Hugging Face's logo

Datasets:
OpenLLM-France
/
Claire-Dialogue-French-0.1

Tasks:
Text Generation
Text2Text Generation
Sub-tasks: language-modeling dialogue-modeling dialogue-generation
Languages: French
Multilinguality: monolingual
Size Categories: 100M<n<1B
ArXiv:
Tags: conversational text-generation conditional-text-generation dialogue-modeling dialogue-generation
License:
Dataset card Viewer Files Community
3

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

YAML Metadata Warning: The task_categories "conversational" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, text2text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, other

Claire French Dialogue Dataset (CFDD)
A collection of French dialogue transcripts and plays

This is the first packaged version of the datasets used to train the Claire family of large language models (OpenLLM-France/Claire-7B-0.1).

The Claire French Dialogue Dataset (CFDD) is a collection of theater plays and transcripts of real French dialogues from various sources, including parliamentary proceedings, interviews, debates, meetings, and free conversations. Each dialogue is split into speech turns, and each speech turn is labeled with the name of the speaker, or a unique identifier if the speaker is unknown.

Dataset composition

CFDD can be broken down into:

  • 37 015 conversations in total (36 731 in train, 284 in test)
  • 2 961 116 speech turns in total (2 934 084 in train, 27 032 in test)
  • around 150M words

It is a collection of several independent datasets, classified by the types of conversations they contain. This categorization is designed to more evenly balance the influence of different styles of dialogue on model training and to facilitate future applications of CFDD for which certain types of dialogue might be more helpful than others.

Note that this categorization leads to multiple cases in which the original corpus is split into subcorpora. When the smaller sets are included in our corpus, they are clearly indicated, e.g., "ESLO (1/5)". Some portions of the original corpora have been excluded entirely because they did not include dialogue between adults (e.g., monologues, read literature).

For more information, you can look at the following documents:

Data sources

Dataset Sub-folder(s) Description Words Turns Conversations License (and conditions)
Parliamentary Proceedings
Assemblée Nationale FR/AssembleeNationale*
(one folder per legislative period)		 Parliamentary proceedings from the French National Assembly 133M 1.6M 4.5k Open License 2.0
Theatre
Theatre Classique FR/TheatreClassique Classic stage plays 12.8M 441k 25k CC BY-NC-SA 4.0 (please cite)
Theatre Gratuit FR/TheatreGratuit Stage plays 2.7M 155k 4k
Interviews
ESLO (1/5) FR/ESLO_interview Guided conversations 4.2M 329k 399 CC BY-NC-SA 4.0 (please cite)
TCOF (adults) FR/TCOF_adults Guided conversations (between adults) 765k 49k 237 CC BY-NC-SA 2.0 (please cite)
CFPP FR/CFPP Interviews of people in Paris in 2000 608k 48k 42 CC BY-NC-SA 3.0 (please cite)
ORFEO/Valibel (1/2) FR/ORFEO_valibel_interview Guided conversations of Belgian French speakers 458k 19k 67 CC BY-NC-SA 4.0 (please cite)
PFC (1/2) FR/PFC_guided Guided interviews 268k 15k 173 CC BY-NC-SA 4.0 (please cite)
ORFEO/CFPB FR/ORFEO_cfpb Interviews of people in Brussels 138k 11k 12 CC BY-NC-SA 4.0
ACSYNT FR/ACSYNT Guided interviews from southwestern France 61k 2.7k 144 CC BY-SA 4.0 (please cite)
Free Conversations
OFROM FR/OFROM Conversations in French-speaking Switzerland 590k 44k 151 CC BY-NC-SA 3.0 (please cite)
ESLO (2/5) FR/ESLO_free Diverse conversation 480k 47k 98 CC BY-NC-SA 4.0 (please cite)
ORFEO/CRFP FR/ORFEO_crfp Diverse conversations 405k 9k 124 CC BY-NC-SA 4.0 (please cite)
ORFEO/C-ORAL-ROM FR/ORFEO_coralrom Diverse conversation 248k 6k 152 CC BY-NC-SA 4.0 (please cite)
PFC (2/2) FR/PFC_free Diverse conversation 230k 14k 146 CC BY-NC-SA 4.0 (please cite)
CLAPI FR/CLAPI Diverse conversation 122k 15k 14 CC BY-NC-SA 4.0
CID FR/CID Dialogues between two friends 118k 9k 8 CC BY-NC-SA 4.0 (please cite)
Rhapsodie FR/Rhapsodie Diverse conversations 28k 1k 41 CC BY-NC-SA 3.0 (please cite)
Paris Stories FR/ParisStories Diverse conversations in Paris 28k 351 54 CC BY-SA 4.0
LinTO (1/3) FR/LINAGORA_free Diverse conversation 26k 2k 4 CC BY-SA 4.0 (please cite)
Meetings
SUMM-RE FR/SUMM-RE Meeting-style conversations (transcribed with Whisper large-v2 ASR) 1.3M 39k 283 CC BY-SA 4.0 (please cite)
ORFEO/Reunions-de-Travail FR/ORFEO_reunions-de-travail Real meetings 210k 12k 29 CC BY-NC-SA 4.0
LinTO (2/3) FR/LINAGORA_meetings Meetings on speech recognition 41k 1.8k 6 CC BY-SA 4.0 (please cite)
Debates
FREDSum FR/FREDSum French political debates 406k 7k 144 CC BY-SA 4.0 (please cite)
ESLO (3/5) FR/ESLO_conf Conferences 76k 2k 4 CC BY-NC-SA 4.0 (please cite)
Assistance
ESLO (4/5) FR/ESLO_assistance In-person assistance and call-centers 95k 11k 143 CC BY-NC-SA 4.0 (please cite)
ORFEO/Fleuron FR/ORFEO_fleuron Interactions created to teach foreign students about university life 33k 2k 51 CC BY-NC-SA 4.0 (please cite)
OTG FR/OTG Dialogues in a tourism office 27k 4k 315 CC BY-SA 3.0 (contact before usage)
Accueil UBS FR/UBS University telephone answering service 7.2k 1k 41 CC BY-SA 3.0 (contact before usage)
Presentation, Formal Address
ESLO (5/5) FR/ESLO_discourse Conference presentations 43k 120 9 CC BY-NC-SA 4.0 (please cite)
LinTO (3/3) FR/LINAGORA_discourse Technical presentations (AI topics) with Q/A 38k 1.5k 4 CC BY-SA 4.0 (please cite)
ORFEO/Valibel (2/2) FR/ORFEO_valibel_discourse Formal university addresses 12k 5 5 CC BY-NC-SA 4.0 (please cite)

Example use (python)

In the following sample_by="paragraph" is important to ensure that each sample corresponds to a full conversation (not just a speech turn).

Load dataset from HuggingFace cache (downloaded under ~/.cache/huggingface/datasets):

from datasets import load_dataset

dataset = load_dataset("OpenLLM-France/Claire-Dialogue-French-0.1", sample_by="paragraph", streaming=True)

Load dataset from raw text files:

from datasets import load_dataset
import glob

path = "path/to/dataset"
train_files = glob.glob(path + "/*/train.txt")
test_files = glob.glob(path + "/*/test.txt")

dataset = load_dataset("text", data_files={"train": train_files, "test": test_files}, sample_by="paragraph", streaming=True)

Iterate on the dataset:

for sample in dataset["train"]:
    train_conversation = sample["text"]
    ...

for sample in dataset["test"]:
    test_conversation = sample["text"]
    ...

Important notes

All datasets were normalized in text files so that:

  • Conversations are separated by a single blank line.
  • Each line corresponds to a single speech turn.
  • Each line begins with a speaker label of the form "[***:]".
  • When speaker names are anonymized or otherwise unknown, speakers are distinguished by numbers in the following format: "[speaker001:]", "[speaker002:]", …
    Otherwise, speakers are labeled with their names or roles, e.g. "[Paul:]", "[François Mitterrand:]", "[M. le président:]".
  • There are no parentheses: special annotations are always between square brackets.
  • Commong tags include:
    • "[PII]": Personally Identifiable Information (anonymized name...)
    • "[NOISE]": distinct ambient noises
    • "[LAUGHTER]": laughter
  • Depending on the data source, data may or may not include punctuation marks and upper case letters.
  • The data were normalized in various ways including unicode NFC normalization, conversion of unbreakable spaces to spaces, and standardization of punctuation marks (… -> ..., «/»/“/”/″/„ -> ").

Those details are described in the paper: « The Claire French Dialogue Dataset » (2023).

License

Given that some of the corpora used for training are only available under CC-BY-NC-SA licenses, Claire-Dialogue-French-0.1 is made available under the CC-BY-NC-SA 4.0 license.

Citations

When using the CFDD corpus, please cite the following paper:

✍ Julie Hunter, Jérôme Louradour, Virgile Rennard, Ismaïl Harrando, Guokan Shang, Jean-Pierre Lorré (2023) The Claire French Dialogue Dataset

@misc{openllm2023claire,
      title={The Claire French Dialogue Dataset}, 
      author={Julie Hunter and Jérôme Louradour and Virgile Rennard and Ismaïl Harrando and Guokan Shang and Jean-Pierre Lorré},
      year={2023},
      eprint={2311.16840},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}

This paper in turn provides the requested citations for all of the original corpora. The same references are also listed below.

Contact

contact@openllm-france.fr

Downloads last month
16
Edit dataset card