metadata

language:
  - fr
license: cc-by-4.0
task_categories:
  - token-classification
dataset_info:
  features:
    - name: ner_tags
      sequence: int64
    - name: tokens
      sequence: string
    - name: pos_tags
      sequence: string
  splits:
    - name: train
      num_bytes: 17859073
      num_examples: 26754
  download_size: 3480973
  dataset_size: 17859073
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*

WikiNER-fr-gold

This dataset is a manually revised version of 20% of the French proportion of WikiNER. The original dataset is currently available here, based on which WikiNER-fr-gold is created. The entities are annotated using the BIOES scheme. The POS tags are not revised i.e. remain the same as the original dataset.

For more information on the revision details, please refer to our paper WikiNER-fr-gold: A Gold-Standard NER Corpus.

The dataset is available in two formats. The CoNLL version contains three columns: text, POS and NER. The Parquet version is downloadable using the datasets library. Originally conceived as a test set, there is no recommended train/dev/test split. The downloaded dataset is by default labeled train.

from datasets import load_dataset

ds = load_dataset('danrun/WikiNER-fr-gold')

ds['train'][0]
# {'ner_tags': [...], 'tokens': [...], 'pos_tags': [...]}

The NER tags are indexed using the following table (see labels.json):

{
 'O': 0,
 'B-PER': 1,
 'I-PER': 2,
 'E-PER': 3,
 'S-PER': 4,
 'B-LOC': 5,
 'I-LOC': 6,
 'E-LOC': 7,
 'S-LOC': 8,
 'B-ORG': 9,
 'I-ORG': 10,
 'E-ORG': 11,
 'S-ORG': 12,
 'B-MISC': 13,
 'I-MISC': 14,
 'E-MISC': 15,
 'S-MISC': 16
}

Citation

@misc{cao2024wikinerfrgoldgoldstandardnercorpus,
      title={WikiNER-fr-gold: A Gold-Standard NER Corpus}, 
      author={Danrun Cao and Nicolas Béchet and Pierre-François Marteau},
      year={2024},
      eprint={2411.00030},
      archivePrefix={arXiv},
      primaryClass={cs.CL},
      url={https://arxiv.org/abs/2411.00030}, 
}