spaCy Custom Spancat trained on Diderot & d’Alembert’s Encyclopédie entries

This model is designed to identify and classify text spans corresponding to named entities (such as Spatial, Person, and MISC), as well as nested named entities (including Spatial, Person, and MISC), spatial relations, and other relevant information within French encyclopedic entries.

The spans detected by this model are:

NC-Spatial: a common noun that identifies a spatial entity (nominal spatial entity) including natural features, e.g. ville, la rivière, royaume.
NP-Spatial: a proper noun identifying the name of a place (spatial named entities), e.g. France, Paris, la Chine.
ENE-Spatial: nested spatial entity , e.g. ville de France, royaume de Naples, la mer Baltique.
Relation: spatial relation, e.g. dans, sur, à 10 lieues de.
Latlong: geographic coordinates, e.g. Long. 19. 49. lat. 43. 55. 44.
NC-Person: a common noun that identifies a person (nominal spatial entity), e.g. roi, l'empereur, les auteurs.
NP-Person: a proper noun identifying the name of a person (person named entities), e.g. Louis XIV, Pline, les Romains.
ENE-Person: nested people entity, e.g. le czar Pierre, roi de Macédoine
NP-Misc: a proper noun identifying entities not classified as spatial or person, e.g. l'Eglise, 1702, Pélasgique.
ENE-Misc: nested named entity not classified as spatial or person, e.g. l'ordre de S. Jacques, la déclaration du 21 Mars 1671.
Head: entry name
Domain-Mark: words indicating the knowledge domain (usually after the head and between parenthesis), e.g. Géographie, Geog., en Anatomie.

Model Description

Developed by: Ludovic Moncla, Katherine McDonough and Denis Vigier in the framework of the GEODE project.
Model type: spaCy Span Categorization
spaCy: >=3.7.2,<3.8.0
Components: tok2vec, spancat
Repository: https://github.com/GEODE-project/ner-spancat-edda
Language(s) (NLP): French
License: cc-by-nc-4.0
Dataset: https://zenodo.org/records/10530177

Bias, Risks, and Limitations

This model was trained entirely on French encyclopaedic entries and will likely not perform well on text in other languages or other corpora.

How to Get Started with the Model

Use the code below to get started with the model.

pip install https://huggingface.co/GEODE/fr_spacy_custom_spancat_edda/resolve/main/fr_spacy_custom_spancat_edda-any-py3-none-any.whl

# Using spacy.load().
import spacy
nlp = spacy.load("fr_spacy_custom_spancat_edda")

# Importing as module.
import fr_spacy_custom_spancat_edda
nlp = fr_spacy_custom_spancat_edda.load()
doc = nlp("* ALBI, (Géog.) ville de France, capitale de l'Albigeois, dans le haut Languedoc : elle est sur le Tarn. Long. 19. 49. lat. 43. 55. 44.")

spans = []

for span in doc.spans['sc']:
  spans.append({
      "start": span.start_char,
      "end": span.end_char,
      "labels": [span.label_],
      "text": span.text
  })

print(spans)

# Output
[{'start': 2, 'end': 6, 'labels': ['Head'], 'text': 'ALBI'},
 {'start': 16, 'end': 21, 'labels': ['NC-Spatial'], 'text': 'ville'},
 {'start': 25, 'end': 31, 'labels': ['NP-Spatial'], 'text': 'France'},
 {'start': 33, 'end': 41, 'labels': ['NC-Spatial'], 'text': 'capitale'},
 {'start': 58, 'end': 62, 'labels': ['Relation'], 'text': 'dans'},
 {'start': 92, 'end': 95, 'labels': ['Relation'], 'text': 'sur'},
 {'start': 9, 'end': 14, 'labels': ['Domain-mark'], 'text': 'Géog.'},
 {'start': 45, 'end': 56, 'labels': ['NP-Spatial'], 'text': "l'Albigeois"},
 {'start': 96, 'end': 103, 'labels': ['NP-Spatial'], 'text': 'le Tarn'},
 {'start': 16,
  'end': 31,
  'labels': ['ENE-Spatial'],
  'text': 'ville de France'},
 {'start': 63,
  'end': 80,
  'labels': ['NP-Spatial'],
  'text': 'le haut Languedoc'},
 {'start': 33,
  'end': 56,
  'labels': ['ENE-Spatial'],
  'text': "capitale de l'Albigeois"},
 {'start': 33,
  'end': 80,
  'labels': ['ENE-Spatial'],
  'text': "capitale de l'Albigeois, dans le haut Languedoc"},
 {'start': 16,
  'end': 80,
  'labels': ['ENE-Spatial'],
  'text': "ville de France, capitale de l'Albigeois, dans le haut Languedoc"}]

Training Details

Training Data

The model was trained using a set of 2200 paragraphs randomly selected out of 2001 Encyclopédie's entries. All paragraphs were written in French and are distributed as follows among the Encyclopédie knowledge domains:

Knowledge domain	Paragraphs
Géographie	1096
Histoire	259
Droit Jurisprudence	113
Physique	92
Métiers	92
Médecine	88
Philosophie	69
Histoire naturelle	65
Belles-lettres	65
Militaire	62
Commerce	48
Beaux-arts	44
Agriculture	36
Chasse	31
Religion	23
Musique	17

The spans/entities were labeled by the project team along with using pre-labelling with early models to speed up the labelling process. A train/val/test split was used. Validation and test sets are composed of 200 paragraphs each: 100 classified as 'Géographie' and 100 from another knowledge domain. The datasets have the following breakdown of tokens and spans/entities.

	Train	Validation	Test
Paragraphs	1,800	200	200
Tokens	132,398	14,959	13,881
NC-Spatial	3,252	358	355
NP-Spatial	4,707	464	519
ENE-Spatial	3,043	326	334
Relation	2,093	219	226
Latlong	553	66	72
NC-Person	1,378	132	133
NP-Person	1,599	170	150
ENE-Person	492	49	57
NP-Misc	948	108	96
ENE-Misc	255	31	22
Head	1,261	142	153
Domain-Mark	1,069	122	133

Training Procedure

For full training details and results please see the GitHub repository: github.com/GEODE-project/ner-spancat-edda

Evaluation

Evaluation is performed using the spacy evaluate command line interface:

Overall model performances (Test set)

	Precision	Recall	F-score
	94.09	79.91	86.42

Model performances by entity (Test set)

	Precision	Recall	F-score
NC-Spatial	96.50	93.24	94.84
NP-Spatial	92.74	95.95	94.32
ENE-Spatial	91.67	95.51	93.55
Relation	97.33	64.60	77.66
Latlong	0.00	0.00	0.00
NC-Person	93.07	70.68	80.34
NP-Person	92.47	90.00	91.22
ENE-Person	92.16	82.46	87.04
NP-Misc	93.24	71.88	81.18
ENE-Misc	0.00	0.00	0.00
Head	97.37	24.18	38.74
Domain-mark	99.19	91.73	95.31

Cite this work

Moncla, L., Vigier, D., & McDonough, K. (2024). GeoEDdA: A Gold Standard Dataset for Geo-semantic Annotation of Diderot & d’Alembert’s Encyclopédie. In proceedings of the 2nd International Workshop on Geographic Information Extraction from Texts (GeoExT'24), ECIR Conference, Glasgow, UK.

Acknowledgement

The authors are grateful to the ASLAN project (ANR-10-LABX-0081) of the Université de Lyon, for its financial support within the French program "Investments for the Future" operated by the National Research Agency (ANR). Data courtesy the ARTFL Encyclopédie Project, University of Chicago.

GEODE
/

fr_spacy_custom_spancat_edda