = {[https://huggingface.co/datasets/nassimjp/Legal-Issues-Multilingual-Reasoning](https://huggingface.co/datasets/nassimjp/Legal-Issues-Multilingual-Reasoning)}
}