CATIE-AQ/frenchSTS

Dataset information

Dataset concatenating Sentence Similarity datasets available in French and open-source.
There are a total of 45,726 rows, of which 12,227 are for training, 3,526 for validation and 29,973 for testing.

Usage

from datasets import load_dataset
dataset = load_dataset("CATIE-AQ/frenchSTS")

Dataset

Details of rows

Dataset Original	Splits	Note
stsb_multi_mt	5,749 train / 1,500 validation / 1,379 test	We only keep the French split (fr)
GEM/opusparcus	0 train / 1,630 validation / 1,674 test	We only keep the French split (fr.100)
mteb/sts22-crosslingual-sts	0 train / 0 validation / 104 test	We only keep the French split (fr)
Lajavaness/SICK-fr	4,439 train / 495 validation / 4,906 test
Lajavaness/STS12-fr	2,234 train / 0 validation / 3,108 test
Lajavaness/STS13-fr	0 train / 0 validation / 1,500 test
Lajavaness/STS14-fr	0 train / 0 validation / 3,750 test
Lajavaness/STS15-fr	0 train / 0 validation / 3,000 test
Lajavaness/STS16-fr	0 train / 0 validation / 1,186 test
OrdalieTech/Ordalie-FR-STS-benchmark	0 train / 0 validation / 10,000 test
musts/french	600 train / 0 validation / 410 test

Removing duplicate data and leaks

The sum of the values of the datasets listed here gives the following result:

DatasetDict({
    train: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 13022
    })
    validation: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 3625
    })
    test: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 30913
    })
})

However, a data item in training split A may not be in A's test split, but may be present in B's test set, creating a leak when we create the A+B dataset.
The same logic applies to duplicate data. So we need to make sure we remove them.
After our clean-up, we finally have the following numbers:

DatasetDict({
    train: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 12227
    })
    validation: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 3526
    })
    test: Dataset({
        features: ['sentence1', 'sentence2', 'score', 'dataset'],
        num_rows: 29973
    })
})

Columns

• the sentence1 column contains the first sentence
• the sentence2 column contains the second sentence
• the score column contains the similarity score between sentence1 and sentence2
• the dataset column identifies the row's original dataset (if you wish to apply filters to it)

Split

• train corresponds to the concatenation of stsb + sick + sts12+ french_musts
• validation corresponds to the concatenation of stsb + opusparcus + sick
• test corresponds to stsb + opusparcus + sts22 + sick + sts12 + sts13 + sts14 + sts15 + sts16 + ordalie+ french_musts

Citations

GEM/opusparcus

@InProceedings{creutz:lrec2018,
  title = {Open Subtitles Paraphrase Corpus for Six Languages},
  author={Mathias Creutz},
  booktitle={Proceedings of the 11th edition of the Language Resources and Evaluation Conference (LREC 2018)},
  year={2018},
  month = {May 7-12},
  address = {Miyazaki, Japan},
  editor = {Nicoletta Calzolari (Conference chair) and Khalid Choukri and Christopher Cieri and Thierry Declerck and Sara Goggi and Koiti Hasida and Hitoshi Isahara and Bente Maegaard and Joseph Mariani and Hélène Mazo and Asuncion Moreno and Jan Odijk and Stelios Piperidis and Takenobu Tokunaga},
  publisher = {European Language Resources Association (ELRA)},
  isbn = {979-10-95546-00-9},
  language = {english},
  url={http://www.lrec-conf.org/proceedings/lrec2018/pdf/131.pdf}}

Lajavaness/SICK-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/SICK-fr

Lajavaness/STS12-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/STS12-fr

Lajavaness/STS13-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/STS13-fr

Lajavaness/STS14-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/STS14-fr

Lajavaness/STS15-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/STS15-fr

Lajavaness/STS16-fr

Dataset by Lajavaness (2023)
Hugging Face repository: https://huggingface.co/datasets/Lajavaness/STS16-fr

stsb_multi_mt

@InProceedings{huggingface:dataset:stsb_multi_mt,
title = {Machine translated multilingual STS benchmark dataset.},
author={Philip May},
year={2021},
url={https://github.com/PhilipMay/stsb-multi-mt}}

OrdalieTech/Ordalie-FR-STS-benchmark

Dataset by OrdalieTech (2023)
Hugging Face repository: https://huggingface.co/datasets/OrdalieTech/Ordalie-FR-STS-benchmark

mteb/sts22-crosslingual-sts

@inproceedings{hercig-kral-2021-evaluation,
    title = "Evaluation Datasets for Cross-lingual Semantic Textual Similarity",
    author = "Hercig, Tom{\'a}{\v{s}}  and  Kral, Pavel",
    editor = "Mitkov, Ruslan  and  Angelova, Galia",
    booktitle = "Proceedings of the International Conference on Recent Advances in Natural Language Processing (RANLP 2021)",
    month = sep,
    year = "2021",
    address = "Held Online",
    publisher = "INCOMA Ltd.",
    url = "https://aclanthology.org/2021.ranlp-1.59",
    pages = "524--529",
    abstract = "Semantic textual similarity (STS) systems estimate the degree of the meaning similarity between two sentences. Cross-lingual STS systems estimate the degree of the meaning similarity between two sentences, each in a different language. State-of-the-art algorithms usually employ a strongly supervised, resource-rich approach difficult to use for poorly-resourced languages. However, any approach needs to have evaluation data to confirm the results. In order to simplify the evaluation process for poorly-resourced languages (in terms of STS evaluation datasets), we present new datasets for cross-lingual and monolingual STS for languages without this evaluation data. We also present the results of several state-of-the-art methods on these data which can be used as a baseline for further research. We believe that this article will not only extend the current STS research to other languages, but will also encourage competition on this new evaluation data.",}

Musts/french

Dataset by Musts (2024)
Hugging Face repository: https://huggingface.co/datasets/musts/french

FrenchSTS

@misc {frenchSTS_2025,   
    author       = { {BOURDOIS, Loïck} },  
    organization  = { {Centre Aquitain des Technologies de l'Information et Electroniques} },  
    title        = { frenchSTS (Revision aae6da7) },  
    year         = 2025,  
    url          = { https://huggingface.co/datasets/CATIE-AQ/frenchSTS },  
    doi          = { 10.57967/hf/5168 },  
    publisher    = { Hugging Face } }

License

cc-by-nc-4.0 if you use the dataset with the GEM/opusparcus' datas, cc-by-4.0 otherwise.