|
Download README.md from proustpunk/languages: direct link, hf CLI and curl.
- Browser
- Download file 1.51 kB
-
https://huggingface.co/datasets/proustpunk/languages/resolve/main/README.md
- Command line
-
hf download hf://datasets/proustpunk/languages/README.md
-
curl -L -H "Authorization: Bearer $HF_TOKEN" -o README.md https://huggingface.co/datasets/proustpunk/languages/resolve/main/README.md
1.51 kB
metadata
license: cc-by-sa-4.0
language:
- new
pretty_name: Newar (Nepal Bhasa) Wikipedia text, cleaned and clustered
Newar (Nepal Bhasa) Wikipedia text
Source
Newar Wikipedia (new.wikipedia.org), from the official Wikimedia dump. Wikipedia text is licensed CC BY-SA. This dataset keeps that license (ShareAlike applies to anything derived from it).
What was done
- Extracted article text from the dump (normal articles only; no redirects or talk pages).
- Removed lines shorter than 20 characters.
- Removed lines where under 80% of letters were Devanagari or Newa script (English, etc.).
- Removed exact duplicate lines (ignoring punctuation and case).
- Clustered lines with character n-gram TF-IDF + KMeans (20 clusters).
- Labelled clusters by reading samples. Clusters [12, 17] were judged to be Sanskrit by a non-speaker reading samples.
Files
newar_candidate.jsonl: 181403 linessanskrit.jsonl: 24290 lines
Each line is one JSON record: title, url, text, cluster, label, source, license.
Known limitations
- NOT verified by native Newar speakers.
- Newar, Nepali, Hindi and Sanskrit all use Devanagari. The script filter cannot
tell them apart, so
newar_candidatemay still contain Nepali or Hindi. - Cluster labels came from a non-speaker reading samples; they may be wrong.
- Near-duplicate bot-made stubs (lines differing by one name) are not removed.
- The share of bot-generated articles has not been measured.