Bodily sensation maps
A bodily sensation map (BSM) is the pixelwise topography of where in the body a person reports feeling activity increase and decrease during an emotion, measured with embody. The theoretical claim attached to them is stronger than the measure: that these maps are discrete, emotion-specific “fingerprints,” and that they determine the qualia of subjective emotional feeling (Nummenmaa et al. 2014, 2018).
That claim is a modern, high-resolution descendant of James — the felt body as the content of emotion — but relocated from physiology to phenomenology. See william-james, friedman-2010-jamesian-perspective.
The empirical profile
Nummenmaa et al. (2014) is the original. Five experiments, 773 subjects, and the key structural facts the later literature reproduces:
- Thirteen emotions plus neutral separate under LDA from word cues (38% vs 14% chance, complete classification for the basic six).
- The maps survive changes of cue. Vignettes that never name the emotion, silent films, and photographs of strangers’ faces converge with word-cued maps at rs = 0.71–0.82 — and classify better than words do. This is the paradigm’s foundation and the reason its self-report data are taken seriously at all. See embody.
- Concordant across Finnish, Swedish and Taiwanese Hokkien speakers (rs = 0.70–0.75), on n = 36 for the Taiwanese sample.
- Shared structure dominates: head activation in every emotion, upper-chest activation in most, and different emotions’ maps correlating at rs ≈ 0.52 on average. The emotion-specific part is a residual on a large common component — worth holding in mind against the word “fingerprint,” and against every concordance figure the paradigm reports, all of which sit on that floor.
Volynets et al. (2020) is the largest test. In 3954 people across 101 countries:
- Each of 13 emotions plus neutral yielded a distinguishable map. Anger and happiness fill the upper body and head; sadness and depression are dominated by whole-body deactivation with limb “emptying”; love and pride concentrate in head, chest and (for love) genitals; shame and jealousy pair facial/chest activation with limb deactivation.
- The maps cluster by emotion, not by country (t-SNE), with inter-nation similarity rs > 0.82.
- Emotion category dominates the variance; culture, language, sex, education and BMI do not.
- Intensity dampens with age across all emotions (age-related-interoceptive-decline).
- Discreteness is not confined to Ekman’s six: anxiety, love, depression, contempt, pride, shame and jealousy have fingerprints too, which the authors read as universality extending past the canonical basic set.
The clinical profile — where the construct earns its keep
Lyons et al. (2021) add the first psychiatric data here, and a double dissociation that is better evidence for BSMs being about something real than any amount of cross-cultural concordance:
| are emotions distinguishable? | what changes | |
|---|---|---|
| schizophrenia (Torregrossa et al. 2018) | no — maps indistinguishable between emotions | the structure collapses |
| depression, unmedicated | yes — as accurate as controls | localised reductions (stomach, chest in sadness; legs in fear) |
| depression, medicated | yes — highest accuracy of the three groups | activation inverts to deactivation (63% of painted area vs 24% in controls) |
Two disorders, two different failures, one instrument. And a classifier trained on healthy controls transferred to both depressed groups above chance — depression does not relocate emotion in the body. Whatever is wrong with emotion in depression, sadness has not moved.
This matters for the word-cue objection below. If BSMs were simply shared conceptual knowledge about where emotions belong, they should not fracture along diagnostic lines — everyone in these samples knows the same folk theory of where sadness lives. See embody.
A problem for the constitutive claim
The strongest thing Nummenmaa’s group says about BSMs is not that they are discrete but that they determine the qualia of felt emotion. Two findings now put pressure on that, and both come from inside the paradigm.
Deactivation is induction-dependent. In Volynets et al., sadness and depression are carried primarily by deactivation — limb “emptying” is most of the signature. Under picture induction, “almost no deactivation reached significance” in healthy controls (lyons-2021-body-maps-depression), and the authors note the discrepancy without explaining it. If half the fingerprint appears or vanishes depending on whether you cue with a movie, a word, or a photograph, then the map is partly a property of the elicitation rather than of the emotion. A fingerprint that changes with the ink is a weak candidate for constituting qualia.
Sharp maps coexist with blunted feeling. Medicated patients reporting the felt bodily character of emotion as deactivation nonetheless classified at the highest rate in the study. The map stayed sharp and turned blue. If BSMs determined qualia, a large change in felt quality should have shown up as a change in map structure, not just its sign. See antidepressant-emotional-blunting.
Fear’s map is not legible to people who have felt fear. In Nummenmaa et al. (2014)‘s fifth experiment — rarely cited, including by the programme — naive observers shown the averaged basic-emotion maps named them at 46% against 14% chance, but fear at 8%, below chance and nonsignificant, and happiness at only 22%. (Neutral, at 99%, carries much of that average, and neutral is the one map that is mostly blank.) A linear classifier on 30 principal components tells fear apart from the other maps; a human being cannot. If the map of fear were what fear feels like, that gap should not be there. The defence — reading a heatmap is not having an experience — works, and costs something: it puts distance between the map and the feeling at exactly the joint where the constitutive claim needs them fused.
None is fatal — a defender can say the maps track qualia faithfully, that the qualia really did change sign, and that recognition is a different task from experience — but all three are places the constitutive claim (2) on lauri-nummenmaa has to do work it has not yet done. Note that the founding paper is more modest than the programme became: it explicitly does “not argue that the BSMs… would be the only components underlying emotional experience,” only that they may be “the most reliable and systematic consciously accessible bodily states during emotional processing.”
Why the construct is contested
The interesting question is not whether BSMs are reproducible — they plainly are — but what they are maps of. Three readings are live in this wiki, and no source here adjudicates between them:
| reading | BSMs are… | fits |
|---|---|---|
| Nummenmaa’s | the felt output of emotion-specific bodily states; the substrate of qualia | basic-emotions, interoception |
| Constructionist | reports generated from shared concepts of emotion, cued by shared words | core-affect, lisa-feldman-barrett |
| Deflationary | a real and reliable phenomenological measure that simply has no physiological commitment either way | the authors’ own limitation section |
The third is the one the source itself endorses when pressed: “the BMs are thus subjective feeling maps.” That concession is easy to lose, because the discussion sections of these papers argue for a biological basis to emotion, while their limitations sections disclaim any physiological reading of the data. Both are in volynets-2020-cultural-universality.
The second reading now has a measurement
The constructionist row above used to be a possibility argued from first principles. Ferré et al. (2024) put a number on it from a different literature entirely (Spanish psycholinguistic norms, 1,286 words, no bodies involved): interoceptive content is the largest single determinant of whether a word counts as an emotion word at all — R² increment .417, ahead of feeling, and ahead of appraisal, action tendency and body expression, which contribute nothing or contribute negatively.
If the concept EMOTION is partly constituted by “felt inside the body,” then showing someone the word sadness and asking where they feel it hands them an interoceptive expectation before they introspect. The shared conceptual knowledge the constructionist reading needs is not hypothetical; it has been measured and it is large.
The honest complication, recorded because it is genuine and not a hedge: this runs in both directions. Concepts come from somewhere, and the most natural explanation for an interoceptive folk concept of emotion is that emotions really are felt in the body and the lexicon recorded it — on which reading Ferré et al. is a semantic fossil of exactly what Nummenmaa measures directly, and it supports row one. Ferré et al. themselves lean that way. Nothing in a rating study can adjudicate it. What changes is the burden: the word-cue objection is no longer a possible confound to be noted and passed over. See emotion-prototypicality, embody.
The counterweight remains the clinical dissociation above. Spanish semantic norms do not know who is medicated, and shared concepts should not fracture by diagnosis.
Not a cortical body map — a distinction to hold
A BSM is a felt topography, reported by painting on a silhouette. It is not the cortical somatotopy of the classic Penfield homunculus — the orderly mapping of the body surface onto motor/sensory cortex — even though both get called “body maps.” Zeharia et al. (2019), mapping a whole-body somatotopic gradient in the precuneus, are the wiki’s clearest example of the cortical sense. The two are not known to be related: nothing in this wiki shows the emBODY topographies being read off any somatotopic cortical map, and the founding papers disclaim a physiological reading of the felt maps entirely (see the deflationary row below). A somatotopic homunculus tells you which cortex represents the toe; it does not tell you why sadness is felt as heavy legs. See cortical-somatotopy for the three-way disambiguation of “body map” (cortical somatotopy vs. felt sensation map vs. interoceptive/visceral map).
Relation to autonomic specificity — a distinction to hold
BSM specificity and autonomic-specificity are separate claims and can come apart in either direction, which is worth stating explicitly because the vocabulary invites conflation (“fingerprints” is used in both literatures — compare Siegel et al. 2018, “Emotion fingerprints or emotion populations?”).
- Distinct felt maps without distinct autonomic patterns: what you would expect if felt maps are conceptually or attentionally organised. This is consistent with everything Volynets et al. report.
- Distinct autonomic patterns without distinct felt maps: what you would expect given imperfect interoceptive accuracy (Critchley et al. 2004).
So evidence for one is not evidence for the other. See autonomic-specificity-of-emotion, where BSM findings are recorded as re-describing rather than deciding the question.
Developmental arc
Two data points, one lifespan trajectory, both from this group: maps become more discrete across childhood into early adulthood (Hietanen et al. 2015), then weaken in intensity from adulthood to old age (Volynets et al. 2020). Sharpening then fading. Both legs are cross-sectional.
Note the shape of both the ageing and the depression results: what moves is intensity, not topography. Emotions are still felt in the same places, less strongly — in ageing, in depression, and on antidepressants alike. No source in this wiki reports a population whose maps are the same strength but a different shape, and only schizophrenia produces a loss of shape. Whether that is a fact about emotion or a ceiling of the instrument is untested. It also sets up is-more-interoceptive-awareness-better: if the only thing that varies is how loudly the body speaks, the interesting question is what the right volume is, and the wiki’s sources disagree.