Human Emotion Ontology (HEO) · Schema version 1.0.0
| Capability (from the project brief) | Status | How |
|---|---|---|
| Isolated emotions | ✅ Strong | 242 emotion nodes with definitions, dimensions, appraisals, physiology, behavior |
| Subtle feelings | ✅ Strong | feeling nodes, gradation families (irritation→rage; unease→panic), low-arousal negative states |
| Emotional blends | ✅ Strong | blend_of edges with confidence-graded support (Plutchik dyads + modern blends) |
| Simultaneous emotions | ✅ Strong | co_occurs_with edges, meta.mixed_feelings, grp.mixed_emotion, evaluative-space design note |
| Emotional intensity | ✅ Partial | intensity poles as distinct nodes + arousal/dominance dimensions + free parameter in the state-vector model |
| Transitions between states | ✅ Strong | 159 transitions_to edges forming paths; often_follows/often_precedes for weaker regularities |
| Contextual changes | ✅ Partial | appraised_as edges tie states to eliciting appraisals; common_triggers on nodes; triggered_by for non-appraisal elicitors |
| Opposing emotional forces | ✅ Strong | 40 opposite_of edges; amplifies/inhibits for causal dampening and escalation |
| Recurring emotional patterns | ✅ Partial | transitions_to chains encode prototypical sequences; meta.emotion_regulation anchors regulatory processes |
| Cross-cultural coverage | ✅ Strong for lexicon | 28 cultural concept nodes, culturally_related_to mappings, cultural_scope field |
subcategory groupers), 23 appraisals, 17 moods, 15 behavioral tendencies, 14 physiological states, 14 traits, 9 abstract constructs, 8 drives, 7 affective systems, 7 mental states, 6 needs, 5 feelings, 4 desires, 4 meta states, 7 motivations.
near_synonym_of (219), transitions_to (159), similar_to (121), co_occurs_with (99).
low edges.
in data/ontology-meta.json.
(arousal + pole nodes). A production model should add explicit intensity parameters and evidence on how intensity changes transitions (e.g., mild annoyance vs. rage lead to different behaviors). Extension: add intensity to the state-vector representation; consider intensity-weighted edge modifiers.
object_directed but does not model objects (the person/event the emotion is about). Representing "anger at my boss" or "grief for my mother" requires an object layer. Extension: introduce instance nodes and is_a edges (already reserved in the schema), plus a directed_at relationship to entity nodes.
moves but not durations, onsets, or decay functions (a fear spike vs. a slow grief). Extension: attach duration distributions; encode rise/plateau/decay templates.
and the Panksepp systems are the limits. Extension: add interoceptive dimensions, more autonomic indices, and neural-region associations with appropriate (low) confidence.
positive well-being words; this ontology includes 28 cultural nodes. The mapping of non-Western concepts onto English-proximate states is approximate (see disagreement registry §4.14). Extension: systematic import of Lomas's list and anthropological lexicons (Lutz 1988; Wierzbicka 1999).
development (e.g., stranger anxiety, the "terrible twos" anger, theory-of-mind based pride) are absent. Extension: a developmental layer with age-of-onset estimates.
single process node; reappraisal, suppression, distraction, acceptance, and rumination are not separate nodes (rumination exists as a mental state). Extension: a regulation-strategy subtree with inhibits/amplifies edges to target states.
the bridge, but the ontology is human-centric. Extension: a cross_species field marking which nodes have homologous evidence in other mammals.
confidence model §5.5). A future release should populate strength_value from embedding-based or norm-based similarity with documented provenance.
named states, but some phenomena resist node-based modeling and need structural extensions rather than more nodes:
oscillation) need a process model.
of components (feeling without appraisal; expression without feeling) needs a component-disruption mechanism.
cognition fire together; represented as co-occurrence, not yet as a synchronized pattern.
cognitive biases they induce (memory bias, attention bias). A cognition layer would complete the picture.
human might experience, rather than merely providing a dictionary of emotion
words?*
Mostly yes, with two caveats.
What makes it more than a dictionary: the multi-relational graph (transitions, amplification, inhibition, co-occurrence, blends, appraisal triggers, behavioral expression), the state-vector representation for concurrency and intensity, the distinction between static similarity and temporal dynamics, and the explicit preservation of scientific disagreement.
What keeps it from being a full model of experience: (1) it lacks an object layer, so it cannot say what a state is about — the single biggest extension; (2) it lacks temporal parameters (duration, decay, onset) — needed for realistic simulation; (3) the numeric layer is deliberately conservative, so quantitative inference is limited. Each gap has a concrete extension path (§6.3). With the object layer and temporal parameters added, the same schema would support agent-based simulation of emotional life, not just lookup of emotion words.