v0.2.0

UFSAC training corpora added on top of v0.1.0's Raganato evaluation bundle.

New configs

Config	Split	Rows	OEWN 2024 coverage
en-semcor	train	229,533	91.71 %
en-wngt	train	496,776	89.69 %
en-masc	train	113,523	40.98 % *
en-senseval2_ls	train + test	12,690	88-89 %
en-senseval3_ls	train + test	11,668	80-82 %
en-semeval2007_t17_ls	test	455	99.78 %

• MASC mixes WN 3.0/3.1 sensekeys with ANC-specific senses; the lower resolution is expected and surfaced by the coverage report so downstream evaluation can filter.

Total across all 11 configs: 871,898 rows, 83.83 % OEWN 2024 resolution.

Implementation

• loaders/ufsac.py streams UFSAC XML via iterparse (WNGT / OMSTI-scale files don't need to fit in memory).
• scripts/build_all.py takes --ufsac --ufsac-root to regenerate.
• loaders/xl_wsd.py + mappers/babelnet_to_wn.py are explicit stubs raising NotImplementedError — XL-WSD is blocked on a BabelNet academic-license mapping file.

Links

• HF Hub: https://huggingface.co/datasets/alvations/pywsd-datasets
• PyPI: https://pypi.org/project/pywsd-datasets/0.2.0/

v0.1.0

Initial release: English Raganato bundle normalized to oewn:2024.

Bundled data mirror

WSD_Unified_Evaluation_Datasets.zip (Raganato, Camacho-Collados, Navigli 2017)
mirrored here because upstream http://lcl.uniroma1.it/wsdeval/ serves a
mismatched TLS cert and is a single point of failure.

The loader in pywsd_datasets.loaders.raganato fetches from this release
by default.