Word-Colour Associations: Crowdsourcing a Colour Lexicon

# Word-Colour Associations: Crowdsourcing a Colour Lexicon
# Based on Mohammad (ACL 2011, CMCL Workshop), "Colourful Language:
# Measuring Word-Colour Associations."
#
# Goal: build a lexicon of the colours that words evoke. Many concepts have
# strong colour associations (danger->red, growth->green, royalty->purple),
# and - crucially - even ABSTRACT concepts and emotions carry consistent
# colour associations. Such a lexicon supports information visualization,
# text-to-image, product/marketing design, and colour-emotion research.
#
# Annotation design (two parts per word):
#   1) A word-choice QUALITY-CONTROL / SENSE question. The annotator picks
#      the phrase closest in meaning to the target word. This (a) pins down
#      the intended word SENSE before the colour judgment, and (b) filters
#      out inattentive or unqualified annotators - responses that miss this
#      question are discarded. (This is the same guard used in Mohammad &
#      Turney's NRC Emotion Lexicon crowdsourcing.)
#   2) The COLOUR judgment. The annotator picks the single basic colour
#      most associated with the word (from the 11 Berlin & Kay basic colour
#      terms, plus "no colour association"), and rates how strong the
#      association is. Aggregating across annotators yields the word's
#      dominant colour and association strength.
#
# Annotation guidelines:
#   1. Answer the meaning question first, for the sense it disambiguates.
#   2. Choose the ONE colour you most associate with the word in that sense.
#      Go with your first instinct; there is no single "correct" colour.
#   3. If the word truly evokes no colour, choose "No colour association"
#      and set strength to "None".
#   4. Use Skip for unfamiliar words rather than guessing.

annotation_task_name: "Word-Colour Associations"
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:
  # ---- Part 1: sense / quality-control word-choice question ----
  - annotation_type: radio
    name: closest_meaning
    description: "Which phrase is CLOSEST in meaning to the target word? (sense & attention check)"
    labels:
      - "Choice 1"
      - "Choice 2"
      - "Choice 3"
      - "Choice 4"

  # ---- Part 2: the colour judgment ----
  - annotation_type: radio
    name: associated_colour
    description: "Which colour do you MOST associate with this word?"
    labels:
      - "Black"
      - "White"
      - "Red"
      - "Green"
      - "Yellow"
      - "Blue"
      - "Brown"
      - "Pink"
      - "Orange"
      - "Purple"
      - "Grey"
      - "No colour association"

  - annotation_type: radio
    name: association_strength
    description: "How strongly do you associate this word with that colour?"
    labels:
      - "Strong"
      - "Moderate"
      - "Weak"
      - "None"

allow_all_users: true
instances_per_annotator: 100
annotation_per_instance: 4
allow_skip: true
skip_reason_required: false