nassimjp/pashto-sociology

# Dataset Card for Pashto Sociology Dataset

## Dataset Description

- **Homepage:** [N/A]
- **Repository:** [Nassimjp/pashto-sociology](https://huggingface.co/datasets/nassimjp/pashto-sociology)
- **Paper:** [N/A]
- **Leaderboard:** [N/A]
- **Point of Contact:** [N/A]

### Dataset Summary

This dataset contains a collection of 100 sociological dialogue samples in the Pashto language. It is designed to facilitate research and development of conversational AI, natural language understanding, and cultural studies. The dialogues cover a wide range of topics related to Pashtun society, including tribal systems, cultural practices, social issues, and contemporary challenges.

The dataset is structured in JSONL (JSON Lines) format, making it easy to load and process for various machine learning and NLP tasks. Each entry simulates a question-and-answer interaction between a user and an AI assistant, focusing on providing informative and culturally aware responses.

### Supported Tasks and Leaderboards

- **Tasks:** Conversational AI, Question Answering, Social Science Research, Cultural NLP
- **Leaderboards:** Not applicable.

### Languages

The dataset is exclusively in **Pashto** (ps). The script primarily used is the Pashto alphabet (Arabic script).

## Dataset Structure

The dataset is provided as a single JSONL file where each line is a JSON object with the following fields:

- **`instruction`**: A string containing the user's question or prompt.
- **`input`**: An optional string providing additional context. This field is left empty (`""`) in all samples.
- **`output`**: A string containing the AI-generated response to the instruction.
- **`metadata`**: An object containing two keys:
  - **`platform`**: A string indicating the source or type of interaction (e.g., "پوښتنه او ځواب" which means "Question and Answer").
  - **`topic`**: A string categorizing the main topic (e.g., "ټولنپوهنه" which means "Sociology").
  - **`subtopic`**: A string providing a more specific topic (e.g., "جیرګه", "ښځې").

### Example

```json
{"instruction": "د پښتنو په ټولنه کې د جیرګې سیسټم څومره اغېز لري؟", "input": "", "output": "جیرګه د پښتنو په ټولنه کې یو بنسټیز او مهم رول لري. دا سیسټم د شخړو حل، د ټولنیزو ستونزو د هواري او د قبیلوي تصمیمونو د نیولو لپاره کارول کیږي. جیرګه د پښتنو د ټولنیز ژوند یوه مهمه برخه ده چې د پخوانیو وختونو څخه تر اوسه پورې دوام لري او د ټولنې په ثبات کې مهم رول لري.", "metadata": {"platform": "پوښتنه او ځواب", "topic": "ټولنپوهنه", "subtopic": "جیرګه"}}

Data Splits

The dataset contains a single split, presumably the training set, with 100 samples.

Split	Samples
train	100

Dataset Creation

Curation Rationale

This dataset was created to address the lack of structured, sociological conversational data in the Pashto language. It aims to provide a resource that can be used for:

1. Training and evaluating conversational AI models that are culturally sensitive and knowledgeable about Pashtun society.
2. Supporting academic research in sociology, anthropology, and linguistics related to the Pashtun culture.
3. Developing educational tools and information systems for Pashto-speaking communities.

Source Data

The data was generated by a language model, drawing upon a broad understanding of Pashtun culture, history, and social structures.

Initial Data Collection and Normalization

The generation process was guided by prompts focused on key sociological themes within Pashtun society, such as the Jirga system, Pashtunwali, gender roles, and contemporary issues. No external raw text data was used.

Who are the source language producers?

The responses were synthetically generated by a large language model based on a curated list of sociological topics.

Annotations

Annotation process

No manual annotation was performed on the generated text. The focus was on structuring the data in a clean, machine-readable format.

Who are the annotators?

The data was generated synthetically, so there are no human annotators for the content.

Personal and Sensitive Information

The dataset does not contain any personal or identifiable information, as it consists entirely of general, synthetic dialogues about societal topics.

Considerations for Using the Data

Social Impact of Dataset

This dataset can serve as a valuable tool for promoting language technology in a low-resource language and fostering better AI systems for Pashto-speaking users. It can also aid in preserving and disseminating knowledge about Pashtun culture.

Discussion of Biases

While the dataset covers a range of perspectives, it is important to note the following:

1. Synthetic Nature: The data is generated by an AI, which may reflect inherent biases from its training data.
2. Cultural Representation: The views expressed represent general sociological overviews and may not encompass the full diversity of opinions within Pashtun communities worldwide.
3. Language Nuance: The generated text may lack the full richness, dialectal variation, and idiomatic expression of native Pashto.

Other Known Limitations

• The dataset is relatively small (100 samples).
• The "input" field is always empty, limiting its use for tasks requiring complex multi-turn dialogues or external context.
• All metadata topics are currently labeled as "ټولنپوهنه" (Sociology).

Additional Information

Dataset Curators

The dataset was curated and uploaded by nassimjp.

Licensing Information

This dataset is licensed under the Apache License 2.0.

Citation Information

If you use this dataset, please consider citing it by referencing its Hugging Face repository:

@misc{pashto-sociology-dataset,
  title = {Pashto Sociology Dataset},
  author = {nassimjp},
  year = {2026},
  publisher = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/nassimjp/pashto-sociology}}
}

Contributions

Contributions to expand or improve the dataset, such as adding more samples or correcting errors, are welcome. Please contact the dataset owner via Hugging Face.