This document summarizes research on constructing a lexicon called CLex that explores associations between colors, concepts, and emotions based on crowdsourced data. The lexicon contains over 15,000 annotations linking colors to concepts and emotions. The research found cultural differences in color-emotion associations between the US and India and identified frequent color-concept pairs. The lexicon could help applications like sentiment analysis by capturing meaning conveyed through color terms.

1. CLex: A Lexicon for Exploring Color, Concept and
Emotion Associations in Language
Svitlana Volkova, William Dolan, Theresa Wilson
I. Motivation III. Data Analysis
People use color terms to describe visual characteristics of objects. Lexicon Statistics Top 10 Color Term Collocations for Berlin and Kay (1988) Colors
sky - blue sun - yellow CLex contains 15,200 annotations: White – 0.62 Black – 0.43 Red – 0.59 Green – 0.54 Yellow – 0.63 Blue – 0.55
§ 2,3K color term types • off, antique, • light, • dark, light, • dark, light, • light, dark, • light, sky,
snow - white grass - green § 1,9K concept types half, dark, blackish dish brown, olive, green, pale, dark, royal,
black, bone, brown, brick, yellow, lime, golden, navy, baby,
§ 3,4K affect term/emotion types milky, pale, brownish, orange, forest, sea, brown, grey,
People take advantage of color terms to strengthen the message and pure, silver brown, jet, brown, dark olive, mustard, purple,
dark, green, indian, dish, pea, dirty orange, cornflower,
convey emotion in natural interaction. Color dimensions: off, ash, crimson, deep, bright violet
§ hue, darkness, brightness, saturation black grey bright
Colors are indicative and have an effect on our feelings and emotions. NA Other *overall agreement on color terms (exact match – 0.49, substring – 0.46), free-marginal kappa (exact match – 0.46, substring – 0.42)
§ positive emotions – joy, trust, admiration (Ortony et. al, 1988); 8% 12%
§ negative emotions – aggressiveness, fear, boredom. Male
India
Female
51%
41%
US
38%
49% Cross-Cultural Differences in Color-Emotion Associations: US and India, %
Color-concept-emotion associations [brown-darkness-boredom] vary
by culture and are influenced by the beliefs of the society:
§ green – danger in Malaysia, envy in Belgium, happiness in Japan. Affect Terms
(Sable and Akcay, 2010). § feelings, emotions, attitudes, moods
Trust
Some expressions involving colors share meaning across languages:
anticipat.
Joy
surprise
US: 33 Surprise
sadness
disgust
§ “read heat”, “blue blood”, “white collar”. US: 331
anger
I: 47
US: 18
trust
fear
I: 154
joy
I: 12
Deeper understanding of the associations between colors, concepts n - 0.3 - 0.8 1.0 - - -
and emotions [red – blood – anger] may be helpful for: n 3.6 24.0 2.3 43.0 0.2 - - -
§ Sentiment analysis: understanding emotion new article evokes. n 43.4 0.3 1.1 8.9 0.2 1.2 - -
§ Machine translation: as a source of paraphrasing. n 0.3 0.6 11.2 2.0 3.4 3.5 3.3 5.3
§ Textual entailment: question answering in dialog systems n 0.3 0.3 1.1 1.2 5.7 1.2 6.7 5.3
§ Human-computer interactions in real- and virtual word domains: n 0.3 4.2 1.1 0.4 4.2 17.4 6.7 -
online shopping, avatar construction in gaming environments n 3.3 11.4 24.7 6.1 4.2 8.1 3.3 5.3
Anger Sadness Fear
(clearer & natural descriptions “sky-blue shirt” vs. “blue shirt”). n 0.6 0.3 1.1 0.4 9.1 1.2 3.3 5.3 US: 133 US: 171 US: 95
n 0.3 2.2 3.4 0.8 4.4 1.2 6.7 - I: 160
I: 142
I: 105
n 1.5 0.3 1.1 0.4 4.0 5.8 13.3 15.8
II. Data Collection n 2.1 10.3 - 2.0 0.6 1.2 3.3 5.3
We use Amazon Mechanical Turk to collect color-concept and color-
emotion associations: Basic Emotions in Colors, % Top 10 Color-Concept Associations for Ambiguous Colors
8
Task 1: showed swatch of RGB value 6
Lips, 27 Hair, Task 2: showed swatch + facial feature 4
Affect Term Polarity, %
eyebrow
36 2
0
-2
Facial
mask, 27 -4
Eye
shadow, -6
26 -8
Skin, 18
Q1. How would you name this color? -10
Eyes, 18
Q2. What emotions does this color evoke? -12
152 RGB values of facial Q3. What concepts do you associate with it?
features of avatars
Example annotations: IV. Summary V. Related Work
Ø [R=222, G=207, B=186]
light golden yellow è purity, happiness è butter cookie, vanilla § Constructed large-scale color-concept-emotion lexicon by crowdsourcing § EmoLex (Mohammad, 2011)
gold è cheerful, happy è sun, corn § Studied cross-cultural differences in color-emotion associations (US and India) § WordNet-Affect (Strapparava and Valitutti, 2004)
golden è sexy è beach, jewelry § Identified frequent color-concept associations (may help to identify sentiment § General Inquirer (Stone et. al., 1966)
Ø [R=218, G=97, B=212] expressed by a concept) § Affective Forms of English Words (Stone et. al., 1966)
pinkish purple è peace, tranquility, stressless è Justin Bieber’s
Acknowledgements
headphones, someday perfume
§ Everyone in the NLP group at Microsoft Research for helpful discussion and feedback especially Chris Brocket, Piali Choudhury, and Hassan Sajjad.
pink è happiness è rose, bougainvillea
§ Natalia Rud from Tyumen State University, Center of Linguistic Education for helpful comments and suggestions.
§ Johns Hopkins University Human Language Technology Center of Excellence.