Lexicons for Sentiment, Affect, and Connotation

23.8 Lexicon-based methods for entity-centric affect

1. How does the entity-centric affect method of Field and Tsvetkov (2019) work?
   Training: for each word, average its contextual embeddings (e.g. BERT, no fine-tuning) and fit three regression models predicting its NRC VAD scores. Inference: embed an entity mention in context, apply the regressors, and average over coreferent mentions.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 23, §23.8, p. 535
2. What do Valence, Arousal and Dominance become when applied to people in entity-centric affect?
   Sentiment, Agency and Power.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 23, §23.8, p. 535
3. What does entity-centric affect show for characters in a film plot summary?
   Archetypes: protagonist and antagonist both have high power and agency but differ in sentiment; a love interest may have low power and agency but high sentiment.
   • Jurafsky & Martin, Speech and Language Processing, 3rd ed. draft, ch. 23, §23.8, pp. 535–536