README.md

---
language:
- en
- ar
license: apache-2.0
task_categories:
- question-answering
- translation
- text-generation
tags:
- agriculture
- farming
- egypt
- quickmt
- slm-finetuning
size_categories:
- 10K<n<100K
source_datasets:
- KisanVaani/agriculture-qa-english-only
author: Abdulhamed Eid
pretty_name: Agriculture QA English-Arabic Pairs
dataset_info:
  features:
  - name: question
    dtype: string
  - name: answers
    dtype: string
  - name: question_ar
    dtype: string
  - name: answers_ar
    dtype: string
  splits:
  - name: train
    num_bytes: 12358367
    num_examples: 22615
  download_size: 813265
  dataset_size: 12358367
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
---

# Agriculture QA (English & Arabic) Pairs

This dataset contains **22,600+** Question and Answer pairs related to agriculture, farming techniques, and crop management. It combines the original English data from [KisanVaani/agriculture-qa-english-only](https://huggingface.co/datasets/KisanVaani/agriculture-qa-english-only) with high-quality Arabic translations generated and curated by **Abdulhamed Eid**.

It was created to fine-tune **Small Language Models (SLMs)** for meaning alignment between English and Arabic, specifically to build AI agents for **Egyptian farmers** and agricultural shops.

## Dataset Structure

The dataset contains the following columns:

| Column | Type | Description |
|:-------|:-----|:------------|
| `question` | string | The original question in English. |
| `answers` | string | The original answer in English. |
| `question_ar` | string | The translated question in Arabic. |
| `answers_ar` | string | The translated answer in Arabic. |

## Creation Process

### Translation Methodology
To ensure high semantic accuracy and context preservation, the translation was performed using **[quickmt/quickmt-en-ar](https://huggingface.co/quickmt/quickmt-en-ar)** (a Transformer-Big model).

Instead of translating columns in isolation, a **Context-Aware Batching** strategy was used:
1. The Question and Answer were combined into a single prompt separated by a newline (`\n`).
2. This allowed the Neural Machine Translation model to see the context of the answer while translating the question (and vice versa), reducing ambiguity.
3. The output was then rigorously split back into `question_ar` and `answers_ar` using a robust parsing pipeline.

### Data Quality
- **Source:** 22.6k rows from the KisanVaani dataset.
- **Verification:** Post-translation filtering was applied to ensure no rows were corrupted or merged during the batching process. The dataset is free of structural parsing errors.

## Intended Use

This dataset is ideal for:
1. **Fine-tuning SLMs:** Training small models (Llama, Mistral, Qwen) to understand agricultural terminology in both Arabic and English.
2. **Semantic Alignment:** Helping models map English agricultural concepts to their Arabic equivalents.
3. **Agri-Tech Chatbots:** Building Q&A agents for farmers or assistants for agricultural supply shops.

## Sample

```json
{
  "question": "What are the benefits of drip irrigation?",
  "answers": "Drip irrigation saves water and fertilizer by allowing water to drip slowly to the roots of plants.",
  "question_ar": "ما هي فوائد الري بالتنقيط؟",
  "answers_ar": "الري بالتنقيط يوفر المياه والأسمدة عن طريق السماح للماء بالتنقيط ببطء إلى جذور النباتات."
}
```

## Limitations

- **Machine Translation:** While QuickMT is a high-quality NMT model, the translations are machine-generated and have not been manually verified by human linguists.
- **Regional Dialect:** The translation is in Modern Standard Arabic (MSA). While understandable by Egyptian farmers, it does not use the Egyptian colloquial dialect (*Ammiya*).
- **Source Context:** The original dataset (KisanVaani) may contain farming practices specific to India or general global practices. Users should verify specific pesticide/fertilizer recommendations for Egyptian soil and regulations.

## Citation

If you use this dataset in your work, **please cite both the creator of this En-Ar version and the original data source.**

### Cite this Dataset (Arabic/English Version)
```bibtex
@dataset{eid2025agriculture,
  author       = {Eid, Abdulhamed},
  title        = {Agriculture QA English-Arabic Pairs},
  year         = {2025},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/abdulhamed/agriculture_qa_en_ar_pairs}},
  note         = {Fine-tuning dataset for SLM alignment in Egyptian Agriculture}
}
```

### Cite the Original Data Source
```bibtex
@dataset{kisanvaani2024,
  author       = {KisanVaani},
  title        = {Agriculture QA English Only},
  publisher    = {Hugging Face},
  howpublished = {\url{https://huggingface.co/datasets/KisanVaani/agriculture-qa-english-only}}
}
```

### Acknowledgements
Translation performed using the **QuickMT** library and models.