RESEARCH HIGHLIGHT

Teaching AI to draw a “map of emotions”

A new way to organize 99 emotion words — simply by prompting GPT-4

How similar are “joy” and “surprise”? And how far apart are “joy” and “sadness”? We naturally organize the words for our emotions according to how close or how different their meanings are. Yet as the number of emotion words grows, the combinations that must be compared increase sharply, making the whole picture hard for people alone to examine in detail.

Researcher Ke Han and Associate Professor Eiji Watanabe of the Laboratory of Neurophysiology at the National Institute for Basic Biology had GPT-4 arrange emotion words in space to investigate how humans organize them. As a result, “prompting” — asking GPT-4 in ordinary sentences — reproduced a structure closely resembling the arrangements made by people. Moreover, when the set was expanded to 99 words, a structure related to emotional “arousal”, hard to see with only a few words, emerged.

The study suggests that the structure of emotion concepts can be examined using nothing but questions posed in words — with no specialized AI training and no large-scale computing facilities.

01Emotions can be expressed not only by “type” but by “position”

Emotions can be sorted into types such as “joy”, “anger”, and “sadness”. Psychology also offers an approach that represents emotions along several continuous axes. Three of them are representative.

Pleasure
PLEASANT / UNPLEASANT
Is it pleasant, or unpleasant?
Arousal
EXCITED / CALM
Is it excited, or calm?
Dominance
IN CONTROL / OVERWHELMED
Do you feel in control of the situation, or overwhelmed by it?

For example, “excitement” and “serenity” are both relatively pleasant, yet they differ greatly in arousal. “Rage” and “sadness” are both unpleasant, but again differ in arousal. Such structure has mainly been studied by asking people to rate how similar emotion words are. However, an experiment comparing all of many emotion words places a very heavy burden on participants.

02Placing three emotion words on a map

The research team devised a task in which three emotion words are placed at once on a grid on screen. Words felt to be similar in meaning go close together; words felt to be different go far apart. Placing three words at once expresses three relationships simultaneously. In the first study, 89 participants and GPT-4 performed the same task using six basic emotions, and the results were compared with the conventional “word embedding” method that represents emotion words as arrays of numbers.

FIG. 1 The study at a glance — placing words on a “map” to measure distance
1
Present three emotion words
Joy Surprise Sadness
2
Place them on a grid
similar close, different far
Joy Surprise Sadness
3
Measure with three methods
Humans
89 participants place them
GPT-4
prompted to place them
Word embedding
distance from numbers
4
Build a map of emotions from the “semantic distance” between words
Joy Surprise Sadness
Three emotion words are placed on a grid at once. The task is run by humans, by GPT-4, and by word embeddings, and the “semantic distance” between words is used to draw a map of emotions.

03GPT-4’s arrangement closely resembled that of humans

For the six basic emotions, humans and GPT-4 both formed two groups: “joy · surprise” and “anger · fear · disgust · sadness”. By contrast, word embeddings placed “joy” and “anger” in the same group, giving a structure different from humans. The distances produced by humans and GPT-4 were strongly correlated, and the responses among human participants themselves agreed very closely.

This does not mean, however, that GPT-4 experiences the same emotions as humans. GPT-4 is thought to have produced human-like arrangements by drawing on the relationships among emotion concepts contained in human language.

FIG. 2 How the six basic emotions group together — humans, GPT-4, word embeddings
Humans
Joy Surprise
Anger Fear Disgust Sadness
GPT-4 prompting
Joy Surprise
Anger Fear Disgust Sadness
Word embedding
Joy Anger
Surprise Fear Disgust Sadness
Humans and GPT-4 grouped “joy” and “surprise” together. Word embeddings, by contrast, placed “joy” and “anger” in the same group — a structure different from humans. (Dashed outlines mark clusters of semantically close emotion words.)

04Lining up 99 emotion words revealed a hidden axis

Next, the team expanded the set to 99 emotion words. Having human participants repeatedly compare all relationships among 99 words is practically very difficult, but with GPT-4 the same task can be repeated across many combinations. The analysis showed that, with both prompting and word embeddings, the emotion words split mainly into two large groups related to “pleasure” and “dominance”, and that GPT-4’s arrangement separated the two groups more clearly.

On closer inspection, GPT-4’s arrangement also revealed changes corresponding to “arousal”. This appeared not as the main axis dividing the large groups, but as an axis that finely divides the interior of each group. The structure was not clear with only six basic emotions and was detected only once the set was expanded to 99 words.

FIG. 3 Vocabulary size changes the structure you see — 6 vs. 99 words (conceptual)
6 words Two large groups visible / arousal unclear
Pleasant emotions
Unpleasant emotions
99 words Arousal appears inside each group
High
Arousal
Low
Pleasant emotions
ExcitementEcstasyHigh
ContentmentSerenityLow
Unpleasant emotions
RageTerrorHigh
SadnessGloomLow
With only a few emotion words, the large “pleasant vs. unpleasant” difference is easy to find, but finer differences in “arousal (intensity)” are easily missed. Expanding to 99 words reveals high/low arousal inside each group. *A conceptual diagram illustrating the meaning of the results, not a faithful reproduction of the scatter plot.
FIG. 4 A map of the 99 emotion words (representative words shown)
Unpleasant Pleasant ← Unpleasant  horizontal: pleasure  Pleasant → ← Low  vertical: arousal  High → Panic Rage Terror Gloom Sadness Disappointment Excitement Ecstasy Elation Relief Contentment Serenity
Pleasant emotions Unpleasant emotions Thick outline = high arousal Thin outline = low arousal
Representative words drawn from the 99. The horizontal direction corresponds to “pleasure”; the vertical direction and outline thickness correspond to “arousal”. Inside the two large clusters (dashed), arousal spreads from low to high. Positions are a schematic layout based on the analysis.

05The number of words you examine changes the structure you see

These results show that the scale of the vocabulary matters when studying the structure of emotion. Using only a few representative emotion words makes large differences such as “pleasant vs. unpleasant” easy to find, but finer differences such as “excited vs. calm” may be overlooked. Part of the reason existing emotion theories differ from one another may lie in the number and choice of emotion words used in each study.

06Using AI as a “measuring instrument”, not an “answer machine”

Tuning word embeddings to a research purpose can require expertise in machine learning, training data, and a computing environment. In the present method, by contrast, researchers give GPT-4 a task in ordinary sentences and analyze its responses. This may make it an accessible method for researchers who do not specialize in machine learning — for example, in psychology or linguistics.

In the future, the team plans to examine whether the same structure appears in different languages, including Japanese, and how the arrangement of emotion words changes across cultures. Note that this study addressed not the actual experience of emotion, but the semantic and conceptual structure of the words that express emotion. How far the structure shown by GPT-4 corresponds to actual emotional experience, or to emotional representations in the brain, remains to be verified.

Glossary
Emotion word
A word expressing an emotion or emotional state, such as “joy”, “fear”, or “relief”.
Prompting
Giving a large language model instructions or questions in ordinary sentences to obtain answers. This study used GPT-4 with no additional training.
Word embedding
A method that represents a word’s meaning with many numbers. Words with similar meanings tend to sit at nearby positions in the numeric space.
PAD model
A psychological model that represents emotion along three axes: Pleasure, Arousal, and Dominance.
ACCESS THE PAPER

Ke, H., Watanabe, E. Mapping 99 emotion terms with GPT4 prompting reveals nuanced semantic conceptual structure. Scientific Reports (2026). https://doi.org/10.1038/s41598-026-60536-4