FQuAD: French Question Answering Dataset

# FQuAD - French Question Answering Dataset
# Based on d'Hoffschmidt et al., arXiv 2020
# Paper: https://arxiv.org/abs/2002.06071
# Dataset: https://fquad.illuin.tech/
#
# FQuAD is the French extractive QA dataset from Illuin Technology, built on
# the SQuAD 1.1 protocol. University students read a paragraph from a French
# Wikipedia quality article, write a question whose answer is a span of text,
# and select the smallest span that answers it. FQuAD 1.0/1.1 contain no
# SQuAD-2.0-style unanswerable questions.
#
# Annotation workflow (paper section 3.2):
# 1. Read the French Wikipedia paragraph.
# 2. Write a question whose answer is a span within the paragraph.
# 3. Select the smallest span that contains the answer.
# 4. Repeat until 3-5 questions have been written for the paragraph.

annotation_task_name: "FQuAD - French Extractive QA"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "text"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

port: 8000
server_name: localhost

annotation_schemes:
  # Step 1: Write a French question answerable by a span
  - annotation_type: text
    name: question
    description: "Rédigez une question dont la réponse est un passage du texte (Write a question whose answer is a span in the passage)"

  # Step 2: Select the smallest answer span
  - annotation_type: span
    name: answer_span
    description: "Sélectionnez le plus petit passage qui contient la réponse (Select the smallest span containing the answer)"
    labels:
      - "Réponse (Answer)"
    label_colors:
      "Réponse (Answer)": "#3b82f6"

annotation_instructions: |
  You will be shown a paragraph from a French Wikipedia quality article. Following
  the FQuAD data-collection protocol (d'Hoffschmidt et al., 2020), which reuses the
  SQuAD 1.1 guidelines:

  1. Read the paragraph carefully.
  2. Write a question in French whose answer is a span of text within the
     paragraph. Aim for difficult questions rather than trivial word-matching.
  3. Select the smallest span in the paragraph that contains the answer.

  Repeat for several questions per paragraph. Every question must be answerable
  from the paragraph — FQuAD 1.0/1.1 contain no unanswerable questions.

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="background: #f0f9ff; border: 1px solid #bae6fd; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
      <strong style="color: #0369a1;">Paragraphe (Passage):</strong>
      <p style="font-size: 16px; line-height: 1.8; margin: 8px 0 0 0;">{{text}}</p>
    </div>
  </div>

allow_all_users: true
instances_per_annotator: 50
annotation_per_instance: 2
allow_skip: true
skip_reason_required: false