Centre for Text Technology, North-West University

license

cc-by-2.5(Hugging Face card metadata

access

public(ungated loader

dataset_card

partial(paper abstract only

The loader is ungated and pulls openly downloadable archives from SADiLaR under an attribution license. The card itself documents almost nothing beyond the paper abstract.

• https://huggingface.co/api/datasets/nwu-ctext/nchlt recorded 2026-09-24

  "gated": false; repository files .gitattributes, README.md, nchlt.py.

• https://huggingface.co/datasets/nwu-ctext/nchlt recorded 2026-09-24

  Card sections read "[More Information Needed]"; curator "Martin.Puttkammer@nwu.ac.za".

• https://huggingface.co/datasets/nwu-ctext/nchlt/raw/main/nchlt.py recorded 2026-09-24

  Configs download e.g. "https://repo.sadilar.org/bitstream/handle/20.500.12185/299/nchlt_afrikaans_named_entity_annotated_corpus.zip".

• https://huggingface.co/datasets/nwu-ctext/nchlt/raw/main/README.md recorded 2026-09-24

  Front matter "license: cc-by-2.5", nine languages, "named-entity-recognition".

Hugging Face downloads of the loader repository only; direct downloads from SADiLaR are not counted. Downloads count loader fetches, not users.

• https://huggingface.co/api/datasets/nwu-ctext/nchlt recorded 2026-09-24

  166 downloads in the trailing 30 days for nwu-ctext/nchlt

The NCHLT corpora are an established resource for South African languages, described in an LREC paper, and PuoBERTa lists them among its training data. The Hugging Face entry exposes only the named-entity subset through a loader with an empty card, so it offers far less text than WURA, a documented African web corpus that trained the AfriTeVa V2 models.

• https://huggingface.co/datasets/nwu-ctext/nchlt recorded 2026-09-24

  Models trained or fine-tuned list includes dsfsi/PuoBERTa.

• https://huggingface.co/datasets/nwu-ctext/nchlt/raw/main/README.md recorded 2026-09-24

  config af train num_examples 8961; NER tag set B-PERS, B-ORG, B-LOC, B-MISC.