Contribution
It is the first sustained internal critique of the mechanism of self-categorization theory, as opposed to its scope or its motivational claims. It separates two things the theory ran together: seeing oneself as a member of a category, and seeing oneself as part of a group. It argues that meta-contrast cannot produce the second and is too costly to produce the first in general. For this record it also does a second job. It fixes the theory as the plain reading of Turner et al. 1987, chapter 3 and Oakes’s chapter 6, and quotes the book’s assumptions and hypotheses with page numbers. That gives the record second-hand access to the unread book (LIT-1723).
Key insight
A category is a set, and a set is internally differentiated, so it is not a unitary thing. Making its members look alike, by accentuation, makes them interchangeable, not one. “Trees” do not become a forest by looking more like one another. A forest is picked out by its own configuration, as a face is picked out by the arrangement of its parts rather than their similarity. Groups, on this view, are perceived when people are seen to be bound by shared mental states, not when they are seen to be alike.
Assumptions
- What counts as SCT (p. 2): the plain reading of Turner et al. (1987), chapter 3, with Oakes (1987), chapter 6, for salience. Turner et al. (1994) is “a supplementary statement” that does not alter the theory. Later work is about the theory, not constitutive of it.
- Fodor’s conceptual atomism and a language of thought (pp. 31–39). The paper adopts these against prototype and inferential-role semantics. Much of critiques 4 and 5 stands or falls with them.
- Modular, context-impenetrable perception (pp. 35, 46), after Firestone & Scholl (2016) and Pylyshyn. If perception were context-penetrable, the argument that structural fit must operate only after perception would lose its force.
- Tractability as a constraint on cognitive theory (p. 25), after van Rooij (2008): a cognitive function must be computable in plausible time.
Key results
The book, as quoted (pp. 2–7). Page numbers are Turner et al. 1987’s.
- SCT is a theory of “the functioning of the social self-concept” and of “how individuals are able to act as a group” (p. 42).
- The self-concept is the “set of cognitive representations of self available to a person”, and “any individual possesses multiple concepts of self” (p. 44).
- Self-categorizations are “cognitive groupings of oneself and some class of stimuli as the same … in contrast to some other class of stimuli” (p. 44).
- There are “at least three levels of abstraction”: human, in-group versus out-group, and personal (p. 45).
- Meta-contrast: “within any given frame of reference … any collection of stimuli is more likely to be categorized as an entity … to the degree that the differences between those stimuli on relevant dimensions of comparison … are perceived as less than the differences between that collection and other stimuli” (pp. 46–47).
- Salience accentuates “intra-class similarities and inter-class differences” (p. 49). There is a “functional antagonism” between levels (p. 49).
- Depersonalization is “‘self-stereotyping’ whereby people come to perceive themselves more as the interchangeable exemplars of a social category” (p. 50). “The depersonalization of self-perception is the basic process underlying group phenomena”, including “social stereotyping, group cohesiveness, ethnocentrism, cooperation and altruism, emotional contagion and empathy, collective action, shared norms and social influence processes” (p. 50). The book itself calls this the point of the whole argument.
- Groups form psychologically to the degree that differences among people are less than differences between them and others present (pp. 51–52).
- Salience is accessibility × fit (p. 54). Accessibility is set by input required, past learning and current motives (p. 55). Fit is structural (similarities and differences “correlate with [that] classification”) and normative (p. 55).
- The group is seen “as one whole entity … as if they were real and unitary psychological beings” (p. 3).
Critique 1: categorization is neither sufficient nor necessary for group entities (pp. 9–15). Not sufficient: accentuation produces perceived interchangeability, not unity. Oakes et al.’s (1994) own analogy between kinds of vegetation and nationalities shows this, since “trees” are a category, not a thing. Not necessary: families, orchestras and governments are groups whose members are by definition dissimilar. Resolution: groups are holistic perceptual entities, analogous to faces, emerging from configural relations among members’ mental states.
Critique 2: “self-categorization” is underspecified (pp. 15–20). “Self” is defined circularly and “stimuli” is overinclusive. Resolution: self-categorization is the categorization of intentional agents, and groups require collective intentionality. The evidence is Skorich et al. (under review), four experiments described in one paragraph (p. 20). Groups defined by mental states (chosen T-shirt colour; seeing a duck or a rabbit) drew more in-group favouritism and were tracked more in who-said-what tasks than groups defined by physical properties (assigned T-shirt colour; digit ratio).
Critique 3: the person level is underexamined (pp. 20–24). Following Mavor, persons are categories of “person aspects” (Alex at home, Alex at university), and person and group processing should be equivalent by default. Two studies are reported. Skorich & Mavor (2013): with memory confounds removed, neither group nor person processing is privileged under load. Skorich et al. (2021): individual identity is extracted from faces as fast as age, race and gender, and shows SCT’s prototypicality effects.
Critique 4: meta-contrast is implausible for entity formation (pp. 24–33).
- Inapplicable to holistic entities, which come from configuration, not similarity.
- Computationally intractable. Taken as a formula for selecting the best in/out split, it is a subset-selection problem with 2ⁿ candidates. At 1,000 operations a second, 5 people take about 32 ms, 25 take about 9.3 hours, and 35 take about 13 months (pp. 25–26). The arithmetic checks: 2²⁵ ≈ 3.4 × 10⁷, 2³⁵ ≈ 3.4 × 10¹⁰. The obvious reply is that readiness and normative fit prune the candidates. The paper answers that pruning presupposes already-formed entities, so meta-contrast may still select among them (salience) but cannot form them (pp. 27–28).
- Prototype semantics. Prototypes do not compose. “Black woman” does not inherit the prototypes of “Black” (a Black man) and “woman” (a white woman), as Crenshaw’s intersectionality requires. Context-specific prototypes cannot be known for a novel compound (“Gabonese-Inuit emu farmers”) before meeting one (pp. 28–31).
- Resolution: Fodor’s atomism. Entities are “fingers of instantiation” triggered by “abductive templates”, bound to a mental token, with content in an “object file” (pp. 31–33).
Critique 5: comparative and correlational relations are overemphasized (pp. 34–40). Context effects on perception are weaker than SCT needs, so structural fit must work after perception, as inference. Comparison is then one propositional relation among indefinitely many. Resolution: a language of thought.
SCT2.0, “Self-Conceptualization Theory” (pp. 40–45).
- Kept from SCT: groups are psychologically real and part of the self; perception is dynamic; the same explananda.
- Changed: self-categorization is of intentional agents only; category membership is distinguished from group membership, and category effects (out-group homogeneity) from group effects (crowd behaviour).
- Meta-contrast has “little role” in entity formation or salience. Perception is “not inherently and universally comparative”: an in-group need not appear against an out-group.
- Without comparison there is no need for strict levels, and so no “coherent notion of functional antagonism” (p. 44).
- SCT1.0 becomes a sub-theory: the special case of equivalence classes under comparison.
The three sub-processes (pp. 46–58).
- Salience is staged: an impenetrable perceptual triggering, then inference using goals and background knowledge.
- Depersonalization hides two shifts: away from the person (keep the word) and towards a group entity (“collectivization”).
- Prototypicality is a match between contextually activated object-file content and a member. It need not be comparative (“students during spring break”). It applies to categories, not to group entities, whose members are parts, not exemplars. A part can only be diagnostic (Ronaldo for the Portuguese team) or outsized (Eisenhower).
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Categorization into equivalence classes, even with accentuation, yields perceived interchangeability, not a perceived group entity | moderate: a conceptual argument (sets versus things); no data | pp. 9–12 |
| C2 | Perceiving a group as an entity does not require its members to be similar (families, orchestras) | moderate: argued from cases | pp. 12–14 |
| C3 | Groups are perceived from cues to collective intentionality, and mental-state groups draw more favouritism and tracking than physical-property groups | weak: four experiments under review, summarized in one paragraph; not checked | p. 20 |
| C4 | Meta-contrast as a procedure for selecting the best in/out split over n people is intractable (2ⁿ subsets) | strong as arithmetic, given the exhaustive-search reading; whether SCT ever meant that reading is the open point | pp. 25–28 |
| C5 | Prototypes do not compose, so meta-contrast cannot fix category content | weak to moderate: rests on Fodor’s arguments, which are contested; illustrated, not shown | pp. 28–31 |
| C6 | Context effects on perception are too weak for structural fit to act perceptually | weak: cites a contested literature (Firestone & Scholl 2016) | p. 35 |
| C7 | Person and group processing are equivalent by default; neither is privileged under load, and identity is read from faces as fast as social categories | moderate as reported: two published studies by the author’s group, summarized | pp. 22–24 |
Concepts
- SCT1.0 / SCT2.0 / S-ConcepT: the 1987 theory; the proposed revision; its other name, “self-conceptualization theory”.
- intentional agent: an object perceived to generate its own actions, and so to have a point of view, mental states and a sense of self (pp. 17–18).
- collective intentionality: intentionality spread across individual agents (Tomasello & Rakoczy 2003); held necessary for perceiving a group (p. 19).
- category versus entity: a set of separately instantiated things, versus a bounded, holistic, internally undifferentiated thing (p. 10).
- person aspects: the instances of one person across situations, after Mavor (p. 22).
- FINST, abductive template, object file: an attentional index triggered when a configuration of low-level features matches a stored holistic pattern, with an associated store of content (pp. 31–33).
- collectivization: the shift towards seeing oneself as part of a group entity, distinct from depersonalization, the shift away from the person (pp. 49–53).
Connections
- Turner et al. 1994 (NOTE-1284). Skorich treats it as supplementary. Two of its points are the ones he attacks: metacontrast setting content and prototype (critiques 4–5), and the self as generated in context. He keeps the latter’s dynamism but moves the context effects from perception to inference.
- Hornsey 2008 (NOTE-1253). Hornsey reports a softening of functional antagonism; Skorich’s SCT2.0 removes it.
- León 2023 (NOTE-1309). Cited for the point that close relationships are groups without interchangeable members (p. 14). The two papers converge on that and diverge on the remedy: perceptual collective intentionality here, socially extended personhood there.
- Collective intentionality. THEORY-126 already distinguishes I-mode from We-mode collective emotion by collectively intentional appraisal. Skorich makes collective intentionality the condition of perceiving a group at all. These are different claims that use the same notion.
- Fodor and Pylyshyn. The paper’s positive theory imports conceptual atomism and the language of thought wholesale. The record holds no reading of them.
- Self-models. “Persons as categories of person aspects” sits near THEORY-064 (self-model) and THEORY-691 (self as abstractum) without engaging either.
Bearing on the record
- THEORY-1114. This is its strongest rival. Skorich accepts that self-categories vary with context, but denies that meta-contrast is the mechanism of entity formation and gives it “little role” in salience. The THEORY’s promote_when already asks for a frame-only manipulation. This reading adds a sharper test: whether comparison effects survive when the comparison group is absent, which SCT2.0 says is possible and SCT1.0 does not. Noted on that THEORY.
- Source of THEORY-1112 (Proposed): perceiving a group as an entity requires cues to shared mental states, not similarity.
- CASE-104 and QUESTION-119. Skorich’s line between a category and a group entity cuts the fanbase case in two. “The fans” can be a category (people who support the club) without being a group entity, which would need perceived shared mental states (a chant, a boycott, a supporters’ trust). His notion of an individual “diagnostic” or “outsized” in a group (Eisenhower) is an alternative to the prototypicality reading of a vocal minority. These are inferences, not claims of the paper.
- Stand-in for the book. The quotations above are the record’s second-hand reading of Turner et al. 1987, chapter 3. They are not a reading of the book, which stays Unavailable.
- No ML instruction. The paper cites AI applications of SCT (Bingley et al. 2023) in passing only.
Limitations
- Unreviewed. A preprint. Its key evidence for the positive proposal (Skorich et al., under review) is not available and is summarized in one paragraph.
- The intractability argument assumes exhaustive search. It shows that meta-contrast cannot be computed by brute force over all subsets. It does not show that no tractable approximation, such as clustering on a similarity space, could do what the theory needs. The paper answers only the pruning reply.
- The positive theory is heavily borrowed. Conceptual atomism, context-impenetrable perception and the language of thought are each contested, and the paper does not defend them beyond citation.
- The supplementary predictions were not read, so how SCT2.0 would be tested is not recorded here.
Open questions
- Do comparison-dependent effects on self-stereotyping and prototypicality appear when no comparison group is present, or does salience require one? This is the experiment that separates SCT1.0 from SCT2.0.
- Do mental-state groups reliably beat physical-property groups in favouritism and tracking? The under-review studies, once published and read, would settle C3 for the record.