Coding taxonomies
Robert Plutchik's eight primary emotions, in four opposing pairs, at three intensities each — as something you can operate. Select any emotion to see its opposite across the wheel and the dyads it forms, then take the taxonomy away as a coding label set.
Free · No sign-up · Runs entirely in your browser
Clear expressed pleasure or happiness.
Pleasure, happiness or delight expressed by the speaker.
Intensity ladder
Signal words
4
happiness, delight, elation, pleasure
Dyads formed with its neighbours
Pick a granularity and take the taxonomy with you — every label with a stable code, its family, and a definition written for a coder.
8 labels. Highest agreement, least resolution. The usual choice for a first pass.
| Label | Code | Family |
|---|---|---|
| Joy | JOY | Joy |
| Trust | TRU | Trust |
| Fear | FEA | Fear |
| Surprise | SUR | Surprise |
| Sadness | SAD | Sadness |
| Disgust | DIS | Disgust |
| Anger | ANG | Anger |
| Anticipation | ANT | Anticipation |
More categories is not more accuracy. Every step up in granularity costs you agreement: eight primaries are far easier for two coders to apply consistently than twenty-four intensity terms, and the mild ring is where agreement falls apart first, because words like “interest” and “acceptance” overlap ordinary description. If you add the dyads, decide and write down whether a dyad replaces its two parents or sits alongside them — that single unstated decision produces more disagreement than any other choice in an emotion scheme. Pilot the label set before you commit to it, and measure agreement rather than assuming it.
Structure follows Plutchik (1980, 2001): eight primaries in four opposing pairs, three intensities each, and eight primary dyads from adjacent pairs. Plutchik also described secondary and tertiary dyads from emotions further apart on the wheel; only the primary dyads are included here, because they are the ones with settled conventional names. The category definitions and coding notes are written for annotators and are not reproduced from any copyrighted manual.
Reference · for the curious
Robert Plutchik's wheel is the most reproduced diagram in the psychology of emotion, and it is usually encountered as a picture. That is a shame, because the picture is an argument. The circular arrangement, the pairing across the diameter, the way intensity increases toward the centre and categories blend at their edges — each is a claim about how emotions relate, and each has consequences for anyone using the model to label data.
Plutchik (1980) proposed eight primary emotions on evolutionary grounds: each is a behavioural response with survival value, present across species in some form. Joy, trust, fear, surprise, sadness, disgust, anger and anticipation.
They are arranged in four opposing pairs — joy against sadness, trust against disgust, fear against anger, surprise against anticipation — and the opposition is the part that makes it a model rather than a list. Fear and anger sit opposite each other because they are the two responses to threat: withdraw or confront. Trust and disgust are the two responses to something offered: take it in or expel it. Turn the opposition axes on in the wheel above and the geometry is the theory.
This is also where Plutchik differs most usefully from Paul Ekman, whose six basic emotions were derived from facial expression research rather than evolutionary function and include no oppositional structure. Our comparison of Ekman versus Plutchik works through what you gain and lose choosing between them, and Ekman's basic emotions covers the six-category alternative on its own terms.
Each primary varies in intensity, and Plutchik gave each level a distinct term — serenity, joy, ecstasy; annoyance, anger, rage; apprehension, fear, terror. The wheel puts the most intense terms at the centre and the mildest at the rim, which is why it is drawn as a three-ring target rather than a flat pie.
For a coding scheme this is a genuine offer and a genuine trap. The offer is a principled ordinal dimension: you can record that someone was annoyed rather than enraged without inventing your own severity anchors. The trap is that the mild ring is where agreement goes to die. “Interest” and “acceptance” and “pensiveness” describe states so close to ordinary conversational neutrality that two competent coders will read the same sentence differently, and no amount of guideline writing fully fixes it. If you use the 24-label version, expect your disagreements to cluster there and plan the adjudication for it.
Real speech is rarely one clean emotion. Plutchik's answer is the dyad: two adjacent primaries combining into a nameable third thing. Joy plus trust is love. Anticipation plus joy is optimism — which, in interview data, is probably the single most common emotional stance a person takes, and the one most often flattened into plain “joy” by a scheme that has nowhere else to put it. Disgust plus anger is contempt; fear plus surprise is awe.
Adding the eight dyads to your label set gives you somewhere to put blends. It also forces a decision that schemes routinely leave unstated: does a dyad replace its two parents, or sit alongside them? If a coder can tag a passage as optimism, or as anticipation, or as both, you have three defensible answers to the same passage and your agreement figures will show it. Decide, write it into the codebook, and put an example of each in the guidelines. This is the sort of edge case annotation guidelines that actually work exist to settle, and turning a scale into a codebook covers the mechanics.
The tool exports at three sizes because the choice is real and consequential. Eight primaries is the sensible default: coders apply it consistently, and it is coarse enough that most disagreements are genuine ambiguity rather than terminology. Twenty-four with intensity buys resolution at a measurable cost in agreement. Thirty-two with dyads is appropriate when mixed emotion is the object of study rather than a nuisance.
The general principle is not specific to emotion: finer category systems lower agreement, and the resolution you gain is only worth having if it survives the reliability check. Run the coarse version in a pilot round first, compute agreement with the inter-rater reliability calculator — Krippendorff's alpha if coders skipped items, and a paradox-resistant coefficient if one category dominates, as neutral usually does in interview data — and only then decide whether to go finer. If your emotion labels turn out to be ordered, weighted kappa is the coefficient that credits near-misses, which matters a great deal on the intensity rings.
One more thing that will bite you: disagreement about emotion is often signal, not noise. Two coders reading a passage as sadness and anger respectively may both be right about an ambivalent speaker. When to preserve annotator disagreement rather than reconcile it is worth reading before you build a pipeline that majority-votes it away.
Text loses most of what tells a human which emotion is present. No prosody, no face, no timing. What remains is lexical choice, syntax, what the speaker says about their own state, and what a reader infers from the situation described — and that last one is where coders diverge, because inference is where their own knowledge enters.
Two rules keep it tractable. Code what the speaker expresses, not what you would feel in their position: a flatly delivered account of a bereavement is not automatically grief on the page. And do not code emotion words that appear only in the interviewer's question — a subject who answers “yes” to “were you frightened?” has confirmed fear, which is a different observation from having expressed it, and your codebook should say which one counts.
The taxonomy is hosted in the Tagaroo Scale Library with the eight primaries as an annotation scheme; our overview of Plutchik's wheel covers the model in more depth.
Plutchik's model is a conceptual framework, free to operationalise in your own scheme with a citation to Plutchik (1980). It is not a licensed instrument, and there is no fee or permission to obtain. What is not free is any particular copyrighted manual or scoring dictionary built on the model — so the definitions and coding notes here are written for annotators from the structure of the model itself, not reproduced from a published manual. The label sets you export are yours to adapt.
This is a reference and codebook-building aid for education and research support. It is not a clinical instrument, it does not assess anyone's emotional state, and it is not a self-assessment questionnaire. Everything runs in your browser; nothing is transmitted or stored.
Joy, trust, fear, surprise, sadness, disgust, anger and anticipation. Robert Plutchik proposed them in 1980 as evolutionarily adaptive states, arranged on a wheel in four opposing pairs: joy against sadness, trust against disgust, fear against anger, and surprise against anticipation. The arrangement is the substance of the model rather than decoration — placing each emotion opposite the one it conflicts with is a claim that the eight are not an arbitrary list but a structure with poles.
Each primary exists at three intensities, and Plutchik gave each level its own word. Joy runs serenity, joy, ecstasy; anger runs annoyance, anger, rage; fear runs apprehension, fear, terror; sadness runs pensiveness, sadness, grief; trust runs acceptance, trust, admiration; disgust runs boredom, disgust, loathing; surprise runs distraction, surprise, amazement; and anticipation runs interest, anticipation, vigilance. The wheel is drawn with the most intense terms at the centre and the mildest at the rim, so moving inward is moving up in intensity.
Dyads are blends of two emotions. The eight primary dyads come from adjacent pairs: joy plus trust gives love, trust plus fear gives submission, fear plus surprise gives awe, surprise plus sadness gives disapproval, sadness plus disgust gives remorse, disgust plus anger gives contempt, anger plus anticipation gives aggressiveness, and anticipation plus joy gives optimism. Plutchik also described secondary and tertiary dyads from emotions further apart on the wheel, but only the primary dyads have settled conventional names, which is why this tool shows those eight.
It is one of the two default choices, alongside Ekman's six basic emotions, and it trades agreement for resolution. Eight categories give you more nuance than six and the dyads give you a principled way to handle mixed states, but every added category is a category two coders can disagree about. Agreement typically falls apart first on the mild intensity ring, where terms like interest and acceptance overlap ordinary description rather than marking an emotional state. Start with the eight primaries, pilot them, measure agreement, and only add intensity or dyads if the coarser scheme genuinely fails to capture what you need.
The mild intensity ring, and "interest" and "acceptance" in particular. Those two describe states so close to ordinary conversational neutrality that two competent coders routinely read the same sentence differently — "pensiveness" and "boredom" have the same problem to a lesser degree. The reason is structural rather than a failure of training: the mild terms overlap plain description, so a coder is deciding whether an emotion is present at all, not which emotion it is. If you code the 24-label version, expect your disagreements to cluster on that outer ring and plan the adjudication for it. The eight primaries alone avoid the problem almost entirely, which is why they are the sensible default for a first pass.
Yes. Plutchik's model is a conceptual framework rather than a licensed instrument, so it is free to operationalise in your own coding scheme with a citation to Plutchik (1980). What you cannot do is reproduce a copyrighted manual or scoring dictionary built on top of it. The definitions and coding notes in this tool are written for annotators from the structure of the model, not copied from any published manual, so the label sets you export here are yours to adapt.
Written by Enrique Gutiérrez, PhD (Computer Science) — founder of Tagaroo and Associate Professor of Computer Science, working on inter-rater reliability, measurement and annotation methodology (ORCID).
Last verified: 29 July 2026. Formulas, thresholds and cited figures on this page were checked against their original sources on that date. Every calculation runs in your browser; nothing you enter is transmitted or stored.
Tagaroo loads this taxonomy as an anchored codebook and computes inter-rater agreement while your coders work — so the granularity you choose is a decision you can measure rather than one you find out about at the end.
Interactive Plutchik wheel of emotions · tagaroo.ai/materials/emotion-wheel · figures verified against primary sources 2026-07-29. Educational scoring aid, not a diagnosis, and not reviewed by a licensed clinician.