OffensEval - Offensive Language Target Identification

# OffensEval - Offensive Language Target Identification
# Based on Zampieri et al., SemEval 2020
# Paper: https://aclanthology.org/2020.semeval-1.188/
# Dataset: https://sites.google.com/site/offaborita/olid
#
# This task implements a multi-step offensive language annotation pipeline.
# Annotators first determine whether a social media post is offensive,
# then classify the target type (individual, group, other, or untargeted),
# and finally highlight the specific offensive spans and target mentions.
#
# Annotation Guidelines:
# 1. Read the post carefully and determine if it contains offensive language
# 2. If offensive, classify whether the offense targets an individual, group, or other entity
# 3. Use span annotation to highlight offensive expressions and target mentions
# 4. A post can be offensive without targeting anyone specific (untargeted)

annotation_task_name: "OffensEval - Offensive Language Target Identification"
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: Is the post offensive?
  - annotation_type: radio
    name: offensiveness
    description: "Is this post offensive?"
    labels:
      - "Offensive"
      - "Not Offensive"
    keyboard_shortcuts:
      "Offensive": "1"
      "Not Offensive": "2"
    tooltips:
      "Offensive": "The post contains any form of offensive language, insult, or profanity"
      "Not Offensive": "The post does not contain offensive language"

  # Step 2: What type of target?
  - annotation_type: multiselect
    name: target_type
    description: "If offensive, what type of target is addressed? (select all that apply)"
    labels:
      - "Individual"
      - "Group"
      - "Other"
      - "Untargeted"
    tooltips:
      "Individual": "The offense targets a specific named or unnamed individual"
      "Group": "The offense targets a group of people based on identity, affiliation, or characteristics"
      "Other": "The offense targets an organization, event, or abstract entity"
      "Untargeted": "The post is offensive but does not target any specific entity"

  # Step 3: Highlight offensive spans and target mentions
  - annotation_type: span
    name: offensive_spans
    description: "Highlight the offensive expressions and target mentions in the text"
    labels:
      - "Offensive Span"
      - "Target Mention"
    tooltips:
      "Offensive Span": "A word or phrase that constitutes the offensive expression"
      "Target Mention": "A word or phrase that refers to the target of the offense"

annotation_instructions: |
  You will be shown social media posts. Your task is to:
  1. Determine whether the post contains offensive language.
  2. If offensive, classify the target type (individual, group, other, or untargeted). You may select multiple types.
  3. Highlight specific offensive expressions and target mentions using span annotation.

  Offensive language includes insults, threats, profanity directed at someone, and derogatory language.
  A post can be offensive without targeting anyone specifically (e.g., general profanity or vulgarity).

html_layout: |
  <div style="padding: 15px; max-width: 800px; margin: auto;">
    <div style="background: #fef2f2; border: 1px solid #fecaca; border-radius: 8px; padding: 16px; margin-bottom: 16px;">
      <strong style="color: #991b1b;">Post:</strong>
      <p style="font-size: 16px; line-height: 1.7; 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