NRC Emotion Lexicon (EmoLex): Word-Emotion Association

# NRC Emotion Lexicon (EmoLex): Word-Emotion Association
# Based on Mohammad & Turney (2013), "Crowdsourcing a Word-Emotion
# Association Lexicon," Computational Intelligence 29(3).
#
# Annotation design (per target word):
#   1. WORD-CHOICE / SENSE question: the annotator is shown the target word
#      and four candidate words, and picks the one closest in meaning. This
#      serves two purposes at once - it gently disambiguates the intended
#      sense of the target word, and it screens out inattentive annotators
#      (the "gold" near-synonym is known in advance).
#   2. POLARITY: how positive and how negative the word is.
#   3. EIGHT EMOTIONS (Plutchik's basic emotions): for each, how strongly
#      the word is associated with that emotion.
#
# Association strength scale (used for polarity and every emotion):
#   Not at all  -  Weakly  -  Moderately  -  Strongly
#
# This is Mohammad's rating-scale (NON Best-Worst-Scaling) design; his later
# lexicons (NRC-VAD, Affect Intensity, WorryWords) moved to Best-Worst
# Scaling for finer, more reliable real-valued scores.
#
# Annotation guidelines:
#   1. Answer the word-choice question first; it fixes the sense you should
#      judge for the rest of the item.
#   2. Judge the word's general association with each emotion, not whether
#      it literally names the emotion.
#   3. A word can be associated with several emotions at once (e.g., "gift"
#      -> joy, surprise, trust, positive) or none.
#   4. "Not at all" is a valid and common answer - most words evoke only a
#      few emotions.

annotation_task_name: "NRC EmoLex Word-Emotion Association"
task_dir: "."

data_files:
  - sample-data.json

item_properties:
  id_key: "id"
  text_key: "prompt"
  text_display_key: "prompt"
  list_display_key: "choices"

output_annotation_dir: "annotation_output/"
output_annotation_format: "json"

annotation_schemes:
  # Step 1 - sense disambiguation / attention check
  - annotation_type: radio
    name: closest_meaning
    description: "Which of the four candidate words is CLOSEST in meaning to the target word?"
    labels:
      - "Candidate 1"
      - "Candidate 2"
      - "Candidate 3"
      - "Candidate 4"

  # Step 2 - polarity
  - annotation_type: radio
    name: positive
    description: "How POSITIVE (good, favorable) is the target word?"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: negative
    description: "How NEGATIVE (bad, unfavorable) is the target word?"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  # Step 3 - eight basic emotions
  - annotation_type: radio
    name: anger
    description: "Association with ANGER"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: anticipation
    description: "Association with ANTICIPATION"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: disgust
    description: "Association with DISGUST"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: fear
    description: "Association with FEAR"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: joy
    description: "Association with JOY"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: sadness
    description: "Association with SADNESS"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: surprise
    description: "Association with SURPRISE"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

  - annotation_type: radio
    name: trust
    description: "Association with TRUST"
    labels: ["Not at all", "Weakly", "Moderately", "Strongly"]

allow_all_users: true
instances_per_annotator: 100
annotation_per_instance: 5
allow_skip: true
skip_reason_required: false