Chomsky 1992 (part 2)

Chomsky now addresses the second claim he attributes to Putnam: ‘studying the brain is not going to be helpful in understanding what people do.’ I think it is best to ignore what Putnam actually said, and to focus instead on the positive claims being made by Chomsky: “here’s how studying the brain might be helpful in understanding what people do.”

Take Putnam’s case: the discovery that thinking of cats evokes C. Surely such a discovery might have some relevance to inquiry into what Peter means (or refers to, or thinks about) when he uses the term cat, hence to “a discussion of the meaning of cat.” For example, there has been a debate – in which Putnam has taken part – about the referential properties of cat if cats were found to be robots controlled from Mars. Suppose that after Peter comes to believe this, his brain does, or does not, form C when he refers to cats (thinks about them, etc.). That might be relevant to the debate. Or, take a realistic case: recent studies of electrical activity of the brain (event-related potentials, ERPs) show distinctive responses to nondeviant and deviant expressions and, among the latter, to violations of:

  1. word meaning expectancies;
  2. phrase-structure rules;
  3. the specificity-of-reference condition on extraction of operators; and
  4. locality conditions on movement (Neville et al. 1991).

Such results surely might be relevant to the study of the use of language, in particular, the study of meaning.

Just as we might want to say of Nils Bohr that his use of the word ‘electron’ pre-quantum mechanics referred to the same stuff that his use of the same word post-quantum mechanics referred to, it seems intuitive to want to say that our use of the word ‘cat’ should refer to the same fuzzy little tailed guys wandering around our world even after we discover that they are robots sent from Mars. Putnam observes that this intuition cannot be maintained if we think that everything relevant to determining the reference of our terms is ‘in our heads.’ Instead, he argues, the meanings of our terms can only be determined by “what’s in our heads” in conjunction with facts about the world. Therefore, our mental state, and presumably therefore also our brain state, does not suffice to determine what we’re actually talking about. Chomsky says: well, what would we say if we observed that the state our brain is in when we use the word ‘cat’ changes once we learn they are robots from Mars? (He says this ‘might be relevant to the debate’, meaning: it is really not clear what we would say then.) Then he descends into ‘real-life’ examples, noting that our brains respond differently to the above four (five, if you count well-formed things) categories of stimuli. It is not obvious at all that these will be relevant to studying the meanings of our terms (but I guess they ‘surely might be’).

We can proceed further. Patterns of electrical activity of the brain correlate with the five categories of structure noted: nondeviance, and four types of deviance. But the study of these categories is also a study of the brain, its states and properties, just as study of algorithms involved in seeing a straight line or in doing long division is a study of the brain. Like other complex systems, the brain can be studied at various levels: atoms, cells, cell assemblies, neural networks, computational–representational (C–R) systems, etc. The ERP study relates two such levels: electrical activity of the brain and C–R systems. The study of each level is naturalistic both in the character of the work and in that integration with the core natural sciences is a prospect that can be reasonably entertained. In the context of Putnam’s discussion, discoveries about the brain at these levels of inquiry are on a par with a discovery about the (imagined) configuration C, when Peter thinks of cats.

Linguistics and cognitive science more generally investigate aspects of what Chomsky is here calling the ‘computational-representational’ system of the brain. We could study lots of other aspects of our brain, such as what kinds of cells make it up and how these cells are structured, such as how these cells are connected to one another, such as how the action of a particular cell influences other cells elsewhere in the brain, etc. Neurolinguistic studies attempt to relate patterns at one level (a lower level, i.e. closer to physics) to the computational-representational system; we ask: what is going on at this lower level, when this is happening at the higher level?

In the case of language, the C–R theories have much stronger empirical support than anything available at other levels, and are far superior in explanatory power; they fall within the natural sciences to an extent that inquiry into “language speaking” at the other levels does not. In fact, the current significance of the ERP studies lies primarily in their correlations with the much richer and better-grounded C–R theories. Within the latter, the five categories have a place and, accordingly, a wide range of indirect empirical support; in isolation from C–R theories, the ERP observations are just curiosities, lacking a theoretical matrix. Similarly, the discovery that C correlates with use of cat would, as an isolated fact, be more of a discovery about C than about the meaning of cat – and for that reason alone would shed little light on the controversy about robots controlled from Mars. To take another case, the discovery of perceptual displacement of clicks to phrase boundaries is, for now, more of a discovery about the validity of the experiment than about phrase boundaries. The reason is that evidence of other sorts about phrase boundaries – sometimes called “linguistic” rather than “psychological” evidence (a highly misleading terminology) – is considerably more compelling and embedded in a much richer explanatory structure. If click experiments were found to be sufficiently reliable in identifying the entities postulated in C–R theories, and if their theoretical framework were deepened, one might rely on them in cases where “linguistic evidence” is indecisive; possibly more, as inquiry progresses. (On some misunderstandings of these matters see Chapter 3 of this volume; Chomsky 1991a; 1991b).

This is something that Chomsky says that often engenders pushback: that our theories of the higher level organization of the brain are currently much more developed than our theories of its lower level organization, and that therefore when we can link something at a higher level with something at a lower level that should not be thought of as evidence for the higher level concept, but rather for the fact that the experiment worked. As he says, since we understand the nitty-gritty of the brain (i.e. its biology) so poorly, these experiments tell us that these brain states are importantly different from one another, not that these high level categories are valid. This is an important argument, and I can imagine that we might want to discuss it a bit more in the comments.

Here Chomsky expands a bit more on his argument.

For the present, the best-grounded naturalistic theories of language and its use are C–R theories. We assume, essentially on faith, that there is some kind of description in terms of atoms and molecules, though without expecting operative principles and structures of language and thought to be discernible at these levels. With a larger leap of faith, we tend to assume that there is an account in neurological terms (rather than, say, glial or vascular terms, though a look at the brain reveals glial cells and blood as well as neurons. It may well be that the relevant elements and principles of brain structure have yet to be discovered. Perhaps C–R theories will provide guidelines for the search for such mechanisms, much as nineteenth-century chemistry provided crucial empirical conditions for radical revision of fundamental physics. The common slogan that “the mental is the neurophysiological at a higher level” – where C–R theories are placed within “the mental” – has matters backwards. It should be rephrased as the speculation that the neurophysiological may turn out to be “the mental at a lower level” – that is, the speculation that neurophysiology might, some day, prove to have some bearing on the “mental phenomena” dealt with in C–R theories. As for the further claims of eliminative materialism, the doctrine remains a mystery until some account is given of the nature of “the material”; and given that account, some reason why one should take it seriously or care if successful theories lie beyond its stipulated bounds.

Thus we have a good theory about how the high-level stuff works, a poor theory about how the low-level stuff works, and virtually no idea about how they relate. Indeed, we sort of just assume that they do relate, and thus that the mind is ‘just’ the brain. An example Chomsky is fond of bringing up is Newton’s mechanics, which we think of as a vindication of physicalism, but which (Chomsky says) really was a surrender to the occult: gravity is action-at-a-distance, unmediated by physical interactions. We currently are materialists, thinking that the brain underlies the mental, but Chomsky points out that we could be in the situation of physics immediately pre-Newton, where everyone was expecting that physicalism would win, but ultimately a form of non-physicalism carried the day.

Now we can discuss language a bit.

For the present, C–R approaches provide the best-grounded and richest naturalistic account of basic aspects of language use. Within these theories, there is a fundamental concept that bears resemblance to the common-sense notion “language”: the generative procedure that forms structural descriptions (SDs), each a complex of phonetic, semantic, and structural properties. Call this procedure an I-language, a term chosen to indicate that this conception of language is internal, individual, and intensional (so that distinct I-languages might, in principle, generate the same set of SDs, though the highly restrictive innate properties of the language faculty may well leave this possibility unrealized). We may take the linguistic expressions of a given I-language to be the SDs generated by it. A linguistic expression, then, is a complex of phonetic, semantic, and other properties. To have an I-language is something like having a “way to speak and understand,” which is one traditional picture of what a language is. There is reason to believe that the I-languages (“grammatical competence”) are distinct from conceptual organization and “pragmatic competence,” and that these systems can be selectively impaired and developmentally dissociated (see Yamada 1990; John Marshall 1990).

Language use is a rich and complicated thing, and we don’t know what kind of theory we will end up with, and how it will ‘cut up’ the pre-theoretical ‘pie.’ Our linguistic theory describes something related to our pre-theoretical conception of language, namely, grammar. (This is true of every linguistic theory, not just minimalism/transformational grammar.) The term I-language is used as a name for a particular grammar, as had by a particular person.1

The I-language specifies the form and meaning of such lexical elements as desk, work, and fall, insofar as these are determined by the language faculty itself. Similarly, it should account for properties of more complex expressions: for example, the fact that “John rudely departed” may mean either that he departed in a rude manner or that it was rude of him to depart, and that, in either case, he departed (perhaps an event semantics should be postulated as a level of representation to deal with such facts; see Higginbotham 1985; 1989). And it should explain the fact that the understood subject of expect in example (1) depends on whether X is null or is Bill, with a variety of other semantic consequences: (1) John is too clever to expect anyone to talk to X. And for the fact that, in my speech, ladder rhymes with matter but madder doesn’t. In a wide range of such cases, nontrivial accounts are forthcoming. The study of C–R systems provides no little insight into how people articulate their thoughts and interpret what they hear, though of course it is as little – and as much – a study of these actions as the physiology and psychology of vision are studies of humans seeing objects.

The job of linguistic theory (the role of I-language) is to specify which forms can have which meanings, which entailments they support, which formal relations the sounds have, etc. It is not supposed to tell us how people choose what to say, or how they understand what is said. Why is it supposed to do these things, and not those? Because the theory seems to work for the former but not the latter! This is a case where our pre-theoretic term ‘language use’ has been revealed by scientific inquiry to actually be comprised of multiple ‘real’ things, one of them being I-language (grammar). If we develop our scientific understanding further, we might decide that this picture is wrong and that (for example) rhyming is explained by something completely different.

A deeper inquiry into I-languages will seek to account for the fact that Peter has the I-language \(L_P\) while Juan has the I-language \(L_J\) – these statements being high-level abstractions, because in reality what Peter and Juan have in their heads is about as interesting for naturalistic inquiry as the course of a feather on a windy day. The basic explanation must lie in the properties of the language faculty of the brain. To a good approximation, the genetically-determined initial state of the language faculty is the same for Peter, Juan, and other humans. It permits only a restricted variety of I-languages to develop under the triggering and shaping effect of experience. In the light of current understanding, it is not implausible to speculate that the initial state determines the computational system of language uniquely, along with a highly structured range of lexical possibilities and some options among “grammatical elements” that lack substantive content. Beyond these possibilities, variation of I-languages may reduce to Saussurean arbitrariness (an association of concepts with abstract representations of sound) and parts of the sound system, relatively accessible and, hence, “learnable” (to use a term with misleading connotations). Small differences in an intricate system may, of course, yield large phenomenal differences, but a rational Martian scientist studying humans might not find the difference between English and Navajo very impressive.

Why is it scientifically uninteristing “what Peter and Juan have in their heads?” The answer is that the language we each end up learning (sorry: the I-language we each end up learning) is the result of a whole bunch of random events (like the exact sentences each of them heard, who their role models were, etc). Likewise, while aerodynamics is scientifically interesting (i.e., we can build a general theory of it), no one is interested in building a theory of (or accounting for) the exact motion of a particular feather from 14:00 to 15:00 on June 21, 1999. That feather’s particular motion was influenced by so many interacting things, that it’s hopeless to either try to account for it, or to try to develop a general theory of it. So instead, we are interested not in what I-languages Peter and John actually ended up with, but rather in what enabled them to develop the I-languages they did. As we informally observe that kids seem to be able to learn whatever language (in the naive sense) they are exposed to, irrespective of race, gender, eye or hair color, handedness, IQ, etc, it seems reasonable to start by assuming that we have a common endowment to learn language.2 What is our common endowment, really? This is what we’re trying to figure out. Chomsky says that, currently, it seems that a whole lot of what makes I-language seems to be common across I-languages, and that therefore our common endowment is almost everything about I-language, with just a few ‘holes.’ Backing off a bit from the bleeding edge of then current linguistic theory, we observe that, contrary to early 20th century thought, languages do not vary arbitrarily from one another, and thus our common endowment must somehow restrict our learning options to that which we actually end up seeing. How could this be false? First, we could be wrong about how similar I-languages are to one another; maybe they do vary arbitrarily! Second, we could be wrong about how similar children’s linguistically relevant input is. Maybe we are right that I-languages are in fact very similar, but this is because everyone’s input is so unexpectedly homogeneous.

The I-language is a (narrowly described) property of the brain, a relatively stable element of transitory states of the language faculty. Each linguistic expression (SD) generated by the I-language includes instructions for performance systems in which the I-language is embedded. It is only by virtue of its integration into such performance systems that this brain state qualifies as a language. Some other organism might, in principle, have the same I-language (brain state) as Peter, but embedded in performance systems that use it for locomotion. We are studying a real object, the language faculty of the brain, which has assumed the form of a full I-language and is integrated into performance systems that play a role in articulation, interpretation, expression of beliefs and desires, referring, telling stories, and so on. For such reasons, the topic is the study of human language.

This somewhat weird sounding passage3 expresses in a condensed form the properties of I-language. It is a property of the brain, because it is a matter of individual psychology (and we assume that the brain is where this stuff happens, not the spleen). Because our use (i.e. speaking and understanding) of language conforms to what our I-language says, I-language must have some relation to other systems. If I-language had no connection to other language systems, we wouldn’t call it language. If the very same brain state which underlies Juan’s I-language were had by octopi but connected to their locomotion system, we would be fascinated, amazed, shocked, but we wouldn’t say that an octopus had language.

The performance systems appear to fall into two general types: articulatory–perceptual, and conceptual–intentional. If so, it is reasonable to suppose that a generated expression includes two interface levels, one providing information and instructions for the articulatory–perceptual systems, the other for the conceptual–intentional systems. One interface is generally assumed to be phonetic representation (Phonetic Form, PF). The nature of the other is more controversial; call it LF (“Logical Form”).

What do we do with our expressions, well, we pronounce and understand them. We understand pronunciation pretty well, partly because we can actually see (or hear) the results, but understanding is unfortunately invisible to the eye, and visible only through deed.

The properties of these systems, or their existence, are matters of empirical fact. One should not be misled by unintended connotations of such terms as “logical form” and “representation,” drawn from technical usage in different kinds of inquiry. Similarly, though there is a hint of the notions “deep grammar” and “surface grammar” of philosophical analysis, the concepts do not closely match. What is “surface” from the point of view of I-language is, if anything, PF, the interface with articulatory–perceptual systems. Everything else is “deep.” The surface grammar of philosophical analysis has no particular status in the empirical study of language; it is something like phenomenal judgment, mediated by schooling, traditional authorities and conventions, cultural artifacts, and so on. Similar questions arise with regard to what is termed, much too casually, “folk psychology,” as noted. One should regard such notions with caution: much may be concealed behind apparent phenomenal clarity.

These systems must exist, if our current scientific study of I-language is on the right track, and if they do, they have whatever properties they have, irrespective of whether anyone ever studies them. We choose suggestive terms for them, but if they actually have the properties these terms suggest is not known (and is sometimes thought not to be the case).

The complex of I-language and performance systems enters into human action. It is an appropriate subject matter for naturalistic theories, which might carry us far towards understanding how and why people do what they do, though always falling short of a full account, just as a naturalistic theory of the body would fail to capture fully such human actions or achievements as seeing a tree or taking a walk.

What we actually do is partly explained by the theories of I-language and performance systems. We can investigate these scientifically, but we will never have a theory that accounts for every detail of what really happens (because what really happens is influenced not only by I-language and performance systems, but also our wants, desires, perceptions, etc).

Correspondingly, it would be misleading, or worse, to say that some part of the brain or an abstract model of it (for example, a neural net or programmed computer) sees a tree or figures out square roots. People in an ambiguous range of standard circumstances pronounce words, refer to cats, speak their thoughts, understand what others say, play chess, or whatever; their brains don’t and computer programs don’t – though study of brains, possibly with abstract modelling of some of their properties, might well provide insight into what people are doing in such cases. An algorithm constructed in a C–R theory might provide a correct account of what is happening in the brain when Peter sees a straight line or does long division or “understands Chinese,” and might be fully integrated into a well-grounded theory at some other level of explanation (say, cells). But the algorithm, or a machine implementing it, would not be carrying out these actions, though we might decide to modify existing usage, as when we say that airplanes fly and submarines set sail (but do not swim). Nothing of substance is at stake. Similarly, while it may be that people carry out the action by virtue of the fact that their brains implement the algorithm, the same people would not be carrying out the action if they were mechanically implementing the instructions, in the manner of a machine (or of their brains). It may be that I see a straight line (do long division, understand English, etc.) by virtue of the fact that my brain implements a certain algorithm; but if I, the person, carry out the instructions mechanically, mapping some symbolic representation of the input to a representation of the output, neither I nor I-plus-algorithm-plus-external memory sees a straight line (etc.), again, for uninteresting reasons.

I find this a very Wittgenstein-ian passage. Chomsky is simply pointing out that the terms ‘pronounce words’, ‘see cats’, ‘play chess’, ‘understand what others say’, etc are part of our non-scientific discourse and we use them to talk about people (and maybe in some cases animals). If we want to ask whether a robot ‘understands what others say’, we are asking how we should extend our informal use of language to a case where it does not normally apply. This is not a matter of science, or philosophy, just a cultural negotiation. To make it scientific, or philosophical, we need to do a scientific (or philosophical) investigation of ‘understanding what others say’ (or ‘pronounce words’ or ‘see cats’). The theory we end up with will, for reasons Chomsky has discussed previously, end up drawing different distinctions than our original, naive, term used. We can then ask of this new, perhaps related, scientific term, whether a robot can do that.

It would also be a mistake, in considering the nature of performance systems, to move at once to a vacuous “study of everything.” As a case in point, consider Donald Davidson’s discussion of Peter as an “interpreter,” trying to figure out what Tom has in mind when he speaks. Davidson observes that Peter may well use any information, background assumption, guesswork, or whatever, constructing a “passing theory” for the occasion. Consideration of an “interpreter” thus carries us to full models of human functional organization. Davidson concludes that there is no use for “the concept of a language” serving as a “portable interpreting machine set to grind out the meaning of an arbitrary utterance”; we are led to “abandon . . . not only the ordinary notion of a language, but we have erased the boundary between knowing a language and knowing our way around in the world generally.” Since “there are no rules for arriving at passing theories,” we “must give up the idea of a clearly defined shared structure which language-users acquire and then apply to cases” (Davidson 1986b: 446). “There is no such thing as a language,” a recent study of Davidson’s philosophy opens, with his approval (Davidson 1986b; Ramberg 1989).

Chomsky is here arguing that we need to study parts of our behaviour which lend themselves to study in isolation (those which Marr would say have type 1 theories). Just because when we translate, we use everything at our disposal, doesn’t mean we have to take ‘translation’ as a monolithic object of study (in fact, we shouldn’t, because it is hopless). Instead, we should try to identify individual aspects of the translation process which lend themselves to further study. (Johannes would call this ‘obvious,’ but clearly it is not obvious to everybody!)

Here are his words:

The initial observation about “passing theories” is correct, but the conclusions do not follow. A reasonable response to the observation – if our goal is to understand what humans are and what they do – is to try to isolate coherent systems that are amenable to naturalistic inquiry and that interact to yield some aspects of the full complexity. If we follow this course, we are led to the conjecture that there is a generative procedure that “grinds out” linguistic expressions with their interface properties, and performance systems that access these instructions and are used for interpreting and expressing one’s thoughts.

This is, I think, just the scientific process in general, applied to language; we study frictionless planes, because it turns out that friction is (or can be viewed as) an independent factor that can be abstracted away from.

What about “the idea of a clearly defined shared structure which language-users acquire and then apply to cases”? Must we also postulate such “shared structures,” in addition to I-language and performance systems? It is often argued that such notions as common “public language” or “public meanings” are required to explain the possibility of communication or of “a common treasure of thoughts,” in Gottlob Frege’s sense (Frege 1892/1965: 71). Thus, if Peter and Mary do not have a “shared language,” with “shared meanings” and “shared reference,” then how can Peter understand what Mary says? (Interestingly, no one draws the analogous conclusion about “public pronunciation.") One recent study holds that linguists can adopt an I-language perspective only “at the cost of denying that the basic function of natural languages is to mediate communication between its speakers,” including the problem of “communication between time slices of an idiolect” (so-called “incremental learning”; Fodor and Lepore 1992).

So I-language is a completely individual property, and thus no two people will have the exact same I-language. But we talk about ‘language’ and say that you and I speak English - what do? Less memetically, people succeed in communication with each other, how could this happen unless there were some reality of a common language? The basic issue here is that Chomsky is claiming that linguistics is a branch of individual psychology, but that we use terms in what seems to us an explanatory way that claim that there is some public aspect of language. How can we reconcile this? It seems that either Chomsky is wrong, or humanity is. (Spoiler alert: it’s not Chomsky.)

But these views are not well founded. Successful communication between Peter and Mary does not entail the existence of shared meanings or shared pronunciations in a public language (or a common treasure of thoughts or articulations of them), any more than physical resemblance between Peter and Mary entails the existence of a public form that they share. As for the idea that “the basic function of natural languages is to mediate communication,” it is unclear what sense can be given to an absolute notion of “basic function” for any biological system; and if this problem can be overcome, we may ask why “communication” is the “basic function.” Furthermore, the transition problem seems no more mysterious than the problem of how Peter can be the person he is, given the stages through which he has passed. Not only is the I-language perspective appropriate to the problems at hand, but it is not easy to imagine a coherent alternative.

We are first given an analogy:

that Peter and Mary look alike is not explained by saying that physical looks are not an aspect of the individual

This seems right, and so we should at least think about this being true of language as well. (This will be done in the next passage.)

There was another point raised in the previous paragraph, which I glossed over, which is the idea that language is ‘for’ communication. This is also very intuitive. Chomsky claims that it is not 100% clear what it means to claim that a particular biological system has a certain basic function. Certainly, the heart does in fact help circulate blood through the body, and if a particular heart doesn’t do this, the organism it is in quickly dies. But is the heart for circulating blood, or is it just for pumping, and the blood just happens to be there, or …?

At any rate, here is the proposed theory sketch of how I-language suffices to build the scaffolding of a theory of successful communication, without any ‘public’ language.

It may be that when he listens to Mary speak, Peter proceeds by assuming that she is identical to him, modulo M, some array of modifications that he must work out. Sometimes the task is easy, sometimes hard, sometimes hopeless. To work out M, Peter will use any artifice available to him, though much of the process is doubtless automatic and unreflective. Having settled on M, Peter will, similarly, use any artifice to construct a “passing theory” – even if M is null. Insofar as Peter succeeds in these tasks, he understands what Mary says as being what he means by his comparable expression. The only (virtually) “shared structure” among humans generally is the initial state of the language faculty. Beyond that we expect to find no more than approximations, as in the case of other natural objects that grow and develop.

We assume that people we talk to are just like us. As he quickly notices that assuming that he is talking with himself doesn’t work, he rapidly assembles a theory of how she differs from him. In other words, communication is translation; we translate other’s speech (resp. actions) into our I-language (resp. belief and desire system). We don’t know what their language is, so we start out by assuming that they speak our language, and then little by little modifying this assumption as we interact with them.

Some more explaining of the general point:

For familiar reasons, nothing in this suggests that there is any problem in informal usage, any more than in the ordinary use of such expressions as Boston is near NewYork or John is almost home. It is just that we do not expect such notions to enter into explanatory theoretical discourse. They may be appropriate for informal discussion of what people do, with tacit assumptions of the kind that underlie ordinary discourse in particular circumstances; or even for technical discourse, where the relevant qualifications are tacitly understood.They have no further place in naturalistic inquiry, or in any attempt to sharpen understanding.

We do use these terms, it’s fine to use these terms, but these terms do not seem to have identified anything ‘real’ in the world, that we could study.

Alleged social factors in language use often have a natural individualist–internalist interpretation. If Peter is improving his Italian or Gianni is learning his, they are (in quite different ways) becoming more like a wide range of people; both the modes of approximation and selection of models vary with our interests. We gain no insight into what they are doing by supposing that there is a fixed entity that they are approaching, even if some sense can be made of this mysterious notion. If Bert complains of arthritis in his ankle and thigh, and is told by a doctor that he is wrong about both, but in different ways, he may (or may not) choose to modify his usage to that of the doctor’s. Apart from further detail, which may vary widely with changing contingencies and concerns, nothing seems missing from this account. Similarly, ordinary talk of whether a person has mastered a concept requires no notion of common language. To say that Bert has not mastered the concept arthritis or flu is simply to say that his usage is not exactly that of people we rely on to cure us – a normal situation. If my neighbor Bert tells me about his arthritis, my initial posit is that he is identical to me in this usage. I will introduce modifications to interpret him as circumstances require; reference to a presumed “public language” with an “actual content” for arthritis sheds no further light on what is happening between us, even if some clear sense can be given to the tacitly assumed notions. If I know nothing about elms and beeches beyond the fact that they are large deciduous trees, nothing beyond this information might be represented in my mental lexicon (possibly not even that, as noted earlier); the understood difference in referential properties may be a consequence of a condition holding of the lexicon generally: lack of indication of a semantic relation is taken to indicate that it does not hold.

Even when it seems as though our use of concepts like ‘a common language’ are playing an explanatory role, we can often reformulate them in terms of I-languages. Chomsky gives the example of Peter improving his Italian. On the face of it, this only makes sense if there is a common language, Italian, and Peter’s grammar is getting more and more like the grammar of this common language. Chomsky suggests reformulating this in purely individualistic terms by saying that Peter is becoming more like (each of) a group of people. The talk of arthiritis and elms and beeches refer to philosophical thought experiments which aim at establishing the proposition that the reference of our terms is not just partially fixed by facts about the world, but also by other people. In the beech vs elm example, we are asked to imagine someone (like myself) who has no idea what the difference between beeches and elms is. We might think that my concepts of Elms and of Beeches must be the same (‘non-evergreen tree’). Yet when I use the word ‘beech’, it seems not unreasonable to say that I am talking about beeches, and most importantly, it does seem unreasonable to deny that I have mastered English. The argument is supposed to establish that for many of our concepts, we don’t actually know what we’re talking about, and yet we succeed in talking about them because there are experts who do know this, and their knowledge determines the reference of our words. At any rate, Chomsky suggests that we needn’t conclude that my concepts of elms and beeches are identical; just because I can’t tell you what the differences are, doesn’t mean that the concepts need be the same. The fact that my words ‘elm’ and ‘beech’ seem to refer to different things might be explainable under the assumption that our minds assume that distinct concepts refer to distinct things.

  1. Again, Chomsky is using I-language to mean not just syntax, but the whole shebang: phonology, morphology, syntax, semantics, and how they fit together. ↩︎

  2. This is presumably genetic, because we’re materialists at heart, and don’t know how else we could have a common endowment were it not in the genes. If Chomsky were a geneticist rather than a linguist he would be here arguing that we have a better theory of what our common endowment is, than how it is encoded in the genes, and that we should not blithely assume that our current reductive expectations that all common endowments ultimate spring from genes be vindicated after all is said and done. But he’s not, and so we will. ↩︎

  3. It sounds like a Glaubensbekenntnis! ↩︎