I am returning to the Speech and Language Processing textbook (3rd edition) by Daniel Jurafsky and James Martin. It’s always good to pick up new concepts / areas from a textbook as they are written to cover the most ground of a specific area. You can then dive into more research afterwards 🙂 Over the next few posts, I will be covering Chapter 21: Lexicons for Sentiment, Affect, and Connotation. As usual, I will convert all my notes into questions to better organise the information I am consuming.


What is affective, affective lexicons, and connotations?

Affective means emotion, sentiment, personality, mood, and attitudes. Below is an example of the Schere typology of different affective states. Affective lexicons means words that have a strong affective meanings. Connotations means the aspects of a word’s meaning that relates to a person emotions, sentiment, opinions, or evaluations.

What are the three ways to build lexicons?
  1. Human labelling

  2. Semi-supervised

  3. Supervised

What are the applications of emotion detection?

It can help businesses detect the emotions of customer reviews and responses and identify which areas of the business are going well and bad. Emotion detection can also help teachers measure how engaging the students are. Last but not least, emotion detection can play a major role in healthcare for detecting depression and / or suicidal intent.

21.1 Defining Emotion

What are the two families of theories of emotion?

The first family viewed emotions as fixed atomic units. Ekman (1999) proposed the 6 emotions: surprise, happiness, anger, fear, disgust, and sadness. Another atomic theory is Plutchik’s (1980) wheel of emotion, consisting of 8 basic emotions in four opposing pairs. See below the figure of the wheel of emotion.

The second family viewed emotions as a space of 2 or 3 dimensions. The three dimensions are:

  1. Valence. The pleasantness of the stimulus

  2. Arousal. The intensity of emotion from the stimulus

  3. Dominance. The degree of control exerted by the stimulus

Sentiment analysis is the special case of the second view of emotions. The valence dimension measures how pleasant or unpleasant a word is, which it’s used to measure sentiment.

21.2 Available Sentiment and Affect Lexicons

What’s the most basic lexicons?

The most basic lexicons of labelled words are the ones that only work in one dimension. The simplest lexicon is a list of positive words and a list of negative words and it’s used to measure “sentiment”.

Describe the NRC VAD lexicon.

The NRC Valence, Arousal, and Dominance (VAD) lexicon assigns the valence, arousal, and dominance scores to 20,000 words.

What is EmoLex?

EmoLex is the NRC Word-Emotion lexicon that uses the Plutchik’s wheel of emotion. It includes 14,000 words covering frequent nouns, verbs, adverbs, and adjectives. See figure below for examples. The examples below are binary switch of each emotion for a single word. The NRC Emotion/Affect Intensity lexicon goes a step further and assign intensity score of anger, fear, joy, and sadness to words.

What are concrete and abstract words?

Concrete words are words like apple and house whereas abstract words are words like belief and although.



Data Scientist

Leave a Reply