by Noam Chomsky
Finally, a point about scientific concepts as opposed to those found in common sense. With scientific concepts, it makes no sense to say that somehow they succeed at (we think) tracking the way things are because they have their forms and characters as a result of some kind of evolutionary adaptation. Peirce thought that this sort of explanation was plausible. That might make some sense with our commonsense concepts (although see below) because they – or some of them, or (some of) the system(s) that yield them – might have been in place for millennia, allowing for evolutionary adaptation, perhaps through several species (Hauser, Chomsky & Fitch 2002). However, it makes no sense with concepts such as LEPTON or ALEPH-NULL and the like, or even concepts such as ENERGY, FORCE, or MOTION as these are understood and employed in physics. As suggested earlier in discussion, no adaptive forces could have yielded them. They are recent inventions by human beings, and having them apart from the frameworks in which they are lodged – themselves artifacts – could not have provided reproductive or other advantages to humans or other creatures beforehand. Some of the contributions of our minds to the capacity to ‘do science’ and construct scientific theories are no doubt innate, but that is far from claiming that we humans have ‘devoted’ systems yielding the concepts on triggering occasions. If that were the case, constructing adequate sciences would be a lot easier than it is, and in the case of physics, we would not have needed anything like the several centuries it has taken physics to get to the far from complete state it is now in.
Appendix V: Of concepts and misguided theories of them, and why human concepts are unique
V.1 Concepts and ways of going wrong
Almost everyone who allows that there are concepts agrees that word-like units express at least some concepts, that concepts are at least in part mental ‘entities,’ and that it is through concepts that minds gain access to the world. But there is not much else that people agree on.
Some approaches to concepts – those that adopt anti-nativist and externalist empiricist views – need mention only in order to reject them. A dominant version of an empiricist view is found in what is called “functionalism” in the philosophy of mind. Essentially, the functionalist maintains that concepts are the (epistemic) roles/functions of linguistic tokens in mental transactions that mediate perceptual/sensory inputs and behavioral outputs. Wilfrid Sellars's view of mind, discussed in several other appendices and at some risk of tedium, again below, is an example; he presents what is sometimes called a “conceptual role” view of words and the concepts they express. Words are seen not as Chomsky construes them – lexical items – but as ‘things’ of some sort (perhaps neural nodes, perhaps computer binary coding that is electronically instantiated) that function in computational systems that deal with the world and ‘solve problems’ in epistemically reliable ways. Functionalism can take several forms, including behaviorist and connectionist ones. The basic view is implicit in many psychological and philosophical works, including some of the most famous. Moreover, it is popular not just in the philosophy of mind but the philosophy of language. Because functionalists suppose that the mind/brain is something like a causal system that takes perceptual inputs, subjects them to computational operations thought of in terms of (epistemic) rule-following programs that yield reliable answers to environmentally set problems in order to yield successful behaviors or actions, it is easy to see their empiricist pedigree. One studies the mind as if it were a reliable mediator of external inputs and outputs (a version of externalism). And one conceives of the mind as gaining its capacity to act in a reliable way by virtue of habituating oneself to a linguistic community's settled epistemic habits, a rejection of nativism. Having begun with these assumptions about mind and how to proceed, one is invited to adopt others popular among empiricist approaches – that identifying a concept is a ‘holistic’ matter so that to identify a specific one, you must know its ‘place’ (role, function) in a cluster of concepts; that understanding is a matter of knowing/having a (holistic) theory about the world; that having a language is a kind of know-how; that learning a language is developing a (good, reliable) theory of the world; that learning a language is being trained to produce what the community takes to be correct behaviors/responses; and so on. While these kinds of views dominate philosophical and psychological discussion, they solve no problems. In fact, because they begin with the wrong assumptions about how to proceed in studying the mind and its contents, they create puzzles and mysteries. The empiricist approach commits the error that Wittgenstein warned against, trying to construct a theory of highly variable and heavily context- and user-dependent use or application of language. By attempting this anyway, empiricist philosophers and psychologists end up pursuing one or another form of Wittgensteinian Scheinstreit. Wittgenstein in his Philosophical Investigations pointed out that philosophical problems are not problems; there is no solution to them. They are presented and understood in such a way that solution is impossible. Thus, discussion and dispute is not just endless, but useless. Chomsky appears to be suggesting the same (at least of functionalism and related current philosophical dogmas, such as representationalism and physicalism [Chomsky 1996]), and tying this suggestion to a further one: if you want explanations and evidence rather than speculation and the mining of mixed and flexible intuitions that are based on unsupported and unsupportable assumptions about the mind and how to study it, employ the tools of naturalistic research – natural science – and, using these tools, look inside the head, not at heads and worlds and relations between them. That seems to be the only workable “game in town,” to use Jerry Fodor's phrase, but for a different end than his representationalist-externalist effort.
Because Fodor's (1998) view of mind and concepts assumes nativism, it is a useful counterpoint to Chomsky's views. He offers a variation on a Fregean account of words and the concepts they express. Words express what Frege called “senses,” and these in turn refer to or denote things or properties. Senses can differ even though they denote the same things. Think of Fregean senses as ways in which denotations can be “presented.” A denotation can be the same even though it is differently presented – it is denoted by different senses. A standard example is this: the words “morning star” and “evening star” denote the same thing (Venus), but differ in how they ‘present’ Venus; they differ in sense. Fodor psychologizes Frege's senses and calls them “modes of presentation,” or MOPs. What, then, is a concept? One might think that a concept is a MOP. It is, after all, what is in the head; it also develops automatically (Fodor 1998, 2008). However, Fodor with his externalist inclinations – clearly visible in his “representational theory of mind,” among other manifestations – wants to identify concepts in terms of their “wide content” which is, essentially, what he believes he can show MOPs are of or refer to (their denotations). Essentially, he claims that the MOP for, say water develops automatically in a person as a result of some kind of causal informational relationship to – or predominantly to – water ‘out there,’ and that this causal relationship also establishes an inverse semantic relation, denotation, so that the water-MOP denotes the property being water ‘out there’ (and, he insists, being H2O ‘out there’ too). Some of the errors with this view are taken up in the discussion in the main text. There is nothing wrong with holding that MOPs are acquired by some kind of causal relationship; any nativist account holds that concepts develop as a result of some kind of ‘triggering’ relationship. But there is no reason to believe that a semantic relationship of denotation piggybacks on the world–head causal triggering relationship. For further discussion, see my (2002a and 2010). In any case, because both MOP and (supposed) denotation figure in Fodor's account, we can for our purposes think of Fodor as identifying a concept with a pair consisting of a MOP and some property ‘out there.’
So far as I can tell, Chomsky's view of concepts – one that is nativist like Fodor's but, unlike Fodor's, internalist – differs from Fodor's in three central ways. Unlike Fodor, he seriously doubts
that denotations serve to individuate concepts. For reasons that were well explored by rationalists in the seventeenth and eighteenth centuries, he holds that the only fruitful way to distinguish one concept from another is to look to the concepts in the head themselves and construct a naturalistic theory of them. That is, one should aim to construct a theory of something like Fodor's modes of presentation or MOPs, not of relations to what he claims are their denotations. For MOPs – as even Fodor grants1 – have the natures they do not because of what causes them, but because the mind configures them in ways that suit its own machinery's agenda. If you want to know what a concept is, look at what the mind makes it be; nativism and its implicit internalism are fine, representationalism and externalism are not. Chomsky's view is tied to the idea, often presented and defended by him, that referring or denoting is something that people do – a form of (free) human action – not some kind of ‘natural’ relationship, as Fodor wants to believe. It is also tied to the view that human cognitive resources are limited, and that the only readily available ‘access to the world’ that all humans have available to them is provided by the biophysically based cognitive resources offered by innate concepts – in effect, by the commonsense concepts that children and others readily acquire, without training. No doubt at least some humans now do have other sets of concepts available to them, of course – those found in the various natural sciences, which are to a large extent human inventions, and by no means readily available to the child or even the adult unfamiliar with the theories in which the science's concepts are defined and configured. Pace Fodor (and Putnam, Kripke, and many others) these concepts and the access to the world that they afford reinforce the point that it does little good to try to individuate native commonsense concepts by appeal to denotations. H2O, and even more tellingly the various structurally different states of it investigated by the sciences (see Chomsky on H2O in Appendix I), is not at all what people are talking about when they speak of water, using the commonsense concept WATER. The problem lies in Fodor's externalist-representationalist hopes.2
Another major way in which Chomsky differs from Fodor is in his view of the need for introducing what Fodor calls a “language of thought,” or LOT (Chomsky 2000). Chomsky seems to have several reasons for doubting the value of postulating such a thing/system. One has to do with the complications that a LOT adds to the science of natural language and, presumably, other systems that contribute to concepts. If syntactically described elements at the language faculty's semantic interface must be linked to the ‘right’ concept(s) in a separate system – that is, the LOT – in order to say what a specific set of syntactic elements ‘means,’ the theorist must (1) say what this link is, (2) how it is established (how it is acquired/learned) and (3) what is ‘in’ the LOT to link to. If instead the ‘semantic contribution’ of the language faculty at SEM – expressed in theoretical terms in Chomsky's view by stating which semantic features appear there – is taken (as it is by Chomsky) to constitute what an expression ‘means,’ period, there is no linking, acquisition, or specification problem. In effect, a particular SEM becomes a particular complex sentential “mode of presentation,” or at least, the language faculty's contribution to such a MOP (with the rest of the MOP provided by the mental systems with which language interfaces, if any).3 Further, given that the relevant ‘semantic information’ in the form of semantic features must be lodged in some way in the lexicon, and assuming that these features are taken to constitute (at least a major part of) the ‘semantic information’ offered in a commonsense concept, the issue of how one acquires the capacity to express the semantic resources contained in commonsense concepts of the sort that Fodor focuses on can be understood as the issue of what ‘semantic information’ can be placed in a lexical item, where an association between a linguistic ‘sound’ and ‘meaning’ (a set of semantic features) is assumed. And further still, assuming that one can state what the relevant features are, one also has a way of investigating, at least in principle, how human concepts differ from animals’ concepts, if they do; Fodor simply assumes that they are the same, for he assumes that we share the LOT with other creatures. One can do all this while maintaining the internalist principles that Chomsky holds, without any LOT – assuming there is such a thing, for all we have really been told is that the LOT is English (or French, etc.), and English, French . . . are themselves highly suspect entities, since their provenance is the commonsense notion of language, nothing at all like I-languages. Given all these advantages (economy, etc), for purposes of avoiding what appears to be confusion and the pursuit of dead ends, it is unfortunate that the science of semantic features is still in its early stages and so cannot yet be seen as the only plausible approach for the scientist of language to pursue. Nevertheless, there is some progress (noted below), and to anyone who has reason to believe on independent grounds that the internalist approach – as realized in some form – is the only plausible one, there is no alternative.
A third difference is that by assuming that ‘lexical’ concepts (those concepts expressed by words, not sentences) can be characterized by multiple semantic features, Chomsky allows for the idea that what Fodor calls “lexical” concepts are ‘analyzable,’ or ‘compositional’ rather than (as Fodor insists) ‘atomic.’ In essence, Chomsky allows that natural-language-expressed MOPs can be the targets of a compositional theory of concepts, an internalist one that develops the seventeenth- and eighteenth-century rationalists’ view that if you want to know what a concept is, look in the mind. Comments in the main text reflect an inclination to adopt concept complexity and compositionality. There are various ways of massaging both Fodor's atomic view of concepts and a Chomskyan (potentially) compositional one, but they can for our purposes be ignored. In any case, since semantic features figure heavily in the reconstruction above of Chomsky's view of concepts, I will assume that attempting to construct a theory of them and how they are ‘put together’ in accord with biophysical principles is a reasonable way to proceed in developing a naturalistic nativist and internalist theory of concepts.
While a science of concepts/MOPs expressed in terms of semantic features is in its early stages, it does appear to be a reasonable project. And there has been some progress. Features such as CONCRETE, ANIMATE, and the like have long been the focus of lexical concept research. They act as descriptive terms, efforts to capture differences in ‘readings’ of words and sentences. A descriptively adequate theory of semantic features must provide a way to distinguish differences in linguistically expressed concepts, here understood as MOPs.
Here is one way to conceive of terms such as CONCRETE and ANIMATE. Because they contribute to what Chomsky calls a “perspective” available to “other systems” at the semantic interface SEM, one can think of them in any given case as a contribution to one of a potentially infinite number of “ways of understanding” – understanding where language contributes. Sentences (“expressions”) express these ways; that is, sentences in the technical sense (“expressions”) offer in structured form at SEMs the semantic features of the lexical items of which they are composed. One can think of semantic feature terms as something like adverbial descriptions of how a person can think or conceive of ‘the world’ (including presumably a fictional or discourse or story or abstract world, however minimal) as presented by other systems in the head (cf. Chomsky 1995a: 20). The precise way in which semantic features do this is by no means clear; that is for a theory to decide, although I make some suggestions in Appendix XII. And there is danger inherent in saying that these features offer ways in which persons can understand, for persons do not figure in natural sciences; the way(s) the features ‘work’ at a semantic interface by providing ‘information’ to other systems is presumably unconscious; and “understand” is by no means a well-defined theoretical term. For the moment, though, it suffices: the relevant features surely have something to do with how people comprehend and think – with “understand” taken here to be a general term for all such cases.
Edging a bit into the realm of theory, then, perhaps we can think of the ways lexical items' (LIs’) semantic features contribute as having something to do with the way in which they configure other systems – or perhaps provide instructions to them, offering the semantic information the features constitute.
A caveat: it is a mistake to think of these features as properties of things ‘out there,’ rather in the way that Fodor (2008) speaking of features does. They might appear to have that role in the case of a sentence that a person uses to refer to something, at least where this sentence is held-true of the thing(s) to which that person refers. But referring and holding-true are both acts that a person performs, not by any means something that semantic features ‘do.’ Further, while sentences used to refer and hold-true may have a prominent place in the thoughts of those who would like to maintain that this use is both dominant and paradigmatic, it is neither. And emphasis upon truth-telling distracts attention from the far more prevalent uses of language in thought and imagination, speculation and self-berating, etc. and the primary point that ‘telling the truth’ is at best one of many ways in which a semantic feature can contribute to understanding, and it distracts attention from the fact that where semantic features do contribute in any of these ways or others, it is constitutive of a way of understanding, and thereby possibly of ‘experience’ (cf. Chomsky 1966/2002/2009).
