nassimjp/afghanistan-post-2021-pashto-conversation-3x

🇦🇫 Afghanistan Post-2021 Pashto Conversation 3X

`nassimjp/afghanistan-post-2021-pashto-conversation-3x`

A Pashto conversational dataset focused on Afghanistan after 2021, designed for training and evaluating Pashto language models on multi-turn dialogue, answer diversity, contextual follow-up questions, and conversational continuity.

📌 Overview

This dataset is designed as a conversational extension of the Afghanistan Post-2021 Pashto Dataset.

Instead of providing only a single answer for each question, the dataset uses three response variations (3X) for each question.

The goal is to help Pashto language models learn to:

• —Understand questions in Pashto.
• —Generate natural and informative answers.
• —Express the same topic in different ways.
• —Maintain conversational context.
• —Continue a discussion naturally.
• —Suggest relevant follow-up questions.
• —Avoid producing repetitive, rigid responses.
• —Support multi-turn conversational interactions.

🧠 Dataset Concept

The core structure follows:

Question → Answer → Follow-up Questions

Each source question can have up to three different response variations.

Conceptually:

User Question
      ↓
Response A → Follow-up Questions
      ↓
Response B → Follow-up Questions
      ↓
Response C → Follow-up Questions

The three responses are intended to provide linguistic and conversational diversity, rather than simply repeating the same answer with minor word changes.

🎯 Main Topics

The dataset focuses on post-2021 Afghanistan and may include topics such as:

• —🇦🇫 Afghanistan's political situation
• —🧑‍🤝‍🧑 Society and social conditions
• —👩 Women's rights and restrictions
• —🎓 Education and girls' education
• —📰 Media and freedom of expression
• —🧑‍💼 Employment and economic conditions
• —💰 Poverty and humanitarian conditions
• —🏠 Displacement and migration
• —🌍 Afghan refugees
• —⚖️ Human rights
• —🛡️ Security and governance
• —🌐 International relations
• —🏥 Healthcare
• —📚 Culture and society
• —🚧 Humanitarian challenges
• —🔄 Post-2021 political and social developments

🏗️ Intended Use

This dataset is intended for:

• —Pashto LLM instruction tuning
• —Conversational SFT
• —Multi-turn dialogue training
• —Pashto response-diversity training
• —Follow-up question generation
• —Conversational AI research
• —Pashto language evaluation
• —Context-aware dialogue systems

It can be particularly useful for small and medium-sized Pashto language models that need to improve their ability to maintain a natural conversation rather than producing isolated question-answer pairs.

💬 Conversational Continuity

A key feature of this dataset is conversation continuation.

After answering a question, the model can provide one to three relevant questions that naturally continue the topic.

Example:

{
  "question": "په افغانستان کې د بیان ازادي څنګه محدوده شوې ده؟",
  "answer": "په افغانستان کې د بیان د آزادۍ وضعیت د رسنیو پر فعالیتونو، خبریالانو او د عامه نظر پر څرګندولو د بېلابېلو محدودیتونو له امله اغېزمن شوی دی.",
  "follow_up_questions": [
    "د رسنیو وضعیت څنګه دی؟",
    "خبریالان له کومو ستونزو سره مخ دي؟",
    "په ټولنیزو رسنیو د دې محدودیتونو اغېز څه دی؟"
  ]
}

This structure encourages a model to behave more like a conversation partner rather than a static question-answer system.

🔥 Why 3X?

The 3X design provides three response variations for the same underlying question.

The objective is to expose the model to different:

• —Sentence structures
• —Vocabulary choices
• —Explanatory styles
• —Reasoning patterns
• —Levels of detail
• —Conversational continuations

This can reduce overfitting to a single fixed response pattern and encourage more flexible Pashto generation.

📊 Data Structure

Depending on the dataset version, records may contain fields such as:

id
question
answer
follow_up_questions
messages

For conversational training, the messages field can follow a structure such as:

{
  "messages": [
    {
      "role": "user",
      "content": "په افغانستان کې د بیان ازادي څنګه محدوده شوې ده؟"
    },
    {
      "role": "assistant",
      "content": "..."
    }
  ]
}

⚠️ Data and Factuality

This dataset concerns politically and socially sensitive topics.

The dataset should therefore be treated as training data, not as an authoritative historical or factual reference.

Some answers may reflect generated explanations and should be independently verified before being used as factual evidence.

For time-sensitive claims, users should verify:

• —Date
• —Location
• —Source
• —Population affected
• —Whether the information is current

When used for model training, additional factual filtering and human or automated quality control are recommended.

🧹 Recommended Quality Control

Before using the dataset for final training, it is recommended to check for:

• —Duplicate questions
• —Repetitive answers
• —Irrelevant follow-up questions
• —Incorrect Pashto grammar
• —Persianized or unnatural wording
• —Unsupported factual claims
• —Contradictory answers
• —Excessive verbosity
• —Hallucinated events or statistics
• —Poor conversational continuity

🧪 Recommended Training Use

For SFT, the conversational messages format can be used directly where supported.

A recommended pipeline is:

Source Questions
      ↓
3 Response Variations
      ↓
Follow-up Questions
      ↓
Quality Filtering
      ↓
Deduplication
      ↓
Pashto Language Filtering
      ↓
SFT / Instruction Tuning
      ↓
Evaluation

📜 License

This dataset is released under the Creative Commons Attribution 4.0 International (CC BY 4.0) license, subject to the provenance and licensing of the underlying source material.

Users are responsible for ensuring that their downstream use complies with applicable copyright, licensing, privacy, and other legal requirements.

👤 Creator

Nassim / nassimjp

Hugging Face:

nassimjp

🤖 Project Goal

The broader goal of this project is to contribute high-quality Pashto resources for open-source language-model development, with particular attention to conversational ability, contextual understanding, and natural Pashto generation.

The dataset is part of an ongoing effort to expand the availability of Pashto-language resources for researchers and developers.

⭐ Citation

If you use this dataset in research or development, please reference:

Nassim — Afghanistan Post-2021 Pashto Conversation 3X
Hugging Face: nassimjp/afghanistan-post-2021-pashto-conversation-3x

🤝 Contributions

Suggestions, corrections, quality improvements, and additional Pashto conversational resources are welcome.

Pashto AI development starts with Pashto data. 🇦🇫🌱🤖