metadata

license: cc-by-sa-4.0
task_categories:
  - summarization
language:
  - fr
tags:
  - NLP
  - Debates
  - Abstractive_Summarization
  - Extractive_Summarization
  - French
pretty_name: FREDsum
size_categories:
  - n<1K

Dataset Summary

The FREDSum dataset is a comprehensive collection of transcripts and metadata from various political and public debates in France. The dataset aims to provide researchers, linguists, and data scientists with a rich source of debate content for analysis and natural language processing tasks.

Languages

French

Dataset Structure

The dataset is made of 144 debates, 115 of the debates make up the train set, while 29 make up the test set

Data Fields

id : Unique ID of an exemple
Transcript : The text of the debate
Abstractive_1-3 : Human summary of the debate. Abstractive summary style goes from least to most Abstractive - Abstractive 1 keeps names to avoid coreference resolution, while Abstractive 3 is free form
Extractive_1-2 : Human selection of important utterances from the source debate
Community 1-2 : Abstractive communities linking each of the abstractive sentences to the supporting extractive ones. Community 1 represents the linking between Abstractive 1 and Extractive 1, while Community 2 represents the linking between Abstractive 3 and Extractive 2

Data splits

train : 115
test : 29

Licensing Information

non-commercial licence: CC BY-SA 4.0

Citation Information

If you use this dataset, please cite the following article:

Virgile Rennard, Guokan Shang, Damien Grari, Julie Hunter, and Michalis Vazirgiannis. 2023. FREDSum: A Dialogue Summarization Corpus for French Political Debates. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 4241–4253, Singapore. Association for Computational Linguistics.