metadata

license: apache-2.0
task_categories:
  - text-classification
  - token-classification
  - translation
  - text2text-generation
  - question-answering
tags:
  - dataset
  - balochi
  - baluchi
size_categories:
  - 100K<n<1M

Balochi Language Dataset

Overview

This dataset is a comprehensive resource for training large language models (LLMs) in the Balochi language. It is designed to go beyond basic translation tasks, supporting fully generative text and conversational AI capabilities in Balochi.

The dataset includes monolingual Balochi text, multilingual translation corpora, and various conversational and domain-specific texts, enabling diverse use cases such as:

Generative AI: Building chatbots and generative text systems in Balochi.
Multilingual Translation: Translating between Balochi and other languages like English, Urdu, and Persian.
Cultural Insights: Providing context for cultural and historical texts.
Domain-Specific Applications: Technical, educational, and assistant-oriented tasks.

Key Features

Languages: Balochi, English, Urdu, Persian.
Domains:
- General conversation
- Literature
- News
- Academic texts
- Technical content
- Cultural and historical narratives
- Travel and educational materials
- Assistant and chat scenarios
Dataset Size:
- Monolingual and multilingual pairs totaling tens of thousands of records.
- Includes conversational samples, professional scenarios, and technical terminologies.
Custom Tokenizer: Optimized for Balochi, supporting fine-grained tokenization for generative tasks.

Folder Structure

C:.
├───dataset
│   │   Balochi Multilingual Corpus-v2.json
│   │   Balochi_english-translation_corpus.json
│   ├───splits
│   │   │   balochi_to_english.json
│   │   │   balochi_to_persian.json
│   │   │   balochi_to_urdu.json
│   └───text files
│           balochi-only.csv
│           monolingual_balochi.txt
├───tokenizer
│       tokenizer-balochi-v2.json
└───tools used to create this dataset
        Bal_Tokenizer.py
        create-json-format.py

Custom Tokenizer

This dataset includes a custom tokenizer tailored to the Balochi language. The tokenizer has been optimized for encoding Balochi text effectively for generative tasks.

Sample Outputs:

Original: تو کہاں گئے؟

Encoded: [0, 54, 42, 8, 4559, 3, 1]

Original: یہ کنغ پسند کنئے؟

Encoded: [0, 6, 14, 8997, 2066, 2094, 3, 1]

Original: بلی شتگ کور ء اے کسہ ء دمی کارست اتمان ء کسہ ء دمی بهر انت انچو کہ من اول سرا گشتگ ات کہ اے کسہ دو بهر ء انت یکے نبشتہ کار ء وتی کیز ء بندانی سر گوستیں اے بهر ء اولی کسہ پلی مین ء کپتگ ات ء دمی بهر انت

Encoded: [0, 1516, 575, 430, 5, 91, 1053, 5, 11332, 2940, 18035, 5, 1053, 5, 11332, 1234, 32, 1133, 42, 53, 709, 246, 1410, 29, 42, 91, 1053, 63, 1234, 5, 32, 662, 539, 67, 5, 122, 73, 34, 5, 7822, 43, 545, 10, 91, 1234, 5, 618, 1053, 2504, 636, 5, 579, 29, 5, 11332, 1234, 32, 1]

The tokenizer supports tokenized outputs for various use cases, ensuring compatibility with machine learning frameworks such as Hugging Face.

Dataset Highlights

Monolingual Data

Focused on Balochi texts from cultural, historical, and conversational contexts.
Contains cleaned and formatted versions for model training.

Multilingual Translation Data

Balochi-English: General and technical translations.
Balochi-Urdu: News and conversational translations.
Balochi-Persian: Literary and historical translations.

Conversational Samples

Includes a variety of real-world scenarios:

Daily interactions (e.g., greetings, family discussions).
Professional settings (e.g., interviews, meetings).
Educational contexts (e.g., questions in class).
Travel scenarios (e.g., at airports, booking hotels).

Technical Terminology

Covers domains like:

Computing
Medicine
Physics
Internet technology

Licensing

The dataset is released under the CC-BY-SA 4.0 license. This means you can use, share, and modify it, provided proper attribution is given.

Acknowledgments

This dataset is a result of efforts to preserve and enhance computational resources for the Balochi language. The tools and scripts provided alongside this dataset were instrumental in its creation.

Usage Instructions

Download and Preprocess:
- Use the provided JSON files for training data.
- Scripts in the tools folder can be used to preprocess the data.

Tokenizer Integration:

Load the custom tokenizer using the provided tokenizer-balochi-v2.json.

Example:

from transformers import PreTrainedTokenizerFast

tokenizer = PreTrainedTokenizerFast(tokenizer_file="path/to/tokenizer-balochi-v2.json")
encoded = tokenizer.encode("تو کہاں گئے؟")
print(encoded)

Model Training:
- Use frameworks like Hugging Face to fine-tune LLMs on this dataset.
Applications:
- Develop chatbots, translation systems, or other generative applications in Balochi.

For further details or contributions, feel free to reach out or submit issues to the project repository.