Chapter 23
Lexicons for Sentiment, Affect, and Connotation
Chapter: Lexicons for Sentiment, Affect, and Connotation, in Daniel Jurafsky and James H. Martin, Speech and Language Processing, 3rd edition draft, August 19, 2026.
Topics to learn
Page numbers refer to the linked chapter PDF.
- What is being labeled p. 2
- Definitions of emotion p. 2
- Sentiment and affect lexicons already available p. 4
- Building a lexicon p. 5
- Human labeling p. 5
- Semantic-axis methods p. 7
- Label propagation p. 8
- Other semi-supervised methods p. 9
- Supervised word sentiment p. 10
- Log-odds scoring with a Dirichlet prior p. 12
- Using a lexicon p. 15
- Sentence or document sentiment p. 15
- Affect recognition p. 16
- Entity-centric affect p. 17
- Connotation frames p. 17
FAQ
Is a positive lexicon hit the same as a positive sentence?
No. Composition, negation, and the target of the sentiment can reverse or redirect a word's prior polarity. The lexicon is a feature, not the classifier.
What does label propagation assume?
That words in similar contexts, or linked by a similarity graph, often share an affect label. The assumption fails when neighbors are opposites or when the seed labels are narrow.
Why use a Dirichlet prior in a log-odds comparison?
Raw counts make rare words look extreme. The prior pulls sparse counts toward a background rate so a word seen twice is not treated like a stable signal.
What is a connotation frame adding?
It records who is judged, and how, in a predicate's arguments. A word can be negative for one participant and not for another. A single polarity score hides that.