Datasets:

hmar-lang
/

hmar-corpus

language:
  - hmr
  - lus
  - en
license: cc-by-nc-sa-4.0
tags:
  - hmar
  - mizo
  - northeast-india
  - language-preservation
  - nlp
  - textual-corpus
  - low-resource-languages
pretty_name: Hmar Digital Corpus (WIP)
size_categories:
  - n<1K
configs:
  - config_name: default
    data_files:
      - split: train
        path: metadata.jsonl

📚 Hmar Digital Corpus & Research Archive

🚧 Status: Work in Progress

This archive documents and preserves Hmar-related literature through manual digitization of physical volumes and the curation of rare digital materials. Each work is processed individually—from scan to metadata—to prioritize long-term preservation and research usability over scale.

Current Workflow:

Digitization: High-quality scans of physical books and documents.
Curation: Collection and verification of existing digital research papers.
Standardization: Structured metadata (metadata.json) for every entry to ensure consistent indexing and retrieval.

📋 Project Philosophy

The Hmar Digital Corpus exists to prevent the loss of Hmar literature and historical records. By converting limited-circulation books, community publications, and local research into a structured digital archive, the project ensures that the Hmar language (ISO 639-3: hmr) remains accessible for academic, linguistic, and cultural research.

This project prioritizes preservation, provenance, and transparency over textual polish or scale.

📂 Repository Structure

The corpus follows an Atomic Folder model: each book or document is a self-contained unit.

books/
├── hmar/          # Hmar language literature & textbooks
├── mizo/          # Mizo-language works concerning Hmar history
├── english/       # Academic research, ethnographies, and linguistics
└── bilingual/     # Dictionaries, primers, and parallel texts

Each document folder typically contains:

the scanned PDF (authoritative artifact),
a metadata.json file,
and, when available, an ocr/ directory.

🔍 Metadata & Searchability

Every document directory includes a metadata.json file, enabling reliable organization and search at scale.

Core Fields: Title (bilingual where applicable), Author, Year, Publisher.
Extended Metadata: Contextual information such as institutional approvals, author credentials, edition notes, or physical source details.

Metadata is treated as first-class data and is required even when OCR is absent.

📁 OCR Directory Structure

When available, OCR output is stored in a dedicated ocr/ subdirectory within each book or document folder.

Example structure:

Bilingual/
└── hmar_tawng_inchukna/
    ├── hmar_tawng_inchukna_2012.pdf
    ├── metadata.json
    └── ocr/
        └── hmar_tawng_inchukna_2012.txt

OCR File Semantics

The ocr/ directory contains raw, untouched OCR output.
In the current corpus, OCR is typically stored as a single consolidated .txt file corresponding to the scanned PDF.
In future entries, OCR may also be organized per page or per chapter where useful.
No semantic corrections, spell-checking, normalization, or stylistic cleanup is applied unless explicitly documented in metadata.

The presence of an ocr/ directory does not imply textual accuracy or verification. OCR text exists to support search, reference, and future annotation—not as a definitive transcription.

The scanned PDF remains the authoritative source unless a text has been explicitly marked as verified or curated in its metadata.

Any corrected or curated text—if introduced—will exist as a separate layer and will retain explicit references to the original OCR output.

🛠 Progress & Limitations

Preservation first: Priority is given to securing high-quality scans and accurate metadata.
OCR is secondary: OCR is generated opportunistically and may be incomplete or error-prone.
Incremental growth: Materials are added as they are processed and documented.

⚖️ Rights & Usage

This archive is provided for non-commercial research, linguistic study, and cultural preservation.

Copyright remains with the original authors, societies, and publishers (e.g., HLS, HSA, AIRTSC). Rights holders may request removal or modification of materials through the repository’s discussion channel.