THEORY-057: Emotion categories are populations of variable instances grouped by the perceiver's concepts, not natural kinds with a shared neural or appraisal mechanism

Source

Barrett (2017), LIT-542, read in full in NOTE-434: “The theory of constructed emotion”, Table 1, the section on degeneracy, footnote 15 and the list of claims the paper does not make.

What was actually shown

The argument has three steps. None of them is a new experiment.

1. Degeneracy. Instances of one category, such as anger, are produced by dissimilar sets of neurons “in the same context” (the wording as corrected in nsx060). This is many-to-one mapping of structure to function, which Barrett says is “ubiquitous in biology”.
2. No consistent signature (Table 1; C4–C5 in NOTE-434). Cited meta-analyses find no region or intrinsic network that is consistent and specific for a category. Multivariate patterns that classify a category need not occur in any of its instances. Stimulation finds no emotion-specific receptive fields. Amygdala lesions have variable effects. Classifiers do not replicate across studies, and a classifier pattern is “a statistical summary of a highly variable set of instances”. Taking it for the emotion’s brain state is the “essentialist error”.
3. Population thinking. After Darwin as read by Mayr, a category is a population of variable instances grouped “by ‘a goal’ rather than by similar features or a single, shared underlying cause”. Its summary “need not exist in nature”. An emotion is constructed when sensation is categorized with an emotion concept (C2).

The conclusion is that no emotion category has an essence, neither a Platonic fingerprint in face, body or brain nor a Lockean mechanism, which for appraisal theories is “a set of evaluative mechanisms”. Descriptive appraisals survive as products of categorization.

What the philosophy holdings add

• Degeneracy alone does not carry the conclusion. Pernu (LIT-147, NOTE-091) shows the parallel mistake in arguments for mental causation. Many single-neuron patterns realise one cortical variable, and each intention corresponds to one value of that variable. So variability at the fine grain is micro over macro within the physical, and leaves identity at the coarser grain intact (C1, C7). Applied here, step 1 is consistent with a classical view that puts the emotion’s mechanism at a coarser grain than neurons. The account therefore rests on step 2, the claim that no grain shows a consistent signature, and that step is cited from other papers that the record has not read. The application of Pernu is the record’s own, and it is the reason promote_when asks about a variable at any grain.
• Categories without definitions are coherent. Minsky’s frame memo (LIT-410, NOTE-337, C5) argues that family resemblance “needs no definition” and can come from local similarity links. That is an existence argument, not evidence about emotions. It removes one objection: that a category with no shared feature could not be learned or used.
• “Social reality” is narrower than it sounds. Footnote 15 counts emotion categories among “the categories of social reality, such as flowers and weeds”, whose functions are imposed on physically disparate instances by collective agreement, like currency. Searle (LIT-482, NOTE-388) makes such functions observer-relative (C1, C3). But his mark of institutional reality is a status function that carries deontic powers (C4), and Barrett names none for anger or fear. So her comparison is to observer-relative function in general, as with weeds, and not to institutions such as money. Epstein’s split (LIT-487, NOTE-393) states what the theory needs. A particular instance is grounded in brain and body states. What makes it fall under “anger” is anchored in a community’s concept. Neither paper discusses emotion, so this mapping is the record’s own.
• The rival has the same shape. Frijda (LIT-540, NOTE-433) also lets instances vary. “When angry one can kick, shout insults, turn one’s back”, and he says that a motive is inferred “from observing a variety of equifunctional actions”. Both accounts group variable instances by function. They disagree about whose function does the grouping: for Barrett it is the perceiver’s goal-based concept, and for Frijda it is the agent’s own readiness to change its relation to the object. That is the precise point of the rivalry with THEORY-055. It is held as rivals, not corrects, because neither paper cites or tests the other (ADR-022).

What this does not say

• It does not say that emotions are unreal. “I am not saying that emotions are illusions”, Barrett writes, but that “emotion categories don’t have distinct, dedicated neural essences”. Instances are real brain events, and the categories are real in the way weeds are.
• It does not deny affect. Affect is felt allostatic state and is present whether or not any emotion concept is applied. That layer is THEORY-045, which this account sits on top of but does not need. Steps 1–3 do not use it.
• It does not establish that appraisal theories are false. Barrett’s case is strongest against neural fingerprints. Her Figure 1 counts causal appraisal as a Lockean essence, but she does not engage any appraisal theory’s predictions (Limitations in NOTE-434). That is why the relation to the appraisal account is rivals.
• Its denial of fear to the fly is perceiver-relative. “Does the fly feel fear?” is answered “no” because the fly has no concept of fear. The fly “may feel affect”.
• It is not Dennett’s real-patterns question. Whether a category with no shared mechanism is still a real pattern in behaviour is a further question. The record’s account of that criterion (THEORY-036) finds that it admits almost every non-random pattern, so it would admit these categories without deciding anything between the rivals.

Note of 2026-10-29: what holds it back

The paper reports no experiment. Its conclusion rests on Table 1’s claim that no grain shows a consistent signature, which it cites from meta-analyses the record has not read; degeneracy alone does not carry it. The record holds neither result promote_when names: no manipulation of concept availability with affect and situation held fixed, and no variable shown common to one category’s instances and absent from others’ at a coarser grain. The reply that could test it, Adolphs’s companion piece, is not filed.