license: pddl
language:
- fr
pretty_name: PIAF v1.2
task_categories:
- question-answering
size_categories:
- 1K<n<10K
PIAF - The French-language Question-Answer Dataset
(Description translated to English; apologies for any mistakes. Obtained from https://www.data.gouv.fr/datasets/piaf-le-dataset-francophone-de-questions-reponses/)
PIAF: Building an Open French-Language Dataset for AI
The use of artificial intelligence in public policy is often seen as an opportunity to query documentary texts and create automated question-and-answer (QA) tools for users. Querying labor laws in natural language, providing a chatbot for a given service, developing high-performance search engines, and improving knowledge management are all activities that require high-quality training datasets to develop question-and-answer algorithms. The PIAF dataset is a public and open French-language training dataset that allows these algorithms to be trained.
Inspired by SQuAD, the well-known English-language QR dataset, we aimed to build a similar dataset that would be open to everyone. The protocol we followed is very similar to that of the first version of SQuAD (SQuAD v1.1). Nevertheless, some modifications had to be made to adapt to the characteristics of the French Wikipedia. Another major difference is that we do not employ micro-workers through crowdsourcing platforms.
PIAF: A Shared Tool of LAB IA
In March 2018, France launched its national strategy for artificial intelligence. Led by the Interministerial Directorate for Digital Affairs, this strategy comprises three components: research, the economy, and public sector transformation.
Since data policy is a major focus of artificial intelligence development, the Etalab mission is leading the establishment of an interministerial “Lab IA,” whose mission is to accelerate the deployment of AI in government agencies through three main activities:
- Building a core team to internalize skills and expertise in AI
- Supporting AI projects in government agencies through calls for expressions of interest
- Co-developing shared tools that can be used as openly as possible
The PIAF project is one of LAB AI's shared tools.
Description of the data made available
After several months of annotation, we have a robust and open-source annotation platform, a sufficient number of annotations, and a well-established and innovative approach to community engagement and collaborative participation within the French administration. The dataset follows the SQuAD v1.1 format. PIAF v1.2 contains 9225 question/answer pairs. It is a JSON-type file. A text file exemplifying the schema is included below. This file can be used to generate and evaluate Question-Answer models.
{
version: "Version du dataset"
data:[
{
title: "Titre de l'article Wikipedia"
paragraphs:[
{
context: "Paragraph de l'article"
qas:[
{
id: "Id du pair question-réponse"
question: "Question"
answers:[
{
"answer_start": "Position de la réponse"
"text": "Réponse"
}
]
}
]
}
]
}
]
}
Thank you to our 500 contributors!
We are deeply grateful to our contributors who have kept this project going on a voluntary basis until now.
Links
Information about the protocol used, project news, the annotation platform, and the associated code can be found here: