GermanQuAD: German Extractive Question Answering

# GermanQuAD - German Extractive Question Answering
# Based on Moeller et al., MRQA 2021
# Paper: https://aclanthology.org/2021.mrqa-1.4/
# Dataset: https://huggingface.co/datasets/deepset/germanquad
#
# GermanQuAD is the German extractive QA dataset from deepset. Trained
# annotators read a passage from German Wikipedia, write a self-sufficient
# question (deliberately reformulated to reduce lexical overlap with the
# passage), and mark the answer span. GermanQuAD 1.0 contains no
# SQuAD-2.0-style unanswerable questions.
#
# Annotation workflow (paper sections 2.1-2.3):
# 1. Read the German Wikipedia passage.
# 2. Write a self-sufficient question, reformulating with synonyms and altered
#    syntax so the answer cannot be found by simple word matching. Complex
#    questions whose answers span multiple sentences are encouraged.
# 3. Highlight the answer span in the passage.

annotation_task_name: "GermanQuAD - German 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 self-sufficient German question
  - annotation_type: text
    name: question
    description: "Formulieren Sie eine eigenständige Frage, die aus dem Absatz beantwortbar ist (Write a self-sufficient German question answerable from the passage)"

  # Step 2: Highlight the answer span
  - annotation_type: span
    name: answer_span
    description: "Markieren Sie die Textstelle, die die Frage beantwortet (Highlight the span that answers the question)"
    labels:
      - "Antwort (Answer)"
    label_colors:
      "Antwort (Answer)": "#3b82f6"

annotation_instructions: |
  You will be shown a passage from German Wikipedia. Following the GermanQuAD
  data-collection protocol (Möller et al., 2021):

  1. Read the passage carefully.
  2. Write a question in German that can be answered from the passage alone. Make
     it self-sufficient — a reader should not need any outside information. To
     avoid the lexical-overlap shortcut, reformulate the question using synonyms
     and altered syntax rather than reusing the passage's wording. Complex
     questions whose answers span multiple sentences are encouraged.
  3. Highlight the span in the passage that answers your question. Answers can
     range from a single entity to several sentences.

  Every question must be answerable from the passage — GermanQuAD 1.0 contains 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;">Absatz (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