According to the strong version of the “mindreading is prior” model (MPM), all metacognition is interpretative. On the weak version, we have non-interpretative access to both sensory states and propositional attitudes. Carruthers' version of MPM is a middle-ground position. In accord with the strong version, he insists that first-person awareness of propositional attitudes is always self-interpretative and, hence, never “introspective” (in his quasi-pejorative sense). However, he denies that self-attribution of “sensory-imagistic” states is interpretative, claiming that such states are introspectively available as data for the mindreading system. On his view, “the mindreading system can receive as input any sensory or quasi-sensory (e.g., imagistic or somatosensory) state that gets ‘globally broadcast’ to all [cognitive] systems” (sect. 2, para. 3). The set of such states includes “perceptions, visual and auditory imagery (including sentences rehearsed in ‘inner speech’), patterns of attention, and emotional feelings” (sect. 2, para. 6).

I argue that Carruthers' appeal to the distinction between sensory states and propositional attitudes involves an error, and that avoiding this error leads to the collapse of his view into one of the competing versions. The preferable collapse is, I argue, toward the stronger view.

The states that Carruthers takes to be introspectively available are supposed to be “sensory-imagistic,” not conceptualized or propositional. But it is not obvious that perceptual judgments satisfy this description. Perceptual judgments are plainly a species of propositional attitude. True, such states may also have qualitative properties, but they are nevertheless constituted by concepts, and have “sentence-sized” intentional contents. Carruthers acknowledges this in discussing how the mindreading faculty would make use of perceptual judgments:

In explaining why perceptual judgments appear on his list of introspectible states, Carruthers's appeals to global workspace theory. That perceptual judgments are introspectible is, in his view, “pretty much mandatory once we buy into a global broadcasting architecture” (sect. 2, para. 8). That's because, in perception, “the initial outputs of the visual system interact with a variety of conceptual systems that deploy and manipulate perceptual templates, attempting to achieve a ‘best match’ with the incoming data. … [T]he result is globally broadcast as part of the perceptual state itself” (sect. 2, para. 10).

Although this is certainly plausible, Carruthers neglects the fact that similar grounds are available for the claim that other propositional attitudes are broadcast as well. Indeed, given his concession that global workspace theories allow for the broadcasting of at least one propositional attitude, one wonders why he assumes that such theories would not allow for the broadcasting of all the rest. On the face of it, the claim that all propositional attitudes can be globally broadcast has much going for it. Intentions, for instance, routinely recruit a wide array of cognitive resources, as do the conceptual-intentional aspects of emotions like fear and anger (e.g., that one is being attacked). Why not count these as instances of global broadcasting? Carruthers does not say.

Pending further argument, we should assume, pace Carruthers, that global workspace theory does allow for the broadcasting of all propositional attitudes. If so, then whatever we say about first-person access to sensory states, we should say the same about first-person access to propositional attitudes.

Do these considerations support the view that the mindreading system has direct, non-interpretative access to all mental states, both propositional and sensory? Not if one also rejects Carruthers' assumption that globally broadcast states are ipso facto available to the mindreading system in a non-interpretative fashion. Below, I explore grounds for adopting the strong version of MPM, according to which self-attribution is interpretative in the case of all mental states.

Interpretation takes place by deploying a propositional attitude that emerges from a background of theoretical commitments. Consequently, the cost of embarking on an interpretative venture is the possibility of partial misconstrual or wholesale error. These characteristics of interpretative activity fit well with Carruthers' usage of the term “interpretative,” as applied to mechanisms of self-attribution.

As Rosenthal (Reference Rosenthal2005) has argued, self-attribution is a matter of tokening potentially erroneous, theoretically loaded propositional attitudes – occurrent higher-order thoughts (HOTs). On this view, confabulation and error occur even with regard to sensory states. Dental fear, for instance, is a phenomenon in which dental patients under the drill mistake fear, anxiety, and a sensation of vibration for pain in a fully anaesthetized or nerveless tooth – a compelling demonstration that HOTs need not be veridical.

Nevertheless, judged on independent grounds, self-attributions of sensory states are often relatively accurate. Doubtless, this consideration compels theorists to posit a reliable monitoring mechanism, such as Carruthers' mindreading system. But, as Rosenthal points out, simply positing such a mechanism amounts to no more than stipulating a solution to the problem of explaining the frequent accuracy of HOTs about sensory states. An explanatory account of the mechanism's accuracy is not provided.

Extending Sellars's (Reference Sellars1956/1997) treatment, Rosenthal argues that HOTs concerning sensory states arise as a result of a creature's reflection on cases in which its perceptual judgments are mistaken. The creature formulates a rudimentary theory, in effect positing qualitative sensory states as the causes of nonveridical perceptual judgments. Against the background of such a theory, the creature is disposed, for instance, to construe itself as having a sensation of red when perceiving a red object.

Carruthers gives no grounds for rejecting this alternative and appealing picture. Global broadcast theory does not, by itself, settle the issue, for it is consistent with the claim that the mindreading system relies on a tacit theory in interpretatively self-ascribing sensory states. Nor does the data from autistic children disconfirm Rosenthal's view, which allows that even nonlinguistic, cognitively unsophisticated creatures may come to have HOTs concerning their sensory states. All that is required is that such creatures take note of their perceptual errors and account for them.