Broadly speaking, there are two general theories of how the content of mental
representations is determined: tracking semantics (also known as “causal-informational” semantics) and functional role semantics (sometimes called “conceptual role” or
“inferential role” semantics56). Roughly, tracking semantics holds that the content of a
mental representation is fixed by the causal-informational relations that hold between brain states and external objects and properties, whereas functional role semantics holds that the content of a mental representation is determined by the logical or functional relations it bears to other mental representations. Thus, on the tracking view, a brain state represents trees if there is a reliable causal or informational co-variance relation between that state and trees in the external environment. In contrast, according to functional role semantics (hereafter, “FRS”), a brain state represents trees if it bears the rights relations to other brain states that represent leaves, branches, forests, plants, wood, and so on.
56 As should be somewhat obvious, I use the term “functional” instead of “conceptual” because I am not
concerned with concepts as such here. Indeed, phenomenal content is often thought to be specifically non- conceptual (although I am neutral on that issue in this dissertation). Moreover, even if functional role semantics is the correct theory of phenomenal content, I do not claim that it also is the best theory of content for concepts.
These two approaches roughly correspond to the difference between adopting a
referential theory of content as opposed to a use theory of content. The referential view is by far the more dominant account of mental content, especially in regards to perceptual or sensory content. In contrast, FRS derives from the “use” theory of meaning (Grice, 1957; Wittgenstein, 1953), according to which the content of a representation (e.g., a word) is determined by the way it is used in communication.
Transposed into the domain of representationalism about sensory phenomenology, functional role semantics holds that a sensory state has a certain phenomenal content by virtue of the functional relations it bears to other sensory representations. This proposal fits very naturally with the quality space approach to sensory phenomenology: each location in quality space is defined entirely by the relations that it bears to other locations in the space. Likewise, each state of our opponent processing system is individuated by the well-defined functional relations that obtain between it and every other possible opponent process channel state. Thus, the response type r itself can be understood as possessing FRS content, which in turn can be identified with the phenomenal content of the state. On this view, a quality space is a representational space, with individual locations corresponding to particular contents. Paul Churchland (1989, 1995, 2005) has made similar proposals about sensory representation with a theory known as state space semantics (SSS). As he puts it,
“The basic idea...is that the brain represents various aspects of reality by a position in a suitable state space, and the brain performs computations on such representations by means of general coordinate transformations from one state space to another." (1989, p.78-79)
Notice that SSS/FRS is a theory about how the brain represents, not what the brain represents. In other words, the theory captures the intensional structure of a
extension of its representations. In short, FRS/SSS is concerned with content that is roughly analogous to sense, not reference.57
However, there is a problem with adopting a version of FRS/SSS that only recognizes internal relations to other representations in the system as being relevant to determining content. This view is known as “short-armed” FRS, and it seems vulnerable to the objection that some aspect of content is constituted at least in part by external factors. (Indeed, most philosophers have come around to some version of the basic Putnam/Burge view on externalism about meaning.) The more immediate problem, however, is the fact that a state is not a representation simply because it has certain relations to other states within a common system. In other words, mere relational structure does not provide extensionality or veridicality conditions. A location in quality space is not itself intrinsically representational, in the sense of being intrinsically ‘about’ external
properties. This is essentially the “symbol-grounding problem”; that is, the problem of how a “semantic interpretation of a formal symbol system be made intrinsic to the system” (Harnad, 1990). (I discuss this problem in more detail below in section 6.6.) Similarly, Fodor & LePore (1996) present a closely related objection when discussing Churchland’s state-space approach to sensory representation:
“…Churchland has confused himself by taking the labels on the semantic dimensions for granted. The label on a particular dimension says how positions along the dimension are to be interpreted; for example, it says that they’re to be interpreted as expressing degrees of F-ness. To label a dimension as the F-ness dimension is thus to invite the questions “In virtue of what do the values of this dimension express degrees of F-ness rather than, say, degrees of G-ness?”.” (1996, p.153)
57 As an example from Churchland (1989) illustrates, there can be clear divergence between the content
attributed by tracking semantics and the content attributed by functional role semantics: some utterance in a foreign language might track thunder (and thus be attributed the extensional/referential content “There is thunder”) yet, by virtue of the role that it plays in the language, it could actually mean something like “God is shouting”.
In short, although FRS provides the framework for understanding a particular aspect of the representational content of sensory states, by itself it is only half of a theory of representation. That is, although it provides a theory of the nature of sense (in terms of phenomenal MoPs), it still requires the addition of a theory of sensory reference. Thus, many proponents of FRS have supplemented the “short-armed” approach to FRS by adopting one of two options: (i) a ‘two-factor’ theory of content in which the internal functional relations determine a type of “narrow” aspect of content roughly analogous to sense or MoP and external relations (such as causal-informational tracking relations) determine the ‘wide’, referential aspect of meaning; or, (ii) a ‘long-armed’ (or “non- solipsistic”; Harman, 1987) version of the theory which includes relations to external properties and objects as part of the content-determining functional relations, “but without any commitment to a separable narrow aspect of meaning.” (Block, 1998) In what follows I will argue for a two-factor theory of sensory content, and against the “long-armed” version of FRS.
A “long-armed” version of functional role representationalism would hold that relations to external states play a role in determining phenomenal content. This is compatible with the claim that my microphysical duplicate and I could be in identical internal states but have different phenomenal experiences, since we could be systematically related to different things in the world (e.g., Block’s “inverted earth” scenario). This result is obviously incompatible with internalism. However, long-armed FRS doesn’t necessarily entail externalism about content; a version that was compatible with internalism about phenomenal character would be possible, but it would have the consequence that content underdetermines phenomenology. That is, representationalism is only committed to the claim that if there is a difference in phenomenal quality of an experience there must be a corresponding difference in the representational content of that experience. A version of internalist representationalism which holds that different contents could have the same phenomenology is thus technically compatible with representationalism; however, it seems to give up on the spirit of it.
In any case, the reason that a long-armed FRS for representationalism about sensory phenomenology doesn’t work is because it fails to make a principled distinction between those function relations which determine phenomenal character and those that determine the referential content. That’s why a two-factor theory of sensory content is necessary. As described above, a two-factor theory of phenomenal content posits two distinct
mechanisms of representation, one determining representational properties analogous to sense, the other determining representational properties analogous to reference (e.g., McGinn 1982). The joint contribution of these two mechanisms determines the representational content of the sensory state.