Contribution
It states what a semantics for cognitive science should explain, as eight desiderata, and argues that only two-factor conceptual role semantics (CRS) is even aimed at all of them. It also gives the first explicit argument that one-factor, long-arm conceptual role semantics (Harman) is equivalent to the two-factor theory, and it gives the systems reply to the Chinese room a theory of what the system’s symbols would mean.
Key insight
Split meaning into what is in the head and what relates the head to the world, and many disputes dissolve. Narrow meaning, conceptual role, is what twins share and what explains their behaviour; reference is what differs between them. Objections that conceptual roles are about symbols rather than the world (Fodor’s Tom Swift, Dretske’s “cream”) hit only a one-factor theory: in two-factor CRS the referential factor does that job (pp. 637–639).
Assumptions
- Representationalism, minimal (pp. 641–642). Thoughts are structured: representational states “constitute a combinatorial system”. This is “empirical and contingent”, motivated by cognitive science’s success and by productivity (entertainable n × m = q sentences number on the order of forty-six digits). Rules need not be represented, and some reasoning may be primitive.
- Token materialism (p. 617) and a focus on the brain.
- A language module that sets up a “standard association” between English and mentalese expressions, characterisable non-intentionally (pp. 632–634); conceptual roles of English are those of their mentalese associates.
- Idiolect narrow meaning, not public language meaning (p. 633).
- No analytic/synthetic distinction; meaning similarity should be a “multidimensional gradient”, as decision theory replaced belief with degrees of belief (pp. 624, 629; n. 16).
Key results
- The eight desiderata (pp. 616–621). (1) meaning and reference/truth; (2) what makes expressions meaningful; (3) relativity to representational system; (4) compositionality; (5) meaning and mind/brain; (6) autonomous versus inherited meaning; (7) knowing, learning and using; (8) why narrow and wide meaning bear differently on reference and explanation.
- Narrow and wide individuation (pp. 619–625). Narrow groups tokens that attribute the same properties by the same descriptions; wide groups those that attribute them to the same individuals. Narrow meaning is better for explaining decisions; wide meaning predicts where world–behaviour laws hold regardless of description. The tour-bus variance argument supports an individualistic scheme in psychology. Narrow meaning may be an “aspect or determinant” of meaning rather than a kind of meaning; the “fallacy of subtraction” charge is “merely verbal” (p. 626).
- CRS (pp. 627–635). Conceptual role is abstracted causal role, illustrated for the material conditional by three inference tendencies; the “crucial question” is what counts as same and different role, faced with the collateral-information problem (pp. 628–629). Social meaning (Burge’s arthritis) goes to the referential factor or a third, social factor (p. 630). Field’s conditional-probability version is set aside: Bayesian change of degree of belief cannot model generation of new hypotheses (pp. 630–631).
- Two factors or one (pp. 635–639). Harman’s long-arm roles need normal context, experts and borrowed-reference chains; chop off the external part and the difference from two-factor theory is “merely verbal”.
- Meaning and reference (pp. 643–644). The conceptual role of names makes the causal theory of reference true (Kripke’s Moses intuitions are facts about how names function), so conceptual role determines the nature of the referential factor, and the function from context to reference, without determining reference.
- Learning (pp. 644–648). CRS explains why evidence of correct use is evidence of knowing meaning, and rejects premise 2 of “Fodor’s Paradox” (learning a word is hypothesising a definition in known terms), so it avoids the conclusion that ‘meson’ and ‘zipper’ are innate. Learning Newtonian mechanics: no definitions, only use.
- Meaningfulness and the competitors (pp. 649–660). Non-reductionist theories “put off the semantic issue”. Gricean theories need a naturalistic account of intentional content; Searle’s “equivalent causal powers” is not one, and the Loar–Schiffer version is as empirically exposed as CRS if there is a language of thought. Fodor refutes Dretske–Stampe; Fodor’s own theory fails on epistemic idealisation (Riemannian space, charmed quarks need the right theory, not only rubbed noses) and on teleology (a swamp-baby has meaning without evolutionary design; meiotic drive shows selection need not aim at truth), and it cannot separate nomologically correlated contents (electrical and thermal conductivity).
- Relativity and pictures (pp. 661–663). Meaning and syntax are both relative to system because both are functional (a reader of odd and one of even characters reads “CDAOTG” as CAT and DOG). So a pictorial representation can express an abstract concept “so long as the processors that act on it ignore the right specificities”, against Piaget and Premack.
- Autonomous and inherited meaning; the Chinese room (pp. 664–667). Both are conceptual role, and external roles are inherited from internal ones. The systems reply lacks a theory of what would make the system’s symbols meaningful; CRS supplies one, and distinguishes simulating a Chinese speaker’s information processing from emulating it. “What makes Searle’s argument sound so convincing is that it is difficult to imagine a version … that is a genuine instantiation or emulation rather than a mere simulation.” Searle’s target, symbol manipulation, is wider than Block’s in 1978, which was Ramsified commonsense functionalism.
- Explaining behaviour by functional states is not circular (pp. 667–669), as genes and reinforcers show: a web of effects, and independent identification of realisers.
Claims
| id | claim | strength | support |
|---|---|---|---|
| C1 | Two-factor CRS is the only approach with the potential to satisfy all eight desiderata | moderate as a comparative claim | desideratum-by-desideratum comparison, pp. 643–670; competitors judged on their own aims |
| C2 | One-factor long-arm CRS is equivalent to two-factor CRS | moderate | Twin Earth visitor, elm/beech, borrowed reference, pp. 635–637 |
| C3 | Conceptual role determines the nature of the referential factor but not reference | moderate | Kripke’s Moses case, pp. 643–644 |
| C4 | Fodor’s teleological–idealisation theory of truth conditions is not naturalistic | moderate | pp. 657–660; three separate problems |
| C5 | The Chinese room trades on the difficulty of imagining emulation as opposed to simulation | moderate–weak | pp. 665–666; Aristotelian-physicist analogy |
| C6 | Functionalism may be right for intentional states though not for experiential ones | assertion with a stated cost | p. 635; concedes possible zombies |
| C7 | Pictorial representations can express abstract concepts | moderate | functional argument, pp. 662–663 |
Concepts
- Conceptual role — “total causal role, abstractly described”, limited to relations mediating inference, decision making “and the like” (p. 628).
- Narrow / wide meaning — individuation by description versus by referent (p. 619); narrow is “in the head” in the sense of supervening on physical constitution (p. 620).
- Autonomous / inherited meaning — meaning that needs no translation into the language of thought, versus meaning that does (p. 617).
- Short-arm / long-arm roles — roles stopping at the skin, versus Harman’s roles reaching into objects and community (p. 636).
- Standard association — the mechanical mapping between English and mentalese expressions set up by the language module (p. 632).
Connections
- Field, Loar and McGinn are the two-factor predecessors; Harman, after Sellars, the one-factor rival (p. 627).
- Searle (LIT-400, NOTE-349) for the Chinese room and intrinsic meaning.
- “Troubles with Functionalism” (LIT-407, NOTE-366): p. 635 explains the change of mind, consistent with NOTE-366’s C6 (a Psychofunctional homunculi-head could have beliefs) and C7 (qualia may not be in psychology’s domain).
- “Functional Role and Truth Conditions” (LIT-625) restates the Harman argument and points back to pp. 643–644 here.
- Fodor’s representational theory of mind as NOTE-117 describes it is the framework; Block’s commitment is the minimal combinatorial one.
Bearing on the record
- NOTE-117’s counter-model to Toon is functional role semantics as Block states it here and in LIT-625. Drayson needs only that CRS can let mental representations take content from social use; p. 630 offers that through the referential factor or a third, social factor, but does not develop it.
- For the record’s AI-consciousness readings (LIT-056, NOTE-052), p. 635 is a philosopher holding functionalism for intentional states while denying it for experience, the split Butlin et al. do not consider.
- No instruction for machine-learning practice.
Limitations
- No identity conditions for conceptual roles; the collateral-information problem is posed and left (pp. 628–629). The paper says so: it is an advertisement.
- The normative-versus-actual question about roles is raised and set aside (p. 631).
- The social factor is mentioned, not built.
- Arguments against the competitors are compressed; Fodor’s theory is criticised in an unpublished version (n. 61).
Open questions
- What makes two conceptual roles the same, and can a similarity gradient do the work of identity for psychological explanation?
- Is the “small part” of conceptual role that fixes the function from context to reference (p. 670) the same for all natural-kind terms, as Block tentatively suggests?