Jerry Fodor, LOT 2: The Language of Thought Revisited. Oxford, OUP, 2008.
Some quick observations on Fodor’s new book:
1. As usual, a great book: ambitious, provocative, full of ideas and arguments, and as a bonus, funny (for those who like his humor).
2. He is still a computationalist, language-of-thought representationalist, atomist and nativist about concepts, referentialist with respect to semantic content, and causal theorist about the origin of semantic content.
3. He stresses more than ever that although his theory is the best available, it fails to explain central cognitive processes. His main reason is that computation is local, whereas central cognition (e.g., inductive inference) is global. What this means is, roughly speaking, that computational processes (as he understands them) only work well when they manipulate a few representations at a time. As soon as computations try to manipulate too many representations (e.g., because they are looking for global properties of a large set of representations), they run into the intractability of the frame problem. But some cognitive processes (e.g., inductive inference) require taking into account too many representations for computational processes to be a feasible explanation of them. Fodor suggests that some “new” kind of computation might be able to explain such processes, but he doesn’t think anyone has any idea how this new kind of computation works. I wish he would have explained better why it doesn’t help to appeal to lots of parallel processes here (as, e.g., Paul Churchland does in his recent work).
4. One novelty is iconic representations, whose content is nonconceptual. Fodor argues on empirical grounds that cognition involves iconic representations as well as linguistic/conceptual ones. At the very least, iconic (nonconceptual) representations are the ones present in the “iconic buffer”, which is a processing stage postulated by some classical cognitive psychological theories. As far as I remember from his previous work, this is a new addition to Fodor’s theory.
5. As usual, Fodor relies on his semantic account of computation, according to which computation is a kind of manipulation of representations that “preserves” (some) semantic properties of the representations. This gets him into trouble when it’s time to explain how the representations acquire their content, because he can’t appeal to computational processes. (According to Fodor, for a computation to be in place, there must already be representations with their semantic properties in place.) So he ends up saying that the notion of concept learning is incoherent. Instead, he suggests that semantic content is acquired through – listen to this! – non-computational brain processes. The story gets pretty mysterious at that point; a “here a miracle happens” moment. Fodor’s own argument that acquiring semantic properties is just something the brain does is quite involved, though – he does not generate his conclusion simply on the grounds of his account of computation. But I stress the connection with the semantic account of computation because I’ve been arguing for some time that contra Fodor and many others, computation does not require representation. If I’m right, computation might help explain the origin of semantic content is ways that are precluded to Fodor et al.
6. The appeal to brain processes to explain the origin of content is surprising and ironic for someone, like Fodor, whose theory of mind seems to have been built (over many decades) by ignoring neuroscience as a matter of principle, and, occasionally, as a rationalization for why it’s ok to ignore neuroscience.
7. I strongly recommend reading this book. Like most of his previous books, it will become a standard reference for philosophers of mind.
Thanks for a concise summary, Gualtiero. As regards his reliance on global processes, I think the main problem is that Fodor always misunderstood what frame problem is and what isn’t. This definitely isn’t the original Hayes & McCarthy frame problem that seems to be quite nicely addressed by Lifschitz & Morgenstern automatically generated axioms for non-monotonic reasoning (so even if you’re doing GOFAI, you don’t have to worry that much about it). I could even write a simple algorithm for extracting such axioms from large corpora (you just need to use the standard summarization algorithms and word-sense disambiguation methods to define frequent contexts and distill some info on situations to see which default resoning axioms seem to hold). With the advent of really large datasets freely available on the web, you could even try this at home.
Thanks for the post.
Is Fodor’s use of triangulation (a form of which he attributes to Davidson) to solve the ‘Which Link?’ problem for causal theories of perceptual thought (pp. 205-215) also a novelty?
I did not find it very convincing.
The ‘Which Link?’ problem he says his approach solves relates to there being many things along a causal chain terminating in a perceptual thought other than that which we would say the perceptual thought is about: a cerebral cortex at one extreme and the big bang at the other.
Fodor’s account of Davidson’s version of triangulation in a situation of radical interpretation has the field linguist and the native encountering a hairy snake. The referent of the native’s resulting utterance is ‘the one where the causal chain from the world to the speaker (/thinker) intersects the causal chain from the world to his interpreter when the speaker and the interpreter are caused to utter tokens of the same type.’
Fodor’s adjustments to his account of Davidson’s version is to deal in thoughts rather than utterances, and to use a counterfactual counterpart of the same person to achieve triangulation, rather than a distinct speaker and an interpreter. Thus, if I perceive a hairy snake, I have a perceptual thought of the hairy snake – and not the big bang – in virtue of the fact that my counterpart possibly 3 feet to my right also tokens a perceptual thought that is the terminus of a causal chain originating at the big bang, but both causal chains intersect at the hairy snake.
First, both me and my counterpart share a cortex, so it is still my cerebral cortex and not the snake that is the first intersecting link in my causal chain and that of my counterpart. If Fodor wants to claim that it is relevant that my counterpart has a different token cortex because there is no transworld identity then the same condition would apply to the hairy snake, so that should make no difference. Thus, it cannot be the nearest intersecting node in the causal chain that determines the referent of our perceptual thought if we are talking about me and my counterpart.
Second, if proximity of intersecting nodes is not relevant then I do not see how triangulation makes any difference. If me and my counterpart truly are perceiving the same hairy snake (or counterparts thereof) then the hairy snake we perceive should have all of the same causal antecedents in the actual world and in all nearby possible worlds.
So, how does triangulation make any difference to solve the ‘Which Link?’ problem?
Thanks for the comment. Yes, I believe triangulation is a novelty in Fodor’s work. However, a somewhat similar idea was floated by Dretske 1981 to solve the same problem. If I remember correctly, Drestke appealed to different sensory channels converging on the same stimulus to solve the “which link” problem.
I agree with you to the extent that I believe Fodor did not give the best possible formulation and exposition of the triangulation idea. But I do agree with Fodor that some version of that idea is a plausible solution to the problem. Triangulation may be a misleading analogy.
I’d imagine a better notion to start with is feedback: the brain tries to respond to the stimuli: after trying to do stuff a few times, it figures out what the environmental invariants actually are – the stuff that gives you feedback when you touch it, etc. It would be interesting to know what neuroscientists say about this – how they think neural tissue “identifies” the actual stimuli. I would guess there is probably some feedback somewhere in there. But maybe this is not as close to Fodor’s solution as I thought it was?