---
dataset_info:
  features:
  - name: inputs
    dtype: string
  - name: choices_list
    list: string
  - name: labels
    dtype: int64
  - name: metadata
    dtype: string
  splits:
  - name: train
    num_bytes: 218023276
    num_examples: 190865
  download_size: 166658363
  dataset_size: 218023276
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
license: other
source_datasets:
- thesofakillers/SemCor
---

# semcor

SemCor native sense keys joined to the matching Princeton WordNet 1.6 glosses.

Original distribution: https://wordnet.princeton.edu/semcor.
Pinned faithful CSV copy: https://huggingface.co/datasets/thesofakillers/SemCor.
Uses only one native gold sense and at least two distinct lemma/POS senses;
unresolved keys, multiple senses and ambiguous duplicate glosses are excluded.
All Brown sections remain training-only, not fabricated evaluation partitions.
Terms: original Princeton WordNet/SemCor license; see upstream LICENSE.

Original data: [thesofakillers/SemCor](https://huggingface.co/datasets/thesofakillers/SemCor), [original distribution](https://wordnet.princeton.edu/semcor). Repackaged as parquet for [tasksource](https://github.com/sileod/tasksource) by [scripts/repackage_dataset/linguistic.py](https://github.com/sileod/tasksource/blob/main/scripts/repackage_dataset/linguistic.py#L91).

[Conversion provenance](provenance.json): source revisions, hashes, configuration and exclusions.