unigrams

A frequency-weighted lexical dataset containing 110,603 Hmar unigrams and active loanwords with occurrence counts compiled directly from the Foundation's verified Hmar corpus (over 4.83 million words).

Maintained by the Hmar Heritage Foundation (hmarheritage.pages.dev).

---

Overview

• Language: Hmar (hmr, ISO 639-3, Glottolog: hmar1241)
• Family: Zo Languages
• Volume: 110,603 unigram tokens (compiled across 4,832,070 words)
• Format: JSONL (data/train.jsonl)
• License: MIT
• DOI: 10.57967/hf/10422
• Coverage: Core grammatical particles (including a, i, e) and digital sentiment tokens are fully included in their true empirical frequencies with standardized boundary spacing.

Dataset Structure

Each record consists of a unigram token and its global occurrence count, sorted in descending order:

{"word": "a", "count": 418374}
{"word": "an", "count": 120530}
{"word": "chu", "count": 117210}

Usage

from datasets import load_dataset

dataset = load_dataset("hmar-heritage-org/unigrams")
print(dataset["train"][0])

Source & Provenance

Derived directly from the Foundation's sentences corpus. Word tokenization, frequency aggregation, text normalization, and linguistic verification were conducted by the Hmar Heritage Foundation using the hmaraniam engine.

Orthographic Variations & Empirical Policy

To reflect genuine language use across different historical, educational, and digital registers, all real-world orthographic variants are intentionally preserved as-is without forced normalization or diacritic folding:

• Plain QWERTY / Standard Latin: Forms without diacritics (e.g., tha, tan, thang).
• Canonical Literary Diacritics: Traditional orthography utilizing the retroflex dot ṭ (\u1E6D) and circumflex vowels (â, ê, î, ô, û) (e.g., ṭha, ṭan).
• Mobile Keyboard Orthography: Digital community writing utilizing the comma-below ț (\u021B) (e.g., țha, țan).

Downstream users and tokenizer developers can choose to fold or strip diacritics according to their own task requirements while having access to the true, unadulterated empirical frequency distribution. In accordance with standard unigram tokenization ($n=1$), words appearing in natural Zo reduplications (e.g., chu chu, bek bek) are tallied per individual space-delimited token occurrence.

Citation

@dataset{hmar_heritage_unigrams_2026,
  author    = {{Hmar Heritage Foundation}},
  title     = {unigrams: A Frequency-Weighted Lexical Dataset for Hmar},
  year      = {2026},
  publisher = {Hugging Face},
  doi       = {10.57967/hf/10422},
  url       = {https://huggingface.co/datasets/hmar-heritage-org/unigrams},
  keywords  = {Hmar, hmr, hmar1241, Zo Languages},
  note      = {Language: Hmar (ISO 639-3: hmr, Glottolog: hmar1241). Language family: Zo Languages.}
}