NOTE-1067: Barrett et al. — The theory of constructed emotion: more than a feeling

Contribution

A reply, not a new statement of the theory. Against van Heijst et al.’s proposal that basic emotion theory (BET) explains emotions as “bioregulatory states of the body” and the theory of constructed emotion (TCE) explains feelings, the sixteen authors argue that the split misdescribes their theory, ignores evidence against BET, and rests on an outdated evolutionary biology. Beyond the 2017 statement (LIT-542) it states four things plainly: feeling and bodily regulation are one mechanism, not two; the theory is holistic, not reductionist; its evolutionary frame is the extended evolutionary synthesis; and the two theories are incommensurable and cannot be integrated. It reports no data.

Key insight

Separating an emotion into an objective bodily state and a subjective feeling of that state is not neutral ground on which both theories can meet. It is “the bifurcation of nature” (Whitehead), and it is already a commitment of BET. On the TCE an instance of emotion is “a complex causal ensemble (i.e., a whole-brain event) whose simultaneous features include bodily regulation and the mental features of experience” (Error 1, “Holism, not reductionism”). So there is no part of the emotion that is the body’s and another that is the feeling’s.

Assumptions

• Allostasis is the brain’s main function. The brain anticipates the body’s needs and prepares to meet them. “Allostasis is predictive and does not involve set points” (Error 1, “A brain regulates and coordinates the systems of its body”). Homeostatic subsystems are coordinated by it.
• Efference copies. Axons from “primary allostatic control neurons” (anterior cingulate is the example) send motor control signals to the spinal cord and, at once, prediction signals that build features of experience. The support is anatomical (Barbas 2015), cited (Error 1, “Feeling is an emergent feature of the brain’s regulation of the body”).
• Population thinking. Variation among instances of a category is structured by context and is real; a prototype is a summary (Mayr). BET is cast as typological thinking (Error 2).
• Relational realism. Mental features are “not properties of the intrinsic brain signals alone or of the incoming sensory signals alone but of their relations” (Error 1). “Relative” means only that instances of emotion “are conditioned on some perspective”; it does not mean subjective or arbitrary (Error 2, “Relational meaning”), after Barrett & Theriault (2025).
• The extended evolutionary synthesis against the modern synthesis: genes are “necessary but not sufficient”, context is “fully causal rather than merely modulatory”, and cultural inheritance and niche construction are evolutionary processes (Error 3).

Key results

There are no new data. By section:

• Error 1: more than a theory of feeling.
  • The TCE contains hypotheses about emotions as “bioregulatory responses”, but “Emotions are not special in this regard”: every moment of life involves bioregulation. “Feeling cannot be meaningfully separated in any biologically justified manner from the brain’s regulation and coordination of the various systems of the body.”
  • Allostasis “does not cause separate instances of emotion but manifests them”, as it manifests perception, cognition and decision.
  • The efference-copy hypothesis: “efferent copies of the same allostatic signals that control and regulate the body simultaneously create feeling and other features of lived experience”, together with afferent and reafferent signals.
  • Mental features are compressions: “lower dimensional summaries of higher dimensional sensory and motor signals once their temporal and spatial redundancies have been removed”.
  • Table 1 sets the TCE, van Heijst et al.’s description of it, and BET against Tinbergen’s four questions. Its causal entry: categorizing creates regularities rather than detecting them, and the meaning it makes “manifests as features of allostasis, action, and lived experience”.
• Bifurcation and the blind spot. The split implies that experience has “no significant causal role in the brain’s bioregulation of the body”. It has “manufactured” a correspondence problem, citing Gross (1998), Mauss et al. (2005) and Ventura-Bort et al. (2022): physical changes during instances of an emotion category rarely covary with its experience. “Heart rate can increase, decrease, or remain the same (relative to a baseline measurement) when people report feeling angry or afraid.” The section also blames the blind spot for the “widespread failure of pharmaceutical companies to develop drugs that cure depression and anxiety” (citing LeDoux 2015) and for reading emotion from faces in hiring and in court.
• Error 2: evidence against typological thinking. Table 2 lists studies of variation within and similarity across emotion categories in peripheral physiology, brain signals, expressions, affective features, culture and words. “One cannot integrate a theory that requires unchanging prototypes with a theory that denies their existence.” Even a single brain biomarker is “a statistical fiction” (after Clark-Polner et al. 2017). The inference is stated as asymmetric: population evidence “disconfirms” emotion types, but commonality among instances would not show a gene or a dedicated circuit.
• Ad hoc categories. “A category is something you do, not something you have.” A category event begins as a sample of brain-wide prediction signals with a similar control plan, “something like an action tendency”, and ends as one situated instance selected by incoming signals. Signals such as facial movements and heart rate “have no inherent, biologically prepared emotional meaning”; meaning is relational. “A perspective-independent emotion does not exist in any meaningful sense.”
• Error 3: the extended evolutionary synthesis. BET is placed in the modern synthesis, where “FEAR’s expression or triggers can be modified or tweaked after the fact by lived experience, culture, or other phenomena such as regulation strategies”. The TCE is placed in evo-devo: caregivers’ allostatic support builds the niche in which infants learn emotion categories, words are “tools of niche construction”, and humans are said to build more abstract categories than chimpanzees for reasons of niche and brain scaling. “All animals categorize”, and “Even single-cell organisms can be said to categorize”. Animal brains categorize incoming signals “and create lived experience (whatever that means for the animals in question)”.
• Incommensurability. “Our actual theory cannot be integrated with BET because the two are incommensurate” (Kuhn, Feyerabend). Table 3 lists eight contrasts. On categorization: “Categorization involves constructing regularities, not detecting them.” On causation: contextual factors are “fully causal and necessary”.
• The James–Lange example. James’s and Lange’s theories were incommensurable, population against typological, and Dewey joined them by the same bifurcation. The authors call this “one of the great mistakes in modern psychology”.
• Conclusion. “Emotions are not part of some perceiver-independent reality.” In a survey of emotion scientists, “80% of the respondents endorsed typological beliefs (Ekman, 2016)”.

Claims

id	claim	strength	support
C1	The TCE is a theory of bioregulation and action as well as of feeling; van Heijst et al.’s reading misdescribes it	strong as a statement of the authors’ own theory	the architects’ testimony, with citations to their earlier work (Error 1, Table 1)
C2	Feeling cannot be separated from the brain’s regulation of the body: efference copies of the same allostatic signals make both	weak: a hypothesis, untested here	anatomical argument cited (Barbas 2015); Error 1
C3	Splitting an emotion into a bodily state and a feeling is a premise of BET (the bifurcation of nature), and it produces the correspondence problem	interpretive	argument; the correspondence problem cited from Gross 1998, Mauss et al. 2005 and others (Error 1)
C4	Within-category variation and across-category similarity are large and structured by context, which disconfirms emotion types	moderate as cited; nothing re-analysed	Table 2’s list, mostly from the authors’ groups (Error 2)
C5	Emotion categories are ad hoc and constructed in the moment; signals have no inherent emotional meaning	weak to moderate: a hypothesis with cited support	Error 2, “Ad hoc categories” and “Relational meaning”
C6	The TCE’s evolutionary basis is the extended evolutionary synthesis, and emotion categories are culturally inherited	weak: argued, with citations to evo-devo	Error 3
C7	The TCE and BET are incommensurable and cannot be integrated	interpretive	Table 3; Figure 1; Kuhn and Feyerabend
C8	The James–Lange theory is Dewey’s integration of two incommensurable theories	weak: a historical claim, briefly argued	”A cautionary example”

Concepts

• Bifurcation of nature: separating perceiver-independent physical states from perceiver-dependent feelings, so that feelings exist “only in a perceiver’s mind” and are taken as epiphenomenal (Whitehead 1920).
• Allostasis: the brain’s predictive regulation of the body for efficient energy use, without set points, as distinct from homeostasis.
• Efference copy (here): the same allostatic control signal sent both to the body and as a prediction that helps construct features of experience.
• Ad hoc category: a momentary, situated grouping of instances that serve one function in one context, built in the brain as a dynamic event.
• Typological and population thinking: Mayr’s terms. The first takes a category’s prototype as real and variation as error; the second takes the variation as real and the prototype as a summary (Figure 1).
• Relational realism: instances of emotion are real and “conditioned on some perspective”; BET is cast as “traditional realism” (Figure 1).
• Abstract feature: a lower-dimensional feature generalizing over different concrete sensory and motor patterns, such as “threat” or “remove an obstacle” (footnote 8).

Connections

It answers van Heijst, Kret & Ploeger (2025) in the same issue, and places them in a line with Adolphs (2017), Damasio (2004) and de Waal (2019), who also separate emotion from feeling. It rests on the authors’ own earlier work, chiefly Barrett 2017b (LIT-542), Barrett & Finlay (2018), Hutchinson & Barrett (2019), Shaffer et al. (2022) and Barrett & Theriault (2025, the philosophy of science behind “relational”). For evolution it draws on Bateson & Laland (2013), Jablonka & Lamb, Laland and Ball (2024). It cites Gross (1998), the Journal of Personality and Social Psychology study of antecedent- and response-focused regulation, not the 1998 review the record holds (LIT-554). It does not cite Seth.

Bearing on the record

• LIT-542, NOTE-434 (Barrett 2017). The same theory. What this reply adds is emphasis and one explicit denial: the 2017 paper already put allostasis first; this one says outright that feeling is not a second phenomenon representing the first (C2, C3). It adds the evolutionary frame (C6) and incommensurability (C7). It does not repeat NOTE-434’s C8, the hypothesis about reappraisal.
• THEORY-057 (emotion categories as populations). Error 2 and Table 2 restate its claim with a longer evidence list, all cited and not re-analysed, much of it from the authors’ own groups. A further source for it, not a test. The asymmetry it states (variation disconfirms types, commonality would not confirm them) is worth adding there.
• THEORY-045 (felt affect). C2 takes the prediction side of THEORY-045’s open question again, now as the efference-copy hypothesis. It offers no case where prediction and map come apart, so it does not meet that THEORY’s promotion test.
• The dispute between THEORY-056 and THEORY-680. The paper does not discuss emotion regulation, reappraisal or therapy. It bears on the dispute in three ways, none of them decisive.
  • Separate regulation is placed on BET’s side. Regulation appears twice, both times as something BET uses: variation is put down to “regulatory efforts” (Error 2), and FEAR is “tweaked after the fact” by “regulation strategies” (Error 3; also the Figure 1 caption). A regulation process that acts on an already-formed emotion is part of the typological picture the authors reject. That agrees with THEORY-056’s clause that there is no distinct regulating system, by implication only. It is not an argument about regulation.
  • THEORY-680’s evidence is read as an artifact. The study cited for the correspondence problem, Gross (1998, JPSP), is the experiment THEORY-680’s cited contrast rests on (suppression lowered expression and raised sympathetic activation without reliably lowering experience). THEORY-680 reads that dissociation as strategies acting on different response channels of one emotion. This paper reads dissociations of that kind as what the bifurcation manufactures. The citation is in passing, one item in a list. The record has not read the 1998 experiment either.
  • The counterfactual gets harder. THEORY-680 needs an unregulated course to compare against. If allostasis “manifests” emotions rather than causing them, and every moment is bioregulation, there is no emotion that is not already the body being regulated. That is my inference, not the paper’s. It sharpens the problem THEORY-680 already names in “What this does not say”; it does not settle it.
  • Neither promotion test is met. No dissociation, mediation or stage comparison is reported.
• LIT-1359, NOTE-1061 (Bo et al.). The exchange set the two against each other: Bo et al. found amygdala, PAG and hypothalamus responses to aversive pictures unchanged by reappraisal while reported negative affect fell. Reading that as an unchanged component of the emotion, apart from the feeling that changed, is the bifurcation this paper rejects. On the TCE the subcortical response is one feature of a whole-brain event, not the emotion’s objective part. So the reply does not refute that reading of Bo et al.; it shows that it presupposes BET’s split. The comparison is mine.
• CLAIM-412 (experience constituted by interpretive organization). The exchange called the TCE the nearest formulation of the constitutive reading for affect. The reading supports that in part. Meaning is relational, mental features are relations among signals, and allostasis “manifests” instances of emotion rather than causing them. That is constitution language. But the paper also says that categorizing gives meaning to signals “to create lived experience” (Table 1), production language, and it does not choose between the two. Its constraint is the one the exchange named: the meaning is not separate from bodily regulation. It does not mention cognitive behavioural therapy; that was the exchange’s example, not the paper’s.
• CLAIM-411 (broad interpretation admits processing not shown to be experiential). The paper is a case of the objection. Categorization, its word for interpretation, reaches “single-cell organisms”, and animal brains categorize “and create lived experience (whatever that means for the animals in question)”. The theory extends categorization without limit and leaves experience unanalysed. It gives no condition under which categorization is experiential, so it offers nothing towards QUESTION-254.
• TERM-073 (interpretation). The paper’s categorization fits the term’s three parts: a distinction (a category that “creates” regularities), a context (situated, ad hoc) and a consequence (it guides allostasis and action). It is not defined without experience, though: the categories are said to make experience.
• TERM-072 (perspective, three senses). “Conditioned on some perspective” is used without saying which sense. In the Conclusion it is the perceiver’s or the scientist’s inherited categories, which is closer to the organizational sense than to the phenomenal one.
• THEORY-767, LIT-1263 (Seth). Not touched. “Interoception” occurs only in Table 1, in van Heijst et al.’s words.
• THEORY warranted? None filed. Its substantive claims are THEORY-057’s and THEORY-045’s, already held from LIT-542. Incommensurability (C7) is a claim about two theories, argued and not shown. Suggested: add this reading as a source of THEORY-057.
• No instruction for machine-learning practice; nothing for the anthology.

Limitations

• A reply with no data. All evidence is cited, and Table 2’s lists draw heavily on the authors’ own groups.
• BET is characterized by its opponents, largely through van Heijst et al., which was not read here. Whether BET as its own proponents state it requires “immutable” prototypes is not argued from their texts.
• The efference-copy hypothesis (C2) is stated, not tested, and nothing in the paper shows that feeling and bodily regulation cannot be dissociated. Incommensurability is asserted from a contrast table.
• Several claims go well past the evidence given: that the blind spot explains failed anxiety drug development, and the history of Dewey and the James–Lange theory.
• “Lived experience” and “feeling” are not defined, and the paper is explicit that what experience is for other animals is left open.

Open questions

• What result would show that feeling and bodily regulation are, or are not, one mechanism? A manipulation that changes the felt features of an instance while its allostatic signals are held fixed, or the reverse, would bear on C2. None is proposed.
• If regulation strategies belong to the typological picture, what is reappraisal on the TCE? The 2017 paper’s answer (NOTE-434, C8) is that it works through the same predictions. This reply does not say.
• How is a population of instances to be tested against a family of prototypes, so that the two theories are not merely redescribed? Incommensurability makes a shared test hard by its own account.