axmeeabdhullo/axya-tech-qa

Axya Tech QA Dataset

Dataset Summary

The Axya Tech QA Dataset is a comprehensive bilingual question-answering and translation dataset focused on the Dhivehi (Divehi) language. This dataset contains 1,000+ examples covering various types of language tasks including question-answering in Dhivehi about Maldives, bidirectional translation between English and Dhivehi, and conversational interactions. The dataset is designed to support the development of multilingual NLP models with a focus on the underrepresented Dhivehi language.

Dataset Structure

Data Instances

Each instance in the dataset contains the following fields:

• instruction: The task instruction or prompt
• input: The input text or question
• output: The expected output or answer
• language: The target language for the output
• task_type: The type of task being performed

Data Fields

• instruction (string): Contains the task instruction, such as "Respond in Dhivehi to:" or "Translate the following English sentence to Dhivehi:"
• input (string): The input text, question, or sentence to be processed
• output (string): The correct output, answer, or translation
• language (string): The language of the expected output ("Dhivehi" or "English")
• task_type (string): The specific task category:
  • qa_dv: Question-answering in Dhivehi
  • translation_en_to_dv: English to Dhivehi translation
  • translation_dv_to_en: Dhivehi to English translation

Data Splits

The dataset contains a single split:

• train: ~1,000 examples

Source Data

Initial Data Collection and Normalization

The dataset appears to be curated specifically for educational and research purposes focusing on:

• Cultural knowledge about the Maldives
• Common conversational phrases and expressions
• Everyday vocabulary and sentence structures
• Bidirectional translation capabilities

Who are the source language producers?

The dataset contains content related to Dhivehi language and Maldivian culture, suggesting involvement of native Dhivehi speakers or language experts familiar with Maldivian cultural context.

Intended Use

Supported Tasks and Leaderboards

• Question Answering: Answering questions about Maldives and general topics in Dhivehi
• Machine Translation: Bidirectional translation between English and Dhivehi
• Text Generation: Generating responses in Dhivehi for various prompts
• Language Understanding: Understanding and processing Dhivehi text

Languages

• Primary: Dhivehi (dv) - The official language of the Maldives
• Secondary: English (en) - Used for translation pairs and instructions

Dataset Creation

Curation Rationale

This dataset was created to address the lack of high-quality bilingual resources for the Dhivehi language. It focuses on:

1. Preserving and digitizing Dhivehi language content
2. Enabling cross-lingual understanding between English and Dhivehi
3. Supporting the development of NLP tools for underrepresented languages
4. Providing cultural and factual information about the Maldives

Source Data

The dataset includes:

• Cultural and factual information about the Maldives
• Common phrases and everyday conversations
• Translation pairs covering various topics
• Question-answer pairs in both languages

Considerations for Using the Data

Social Impact of Dataset

This dataset contributes to the preservation and promotion of the Dhivehi language in digital spaces. It can help:

• Develop better language technologies for Dhivehi speakers
• Support language learning and education
• Preserve cultural knowledge and linguistic diversity
• Enable better cross-cultural communication

Discussion of Biases

Users should be aware that:

• The dataset may reflect cultural perspectives specific to the Maldives
• Translation quality may vary across different domains
• The dataset size is limited, which may affect model generalization

Other Known Limitations

• Limited domain coverage (primarily focused on cultural and general topics)
• Relatively small dataset size for deep learning applications
• May not cover all dialectal variations of Dhivehi

Additional Information

Dataset Curators

Dataset created and curated by axmeeabdhullo.

Licensing Information

This dataset is released under the Apache License 2.0.

Citation Information

@dataset{axya_tech_qa_2024,
  title={Axya Tech QA Dataset: A Bilingual Question-Answering and Translation Dataset for Dhivehi},
  author={axmeeabdhullo},
  year={2024},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/axmeeabdhullo/axya-tech-qa}
}

Contributions

Thanks to the contributors who helped create this dataset to support Dhivehi language processing and preservation efforts.

---

This dataset card was created to provide comprehensive information about the Axya Tech QA dataset. For questions or issues, please use the Community tab.