Fodor has used compositionality to argue against the prototype theory of concepts, connectionism, and the possibility for recognitional concepts. In this recent paper he continues his attack against recognitional concepts in arguing that compositionality issues will make any epistemically based semantics a failed project. I address this argument in section 2 of this paper and argue that Fodor has not given those engaged in this project much to worry about because his notion of compositionality is too strong. In order to establish his thesis that thought rather than language has content in the first instance, Fodor argues that language is not compositional! It is not Fodor’s conclusion that thought has content in the first instance that I disagree with, rather it is the path via the rejection of the compositionality of language that I take issue with in this paper. Fodor even goes so far as to admit that the consequence of this is that the semantics of English is a study of nothing. Rather Fodor suggests that the semantics of English and any other natural language is actually the semantics of thought. In section 3 of the paper I show several places where Fodor’s argument against the compositionality of language goes wrong.

2 Epistemic Semantics

Fodor argues that there cannot be an epistemically based semantics and that this gives us some reason to believe that it is thought, not language, that has content in the first instance. For Fodor, there cannot be content that has epistemic properties as constituents because epistemic properties don't compose. This would be a big blow to critical philosophies. Again Fodor is picking on Aunty^[1]. In this section I hope to show that all this spleen directed at poor old Aunty is unjustified. Below I will briefly explain the relationship between epistemology and semantics as Fodor sees it and summarise his argument against epistemically based semantics. I will argue that Fodor's notion of the principle of compositionality is too strong. Once compositionality is loosened up a bit it will become apparent that critical philosophy is in no danger from Fodor's arguments.

Very roughly, to justify a knowledge claim is to give reasons. (e.g. I know that the glass is on the table because I can see it and the conditions are good for seeing etc.) Not just any reason will do. We need good reasons. What determines whether a reason is good? More reasons? Presumably the reason giving has to end somewhere in order to avoid a regress. What is needed are some self-justifying reasons. This, in broad brush strokes, is the story we get from foundationalist epistemology since Descartes. Descartes' foundation was a God that does not deceive. In this secular age philosophers have been unhappy with this solution and have tried to find other self-justifying reasons. Fodor points out two attempts at this that have been made. One way is to argue that some reasons are self-justifying because they are true by virtue of meaning. So if I say that I am a bachelor then it is true by virtue of the meaning of the concept BACHELOR that I am also unmarried. If you know that I am a bachelor then you know that I am unmarried. The other way is to argue that some reasons can "just be seen" to be true. Fodor's example is that if its true that the concept RED can just be seen to apply, then in the right circumstances etc. its appearing to apply is warrant for applying it. In both cases, Fodor points out that these epistemic properties are suppose to be constitutive of the concepts. "The aspect of this I'd like you please to attend to, is that being the kind of concept that can serve as a constituent of a self-justifying claim is supposed to be concept-constitutive [...] if it's part and parcel of the concept RED that it's 'observational', then having RED is (inter alia) being able (at least some of the time) to apply it just by looking." (Fodor 2001, p. 4 emphasis is Fodor's) In both cases it is the theory of meaning that is providing the foundation for the epistemology.

Fodor argues that there is a problem with making a theory of meaning do double duty as a theory of meaning and as a foundation for epistemology. The essential point to Fodor's argument is the contention that epistemic properties don't compose and the principle of compositionality is essential to any theory of meaning.

Fodor's argument against epistemically based semantics takes the following form.

1. As between the two, at least one of thought and language must be compositional.

2. If only one of them is, then that's the one that has content in the first instance.

3. Epistemic properties do not, in general, compose.

Conclusion: Whichever between language and thought has content in the first instance cannot have epistemically based semantics. Below I will consider each premise in turn. The reason Fodor posits premise one is that we must be able to explain the productivity and the systematicity of language and/or thought. Fodor defines productivity as follows, "productivity is the property that a system of representations has if it includes infinitely many syntactically and semantically distinct symbols." (Fodor 2001, p.6) So as not to beg any questions, productivity ought to be re-formulated in terms of a capacity to produce infinitely many new complex representations from finite semantic and syntactic resources. Connectionists such as Smolensky (1991) have argued that connectionist models using distributed representations rather than symbols can embody compositionality and thus have the capacity for productivity. According to Fodor, systematicity "is the property that a system of representations has (whether or not it is productive) if each of the symbols it contains occurs with the same semantic valued as a constituent of many different hosts."(Fodor 2001, p. 6) An example of systematicity is if someone can think and/or say 'Aunty loves critical philosophers' then someone can think and/or say 'Critical philosophers love Aunty.'

Again this account should be reformulated such that it doesn't presuppose representations are discrete symbols rather than distributed across a set of connections and their weights. It should be said that Fodor and various co-authors (Fodor & Pylyshyn 1988; Fodor & McLaughlin 1990) argue that compositionality, systematicity, and productivity cannot be accounted for in connectionist models and thus give us good reason to believe that they are not a good models of cognition. As for Fodor's account of compositionality itself, he is not too explicit. It is unfortunate that he plays so fast and loose with a concept that is essential to the entire thesis of his paper. "So not-negotiable is compositionality that I'm not even going to tell you what it is." (Fodor 2001, p. 6)

Fortunately he does give us some clue as to what he means. Fodor gives us essentially two versions to choose from, one stronger and one weaker.

Weak Compositionality^[2]: "It says that the semantic value of thought (/sentence) is inherited from the semantic values of its constituents, together with their arrangement." (Fodor 2001, p.6)

Strong Compositionality: "In fact, (and this is no small matter) the connection that compositionality imposes on the relations between the possession conditions of concepts and the possession conditions of their hosts goes in both directions. That is, compositionality requires not just having the constituent concepts is sufficient for having the host concept, but also (and more obviously) that having the host concept is sufficient for having its constituents." (Fodor 2001, p. 9)

This second formulation is stronger because it requires that the possession conditions for concepts go in both directions while the first formulation, strictly speaking, only requires that they go from the constituents to the host but not necessarily from the host to the constituents. One might also think of strong compositionality as bi-conditional compositionality and weak compositionality as supervenience compositionality. With this preliminary discussion out of the way, we are now in a position to assess this argument.

I will consider each of the premises above in turn. I will accept the predominant view that the principle of compositionality in some form is necessary. My only reservation about premise one is that the stronger version of compositionality Fodor gives is questionable as I will argue below. With regard to premise two I wish to keep open the possibility that it is possible that both language and thought are compositional and further that compositionality might admit of degree. Fodor, himself has talked as if this were a distinct possibility. "You need to assume some degree of compositionality of English sentences to account for the fact that systematically related sentences are always semantically related..." (Fodor & Pylyshyn 1988 p. 126 emphasis is mine). I take it then that the formulation of premise two that Fodor gives is a little misleading as to the options available.

The meat of the argument is in Fodor's justification for premise three. Fodor takes the concepts RED, SQUARE, and RED SQUARE as an example. He argues that compositionality, requires that host concepts exhaustively inherit their semantic values from their constituents. So if recognition conditions are constitutive of concepts then being able to recognize RED by itself under favourable circumstances and SQUARE by itself under favourable circumstances should be sufficient for recognizing RED SQUARE. According to Fodor however, this is not true. It could be that the conditions for recognizing the host 'screen off' conditions for recognizing the constituents. If this is the case, then not all of the semantic properties of the constituents are being inherited by the host and this violates the principle of compositionality. Compositionality is non-negotiable because it is needed to explain productivity and systematicity. From this argument Fodor concludes that at least some concepts are not constituted by recognitional capacities and that this argument will apply mutatis mutandis to preclude other putative epistemic possession conditions on the possession of complex concepts.

Fodor has also cast this argument in terms of "good instance-hood." The idea is that instances can compose, but that good instances cannot. However I would like to not here that Fodor seems to be confusing his arguments against prototype theories of concepts (see Fodor 1998) with an attack on epistemically based semantics. This distinction is put nicely by Richard Grandy (1998, p. 23), "The prototype theorist is committed to our being able to recognize good examples of red as good examples. The empiricist, however, is only committed to our recognizing good examples as examples." In other words, Fodor confuses good examples from how we recognize them and it is the recognitional capacity that the epistemic semanticist is interested in.

It is difficult to find this argument convincing, especially given Fodor's example. Perhaps it is a lack of imagination on my part, but I can't for the life of me think of recognition conditions of RED SQUARE that would preclude recognizing either RED by itself or SQUARE by itself. What condition could this possibly be? It certainly isn't any visual condition. Any visual conditions that would prevent someone from recognizing RED or SQUARE would prevent someone from recognizing RED SQUARE. Though conceivable, it is indeed difficult to come up with many examples of complex concepts whose recognition conditions 'screen off' the recognition conditions of any of its constituents. We should expect examples to be easy to come by if this is going to be a convincing argument, but that just doesn't seem to be the case. I said above that it is a possibility that compositionality is a property that can come in degrees. It may well be that epistemically based semantics is compositional to the degree that it avoids such examples. Since this 'screen off' effect seems prima facie a rare phenomena, the critical philosopher need not worry too much about these cases. Fodor needs to show us that a significant number of complex concepts or even all complex concepts in an epistemically based semantics are not compositional. Indeed, this is what he attempts to do by making use of the strong version of compositionality.

What I have termed “strong compositionality” requires that hosts receive their semantic properties from only their constituents and also that constituents transmit all of their semantic properties to the host. Fodor argues while compositionality must have this bi-conditional structure, epistemic conditions on concept possession don't. He argues that one could imagine a person who has a recognitional capacity for RED SQUARE but who may not have a recognitional capacity for RED. If this is true it violates compositionality, compositionality is non-negotiable and thus no concepts can be constituted by epistemic properties. "For example, you might HAVE a concept RED that occurs only in host contexts. In that case, you'd be able to think red square and red triangle but not red tout court..." (Fodor 2001, p. 9) Under Strong compositionality this would not be possible, but isn’t this a plausible story? Fodor seems to think it is. "(That kind of mind wouldn't even be very surprising, since thoughts about red tout court are more abstract than thoughts about red squares.)" (ibid.) Fodor can't have it both ways. Either "that kind of mind" isn't plausible and strong compositionality is true, or "that kind of mind" is possible and strong compositionality does not hold. I contend that it is strong compositionality that doesn’t hold.

This argument relies on the strong version of compositionality, because what is supposed to violate compositionality in this example is that the recognitional capacities are not travelling in the host-to-constituents direction. Another way to think of this is that recognitional constituents do not retain all the semantic properties they had when the complex expression is decomposed. This requirement is too strong. The problem is that complex contents acquire content not just from their constituents, but from the arrangement of those constituents as well. Fodor often formulates compositionality as to include the semantic properties of the constituents together with their syntax. However, he doesn't take the part about syntax seriously. Indeed, in places he seems to forget it altogether. "Or, to put it slightly differently, compositionality requires that host concepts receive their semantic properties solely from their constituents..." (Fodor 2001, p. 9 emphasis mine) There are well known ambiguities in sentences that can only be disambiguated syntactically. For example, consider the following sentence.

Chris saw the man with a telescope.

The sentence is ambiguous. It could mean Chris saw the man by looking through the telescope or it could be Chris saw a man who was in possession of the telescope. This can only be disambiguated syntactically. The contribution of syntax to meaning is straightforward when you go from constituent to host, but when you go from host to constituent (decomposition) it is problematic. The strong version of compositionality would require that each constituent take along with it as content the meaning roles it could play in an indefinite number of syntactically different sentences. This consequence undermines the need for compositionality to explain productivity.

A couple of analogies may be helpful here. If we imagine creation of hosts as a construction process we can see what happens. First we gather up all the constituent parts including all their semantic properties. Next, we arrange these constituents according to various syntactic rules. Now we have a new complex expression. The problem is that this process cannot be reverse-engineered such that the constituents can take all the semantic properties of the host with them. Part of the semantic value of the host consists in the arrangement of the constituents and not just in the constituents themselves. When your computer screen constructs an image, it just uses thousands of little coloured dots. If I separate out these dots and arrange them according to colour, say, it would be impossible for you to predict what the original image on my screen was. Imagine the semantic value of each dot is its colour. When I deconstruct the image the dot retains its colour (semantic value). However it does not bring with it some mysterious property as to how it was arranged relative to the other dots. Likewise, when I deconstruct a host concept, the constituents will retain all the semantic properties they had before they were used to construct the host, but not all of the semantic value of the host itself. This is even true for a straightforward case like RED SQUARE. If we just put RED and SQUARE together we do not capture that it is the square that is red rather than there just being some redness and some squareness around. It is the syntactic role RED takes as a adjective and the syntactic role that a SQUARE takes as a noun that gives us A SQUARE THAT IS RED, presumably this is the concept Fodor had in mind.

The fact that we use English as a mediating representation to express concepts accounts for some of our intuitions^[3]. We don't see the importance of syntax when we represent the concept as "red square" while we do when we represent it as "a square that is red." Just using capitals to suggest that it is the concept that we are talking about rather than the word can still be misleading. While constituents inherit their original contribution to the sentence with them when they are detached from their host, they do not inherit the semantic effects of the syntactic arrangement. Compositionality does not require that the constituents inherit all of the semantic properties from the host. This shows that it cannot be a requirement of compositionality that all of the semantic value of the host transfers to constituents when you move in that direction. Since Fodor's argument relies on this being the case I conclude that he is not warranted in his conclusion against epistemically based semantics.

3 Language and Thought

In this section I consider Fodor's argument in favour of it being thought that has content in the first instance. What I want to argue here is that it is not necessarily Fodor's conclusion that is false, but that his reasons for thinking so are not good reasons. One might have good reason to think that it is true that thought has content in the first instance for independent reasons. I specifically take issue with the premise that natural language is not compositional. Following G.E. Moore, if a conclusion is absurd, it is probably the case that either the conclusion doesn't follow from the premises or that one of the premises is false. Thus I will show where Fodor goes wrong in arguing that language is not compositional.

I begin by summarising Fodor's argument. The general argument of the section takes the following form.

1. As between the two, at least one of language or thought must be compositional.

2. The evidence suggests that language is not compositional.

Therefore, it is thought rather than language that has content in the first instance.

I would again like to mention the possibility that both language and thought admit of some degree of compositionality. It seems a bit simplistic to suggest that compositionality tracks content in a linear fashion, however it is plausible that language and/or thought could be imperfectly compositional and be the bearer of content in the first instance. Fodor argues that natural language is not compositional, but it remains an interesting question what his argument shows if it is sound. Is it that English is not compositional at all, or that it is just not completely compositional. It surely must be the latter. A very difficult question, which I do not propose to answer here, is what level of compositionality is adequate in order to bear content in the first instance? In order to establish his claims Fodor needs to give a some sort of explanation of this and I say good luck. Having noted this problem I now turn to the crucial and strange premise that natural language "just isn't compositional."

So here is the outline of the argument that natural language is not compositional.

1. Compositionality requires that the content of a composite (a thought or linguistic expression) contains all the content of its constituents.

2. "The content of a sentence is, or is the same as, the content of the corresponding thought." (Fodor 2001, p.11)

3. "...the constituents of the sentence would have to correspond in a straightforward way to the thoughts constituents." (Fodor 2001, p. 12)

4. If language is compositional it can't be elliptical and inexplicit about the thought that it expresses.

5. Language is elliptical and inexplicit about the thought that it expresses.

Therefore, language is not compositional.

We are familiar by now with the requirements of compositionality from section II in which I argued for weak compositionality, so I take it that premise one is straightforward. I take premise two to be rather straight forward as well. Though it might serve other purposes, a primary function of language must be to express our thoughts to others. The idea behind premise three is that if there are constituents of your thoughts that don't correspond to constituents of a sentence one utters, then it follows from premise one that there was something in the thought that the sentence left out. Also possible is that there is a constituent in the sentence that does not correspond to a constituent in the thought. In this case the sentence has more content than the thought. Presumably the bearer of content in the first instance will be whichever between the two contains more content or the complete content. This gets us premise four, because saying that language is inexplicit and ellipical about thought would be to say that it has less content than the thought. If this were the case then it would be thought that has content in the first instance. So the picture is that constituents of language and thought will have one to one correspondences.

All that remains is to give examples to show that language is indeed elliptical about the thoughts it expresses. Fodor thinks that it is not tendentious that people don't say everything they are thinking. Fodor gives the following example. If someone asks me what time it is and I say "it's three o'clock" I have left out part of my thought. What I was thinking was something like "it is three o'clock in the afternoon here and now." For Gricean reasons we leave out these details when we communicate with someone. One doesn't bother to say what is obvious in the shared intentional context. So there we have it. According to Fodor there are constituents of our thoughts that our sentences don't express. Since compositionality requires that the content of thought contains all the content of its constituents, language cannot leave out some of these constituents and be compositional.

Suffice it to say this argument goes wrong in lots of places. First I want to examine what looks, prima facie, like a contradiction in Fodor's argument. Fodor states that English is compositional early in his paper and follows this by saying that the evidence suggests that language is not compositional later in his paper. Following a discussion of this issue I will argue that premise three is false because Fodor makes an unwarranted assumption about the representational structure of thought. Finally, I will argue that premise five is false because thoughts need not be as precise as Fodor assumes.

What makes Fodor's argument that natural language is not compositional so shocking is that it is has been a paradigm case of compositionality. Indeed, Fodor agrees in other incarnations. "linguistic capacity is a paradigm of systematic cognition" Fodor & Pylyshyn 1988, p.120). Of course, compositionality is what explains systematicity. Important arguments in cognitive science such as poverty of stimulus arguments for an innate universal grammar rely to some extent on language being compositional. The idea here is that if we did not have an innate compositional grammar, we could not learn language. Fodor's famous 1975 hypothesis is indeed called the language of thought hypothesis because it is supposed to be language-like. One reason it is supposed to be language-like is that this buys us compositionality and thus an explanation for the productivity and systematicity of thought. So the exasperated question is why does Fodor write at the beginning of his 2001 paper, "English being compositional is what explains why so many of the sentences that you can use to say things about pigeons and the weather in Manattan, share some or all of their vocabulary." ( Fodor 2001, p.7) Then he turns around at the end of the very same paper and writes, "The evidence suggests strongly that language is not compositional." (Fodor 2001, p. 14). Without further explanation this is a blatant contradiction.

Fodor's only possible response to this is that thought is compositional and that natural language somehow derives its compositionality from thought. A natural question to ask at this point is 'how do we know that thought is compositional?' In Fodor & Pylyshyn (1988) they argue that sentences are used to express thoughts and since sentences are compositional then thoughts must be. According to Fodor & Pylyshyn, if we can understand various sentences that relate to one another in systematic ways, then this must be because the thoughts or mental representations they express have a constituent structure and are compositional. The compositionality of natural language is being used as evidence for the compositionality of thought. Nonetheless the compositionality of language is derived fromm the compositionality of thought on Fodor's view. The interesting thing here is that what we seem to have direct empirical evidence for, contra Fodor, is the compositionality of language. Usually when one claim is taken as an evidential base for another claim, it is the evidential claim that is already accepted as being true. I don't think I have shown a vicious circularity here, but the evidential relationship between language and thought that Fodor has posited is at the very least a strange and disconcerting one.

It is important to be careful here. It is one thing to say that the content of language can be derived from the content of thought and another to say that the compositionality of that content can be derived from the compositionality of that thought. This assumes premise three is correct. Premise three is that the structure of the content of thought must match, in a straightforward way, to the structure of the content of natural language sentences. All that is needed is a set of representations that allow us to be sensitive to the relevant features of language. The phenomenon being explained is the compositionality, systematicity, and productivity of our language use because language is our evidential basis for positing that thought exhibits these properties. All that need be the case is that thought be sensitive to these features in a way that allows us to understand and use language in systematic and productive ways. Robert Cummins(1996) makes a distinction between a structural representational scheme and an encoding scheme. A structural scheme is one whose structure matches the structure of the content represented. Cummins calls the scheme from premise three a classical scheme. "Another way of getting at the idea of classical representation is that a scheme is classical if representing a complex content requires a correspondingly complex representation, with each constituent of the representation corresponding to each constituent of the content represented." (Cummins 1996, p.592). Structural representation is one way we can account for the compositionality of language. In this case the representation works in exactly the same way as the language to achieve compositionality. Another possibility is encoding. Cummins explains this notion as follows. "By an encoding of a domain D (which might itself be a representational scheme), I mean a recursive mapping of the members of D onto the representations in scheme S whose members do not preserve the internal complexity of the members of D." (Cummins 1996, p. 599) Smolensky, LeGendre, and Miyata (1992) show that encoding is sufficient to show sensitivity to the systematicity and productivity of language by proving an equivalence between a parser written in TPPL, a LISP-like language that uses classical representations (in the above sense) and a connectionist network using fully distributed representations. In such a network there will not be a one to one correspondence between activation vectors and constituents in a natural language sentence. What I want to conclude here is not that there cannot be a one to one correspondence of constituents of thought to constituents of language, but that it is not necessary that there be such a correspondence. The point is that we don’t know what the structure of thought is until we have more empirical evidence and Fodor hasn’t given us a conclusive reason to think that this correspondence relation between language and thought must hold. This is all that is needed to show premise three is false.

To return to the main argument against the compositionality of natural language let us see how the falsity of premise three undermines the argument. If premise three is false it is much more difficult to tell if language is ellipitical and inexplicit in relation to thought, because we can't tell what structure the thought has simply by looking at a linguistic expression of that thought. Perhaps Fodor has been bewitched by language after all. This returns to Robert Cummins (1996) idea that I mentioned section two. One can be fooled about what the structure of a thought is by the way we represent its content in discussion. When writing a paper about content we have to represent that content using natural language. Natural language is the mediating representation. We get our intuitions about what the structure of thoughts might be by means of this representation. I will borrow an example from Cummins (1996). If our representational scheme is pictorial rather than sentence-like then thought might exhibit different systematicities than our lingua-form mediating representations suggest. So one might say, anyone who can think that a face is smiling can think that a face is frowning. This systematicity doesn't look obvious at all from the structure of the sentence I just represented it with. However what if our representational constituents were like these.

Fig. 1

Now it is very obvious that for a mind with this sort of representational scheme the systematicity above holds. Again the problem is we don't know, a priori, what the thought looks like.^[4]

The problem gets even worse for Fodor. It is probably the case that thoughts can be elliptical and inexplicit in relation to the sentences used to express them. Another way to put this is that thoughts can be shorthand for the content of the sentence. Ironically, if this is true and the rest of Fodor's argument from compositionality is sound, then it is natural language that has content in the first instance because by Fodor's lights thought would not be compositional. If you think there are independent reasons for thinking that thought has content in the first instance this would be nothing less than a reductio of Fodor's argument. So what sort of expressions have more content, as it were, than their corresponding thoughts. The short answer is anything the requires deference to experts. Suppose I think E=MC2. Since I am not an expert physicist I don't know the entire content that the thought I had expresses. I might not have even known that "E" stands for energy or that "C" is the speed of light. Another example that doesn't even require deference to experts follows. Suppose I think FODOR IS WRONG FOR ALL THE REASONS I SAY ABOVE. What I mean by this thought is the content of all the sentences above. The thought is being used as shorthand for all of the linguistic expressions I used in the paper. It seems that in this case it is the sentences of the paper that are the bearers of content in the first instance. I conclude from all this that premise five is false and in some cases just the opposite is true.

4. Conclusion

Fodor seems to be willing to follow his version of compositionality wherever it takes him, no matter how extreme the conclusion. We saw above that he argues that there cannot be any recognitional concepts. We saw that on the construction side of complex expression building Fodor shows at best that there are some expressions cannot be recognitional. The epistemic semanticist need not fear this. After all it is not necessary that all primitive concepts are recognitional, all that is needed is that just that some are. Fodor tried to make the stronger argument that there are no recognitional concepts by making use of his strong version of compositionality which requires that the semantic value of the complex concept survives decomposition by splitting off with the constituents. This conception of compositionality was shown to be problematic and thus unnecessary, the weaker conception of compositionality will suffice.

The more radical conclusion Fodor reaches is that natural language is not compositional. Now he never specifies if he means not compositional at all or merely to one-hundred percent compositional. I took him to mean the latter. This argument was shown to be problematic on many fronts. Perhaps the most damning point made against his argument is that it's putative compositionality of language that rely on as evidence that thought is compositional. Fodor's mistakes can be diagnosed to the overly simple relation he sees between the structure of language and the structure of thought.

References

Cummins, R. 1996. "Systematicity." The Journal of Philosophy, 93(12) pp. 591-614.

Fodor, J. 1975. The Language of Thought. Harvester.

1987. Psychosemantics. MIT Press.

1998. Concepts: Where Cognitive Science Went Wrong. Oxford University Press.

2001. "Language, Thought and Compositionality." Mind and Language 16(1). reprinted in A. O'Hear ed., Philosophy at the New Millenium, Cambridge University Press.

Fodor, J. and McLaughlin, B. 1990. "Connectionism and the Problem of Systematicity." Cognition, 35. Reprinted in Macdonald and Macdonald, 1995. pagination from reprint.

Fodor, J. and Pylyshyn, Z. 1988. "Connectionism and Cognitive Architecture." Cognition, 28. Reprinted in Macdonald and Macdonald, 1995. pagination from reprint.

Grandy, R. "Recognitional Concepts and Compositionality." in Villanueva, E. (ed.) Concepts Philosophical Issues, 9, 1998. Ridgeview Publishing Co.

Macdonald, C. and Macdonald, G. 1995. Connectionism: Debates on Pyschological Explanation.

Blackwell.

Schiffer,

Smolensky, P. 1991 "Connectionism, Constituency, and the Language of Thought." in B. Loewer and G. Rey (eds), Meaning in Mind: Fodor and his Critics Oxford, Basil Blackwell. reprinted in Macdonald and Macdonald, 1995. pagination from reprint.

Smolensky, P., LeGendre, G., and Miyata, Y. 1992. "Principles for an Integrated Connectionist/Symbolic Theory of Higher Cognition." Tech Report 92-08, Institute of Cognitive Science, University of Colorado.

^[1] Aunty is Fodor's name for philosophers who do a priori philosophy

^[2] The terms "Strong compositionality" and "weak compositionality" are also used in Smolensky 1988, Fodor & McLaughlin 1990, and in other places in an exchange of papers between these authors. My use of the terms here should not be taken to refer to the concepts in those papers. I used the terms to refer to two formulations that Fodor makes of compositionality in Fodor 2001.

^[3] I owe this idea of mediating representation and its effect on our philosophical intuitions to Cummins 1996.

^[4]A similar criticism can be made against premise five. Fodor claims that language is elliptical and inexplicit about the thoughts that it expresses. But how do we know exactly what the thoughts are? Fodor seems to think that they are very detailed and precise. For example, he argues that when I tell someone that it is "three o'clock" this conveys the thought ITS THREE O"CLOCK IN THE AFTERNOON. He goes on to mention that we don't say "in the afternoon" for Gricean reasons. However, why stop here? Maybe the thought is actually ITS THREE O'CLOCK IN THE AFTERNOON OF MAY 5TH 2002 ON THE GREGORIAN CALENDER. I tend to think that my thought probably was just THREE O'CLOCK. I certainly don't sub-vocalise, "It's three o'clock in the afternoon." It's probably the case that we don't remember precisely what the thought was in the first place. So my question for Fodor is, beyond mere stipulation how do you know the thought must be as precise as ITS THREE O'CLOCK in the afternoon? If this question isn't answered we have no reason to think that the expression "its three o'clock" is elliptical and inexplicit about the thought it expresses. However this enters into a controversial debate in philosophy of language on whether we need to acknowledge unarticulated constituents.

Copyright © 2002 Christopher Jenson. All rights reserved