Hmar & English Parallel Numeral Words Dataset (numeral-words)

A 3-way parallel digital dataset containing 999,999 spelled-out Hmar & English number words mapped to their corresponding numerical integer values (1 to 1,000,000).

Maintained as part of the Hmar Heritage Archival Project by the Hmar Heritage Foundation.

---

Dataset Summary

Category	Value
Languages	Hmar (hmr), English (en)
Record Count	999,999 Parallel Rows
Format	Compressed 1-Line JSONL (data/*.jsonl.gz)
Columns	number (Integer), hmar (Hmar Numeral), english (Standard Spoken English Numeral)
Source / Adaptation	Base Hmar adapted from Mizo Numbers (remruata/mizonumbers); English generated via num2words engine.
License	Apache-2.0

---

Dataset Splits

Split	Share	Record Count	File Pattern
Train	90.0%	899,958	data/train-*.jsonl.gz
Validation	5.0%	50,081	data/validation-*.jsonl.gz
Test	5.0%	49,960	data/test-*.jsonl.gz

---

Data Format

Each line in data/*.jsonl.gz is a JSON object with 3 parallel fields:

{"number": 1, "hmar": "pakhat", "english": "one"}
{"number": 16, "hmar": "sawm paruk", "english": "sixteen"}
{"number": 136, "hmar": "za sawm thum paruk", "english": "one hundred thirty six"}

---

License

Apache-2.0 License. Maintained by the Hmar Heritage Foundation.