README.md

metadata

dataset_info:
  - config_name: alias-resolution
    features:
      - name: form
        dtype: string
      - name: type
        dtype: string
      - name: mentions
        dtype: int64
      - name: entity
        dtype: string
      - name: novel
        dtype: string
    splits:
      - name: train
        num_bytes: 395702
        num_examples: 5985
    download_size: 117587
    dataset_size: 395702
  - config_name: text
    features:
      - name: tokens
        sequence: string
      - name: novel
        dtype: string
    splits:
      - name: train
        num_bytes: 10400464
        num_examples: 7
    download_size: 2673188
    dataset_size: 10400464
configs:
  - config_name: alias-resolution
    data_files:
      - split: train
        path: alias-resolution/train-*
  - config_name: text
    data_files:
      - split: train
        path: text/train-*

7-romans

This dataset contains 7 French novels, entirely annoted for the alias resolution task. See the related NER dataset.

Novel	Author	Publication Year	Number of tokens	Number of characters
Les Trois Mousquetaires	Alexandre Dumas	1849	294 989	213
Le Rouge et le Noir	Stendhal	1854	216 445	318
Eugénie Grandet	Honoré de Balzac	1855	80 659	107
Germinal	Émile Zola	1885	220 273	102
Bel-Ami	Guy de Maupassant	1901	138 156	150
Notre-Dame de Paris	Victor Hugo	1904	221 351	536
Madame Bovary	Gustave Flaubert	1910	148 861	175

This gold standard corpus was created in the context of a project at the ObTIC laboratory, Sorbonne University. The project was directed by Motasem Alrahabi, and annnotations were performed by Perrine Maurel, Una Faller and Romaric Parnasse.

The corpus was then used to train a CamemBERT NER model in collaboration with Arthur Amalvy and Vincent Labatut, from Avignon University.

Usage

To load the alias resolution data:

>>> from datasets import load_dataset
>>> dataset = load_dataset("compnet-renard/7-romans-alias-resolution", "alias-resolution")
>>> dataset["train"][0]
{'form': 'À la belle vue', 'type': 'LOC', 'mentions': 1, 'entity': '?', 'novel': 'BelAmi'}

Only the PER entities are annotated: other types only have a "?" in their entity field.

The novel texts themselves are in a separate configuration:

>>> dataset = load_dataset("compnet-renard/7-romans-alias-resolution", "text")
>>> dataset["train"].features
{'tokens': Sequence(feature=Value(dtype='string', id=None), length=-1, id=None), 'novel': Value(dtype='string', id=None)}

Citation

If you use this dataset in your research, please cite:

@InProceedings{Maurel2025,
  authors = {Maurel, P. and Amalvy, A. and Labatut, V. and Alrahabi, M.},
  title = {Du repérage à l’analyse : un modèle pour la reconnaissance d’entités nommées dans les textes littéraires en français},
  booktitle = {Digital Humanities 2025},
  year = {2025},
}