nassimjp/pashto-ethics-based-reasoning

Pashto Ethics-Based Reasoning

A Pashto-language dataset for ethical reasoning, moral judgment, and responsible AI responses.

Overview

Pashto Ethics-Based Reasoning is a dataset designed to support the development and evaluation of AI models capable of understanding ethical principles, analyzing social situations, and generating responsible responses in the Pashto language.

The dataset focuses on ethical reasoning grounded in human values, fairness, justice, professional responsibility, and respect for human dignity.

Its primary goal is to contribute to the development of more capable Pashto-language AI systems and to strengthen Pashto representation in AI reasoning research.

Objectives

The dataset aims to:

• Improve ethical reasoning capabilities in Pashto-language models.
• Encourage fair, balanced, and responsible responses.
• Improve understanding of ethical principles and human values.
• Support analysis of moral dilemmas and social scenarios.
• Develop Pashto-language instruction-following capabilities.
• Contribute to research on low-resource language AI.
• Provide training and evaluation material for Pashto NLP and LLM development.

Dataset Structure

The dataset uses a JSONL format with conversational messages and associated metadata.

Each record contains:

Field	Description
messages	Conversation containing system, user, and assistant messages
meta	Sample metadata, including sample identifiers and generation information

Message roles

• system: Defines the assistant's ethical and linguistic behavior.
• user: Contains an ethical question, scenario, or request.
• assistant: Contains the model's response.

Reasoning format

Some assistant responses include a <think> section followed by the final answer.

This structure separates the reasoning text included in the dataset from the final user-facing response.

Important: The presence of a reasoning section does not independently establish the correctness or quality of the reasoning.

Example

{
  "messages": [
    {
      "role": "system",
      "content": "ته د پښتو ژبې یو ماهر، اخلاقي او مسلکي ډیجیټل مرستیال یې."
    },
    {
      "role": "user",
      "content": "د عدالت په اړه خپل نظر بیان کړه."
    },
    {
      "role": "assistant",
      "content": "<think>\nد عدالت مفهوم باید د انصاف، برابرۍ او انساني کرامت له مخې وڅېړل شي.\n</think>\n\nعدالت هغه اصل دی چې د هر انسان حقونه خوندي کوي..."
    }
  ],
  "meta": {
    "q_idx": 0,
    "sample_id": 0,
    "attempt": 1
  }
}

Illustrative example of the dataset format.

Language

• Primary language: Pashto
• ISO 639-1: ps
• Script: Arabic-based Pashto script

The dataset prioritizes natural Pashto responses and the development of Pashto-language AI capabilities.

Intended Use

This dataset may be useful for:

• Pashto Large Language Model (LLM) training.
• Supervised Fine-Tuning (SFT).
• Ethical reasoning research.
• Instruction-following experiments.
• Conversational AI development.
• Evaluation of Pashto-language model responses.
• Research into low-resource language representation.

Data Quality and Limitations

Ethical reasoning and language quality are different dimensions of model performance.

Generated responses may contain:

• Linguistic and grammatical errors.
• Incorrect or inconsistent terminology.
• Incomplete ethical analysis.
• Factual inaccuracies.
• Context-dependent interpretations.

The dataset should therefore be reviewed and evaluated before being used as a high-quality gold-standard training corpus.

For legal, medical, and other specialized subjects, responses should be independently verified against relevant professional standards and applicable jurisdictions.

Recommended Evaluation Criteria

We recommend evaluating model outputs across the following dimensions:

Evaluation dimension	Description
Ethical consistency	Alignment with fundamental human values
Reasoning quality	Logical coherence and depth of analysis
Pashto fluency	Naturalness and grammatical accuracy
Factual accuracy	Correctness of information
Fairness	Absence of unjustified bias or discrimination
Instruction following	Compliance with the requested task

Loading the Dataset

Using Hugging Face datasets:

from datasets import load_dataset

dataset = load_dataset(
    "nassimjp/pashto-ethics-based-reasoning",
    data_files="data/ethics_responses_1.jsonl",
    split="train"
)

print(dataset[0])

Research and Development

This dataset is part of ongoing efforts to expand Pashto-language AI research and develop models that can reason about ethical, social, and human-centered questions.

It can serve as a starting point for further dataset curation, linguistic refinement, model training, and comparative evaluation.

Author

Nasibullah Nassim

Pashto AI Educator and Researcher

Project: iPashto.ai

Hugging Face: @nassimjp

License

This dataset is released under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license.

You are free to share and adapt the dataset, provided that appropriate credit is given and derivative works are distributed under the same license.

License: https://creativecommons.org/licenses/by-sa/4.0/

Citation

If you use this dataset in your research or projects, please cite it as follows:

@dataset{nasim_pashto_ethics_reasoning,
  author = {Nasibullah Nasim},
  title = {Pashto Ethics-Based Reasoning},
  publisher = {Hugging Face},
  year = {2026},
  url = {https://huggingface.co/datasets/nassimjp/pashto-ethics-based-reasoning}
}

---

Building better reasoning in Pashto.

iPashto.ai — Advancing Pashto Language AI.