8 The involvement of language in conscious thinking
In the last chapter I presented, and defended, a non-species-specific account of the distinction between conscious and non-conscious mental states. I concluded that RT theory at least provides the best available account of the structure of human consciousness, and that it successfully describes the sort of cognitive architecture that any organism, of whatever variety, would have to instantiate if it is to be capable of conscious episodic thinking. In the present chapter I shall argue that there is a natural necessity attaching to the fact that such conscious thinking in the case of human beings, at any rate often involves natural language. I shall distinguish weaker and stronger versions of this natural necessity thesis, and argue tentatively in support of each.
8.1 An architecture for human thinking
My first task is to sketch the outline of an architecture for human cognition (or rather, of that fragment of human cognition which specifically concerns us). This outline-architecture is designed to do two things. First, it should instantiate the RT model of consciousness, already explained and defended in Chapter 7. Second, it should explain the introspective datum of Chapter 2, that much of human conscious thinking appears to take place in natural language. Not only should it explain that datum, indeed, but it should provide an explanation which accords with our common-sense construal of it namely, that much of our conscious thinking does take place in natural language.
No one should doubt that the stream of inner verbalisation exists. Certainly, nothing in the discussion of theories of consciousness over the last three chapters has lent any support to the extreme suggestion (briefly floated in section 2.5) that the very existence of inner speech is an illusion. What remains a legitimate matter of dispute, however, is the causal, or functional, role of inner speech. My explanation is designed to accord it the role of thinking. So I shall claim that we are aware, not only of what we have just thought, and of the fact that we have just thought it (as RT theory maintains), but also of the way in which that thought was entertained we are also aware of its more or less precise formulation in natural language, or in a combination of language and other images.
The components of the model. There are, in fact, five distinct components to be put into place, before I can develop my model of human cognitive architecture. Two of these have already been mentioned. The first is the introspective datum from Chapter 2, that much of human conscious thinking at least appears to take place in natural language. This is the datum which is to be explained by some suitable arrangement of the other four components. The second component is the RT theory of consciousness, which requires thinkers to have non-inferential access to their own acts of thinking, as such, if those thinkings are to be conscious ones. If I entertain a thought consciously, then I have to be capable of thinking about that thought, and the fact that I have just entertained it, without engaging in self-interpretation.
The third component is a theory of mind or common-sense psychology, which is probably innate to our species, and which is presupposed by the activity of reflexive thinking. In order to be capable of thinking about my own acts of thinking, or of thinking about my own perceptions and feelings, I have to possess the requisite concepts. That is, I must have the general concepts thought and perception, as well as their more specific variants believe, want, judge, wonder whether, see, feel, hear, and so on. As was argued briefly in sections 1.6 and 1.7, these concepts get their life and significance from being embedded in a set of beliefs about the structure and functioning of the mind common-sense psychology.
The fourth component is imagination, and will require some brief exposition before it can be put into place. Recall the thesis of modular mental organisation, outlined briefly in section 2.8 in the course of our discussion of Fodor's arguments for Mentalese. According to this thesis, the mind contains various input and output modules whose functions are, respectively, to process perceptual information before passing on the results to central cognition, and to take instructions from central cognition and transform them into detailed bodily movements. (See Fodor 1983. See also Shallice 1988.) The principles of operation of the various modules are, according to Fodor, mostly innately specified and extremely fast. They are also supposed to be isolated from central cognition, at least in the sense that they are impervious to changes in our beliefs (hence the robustness of perceptual illusions). But it is important to see that this isolation need not, and in fact does not, mean that central cognition can have no access to the structures within a module at all.
Consider mental imagery, in particular. It is now well-established that imagery, of a given sensory type (visual, auditory, kinaesthetic, and so on), employs some of the same cognitive resources as does perception in the same sense-modality. For, as we noted in section 2.8, cognitive tasks which require visual imagination (such as counting the right-hand corners in an imaged 'F') will be interfered with if the results have to be indicated visually (by pointing to a number, say), but not if they can be delivered orally. Similarly, tasks which require auditory imagination (such as counting the nouns in a previously heard sentence) will be interfered with if the results have to be delivered orally, but not if they can be given visually. (See Fodor, 1975, and Tye, 1991.) These results are robust.
The best explanation of these and similar results is that visual imagination employs some of the same cognitive resources as does vision, auditory imagination employs some of the same cognitive resources as does hearing, and so on. The hypothesis which naturally suggests itself is that central cognition partly operates by accessing, and by activating and manipulating, some of the higher levels of representation within the input and output modules. Some of the circuits within those modules can, as it were, be run autonomously, without stimulation from the environment. This has obvious advantages for the organism, in that it can then represent to itself states of affairs which are not yet actual, and either try to avoid them or try to bring them about.
The fifth and final component is language. This, too, is very probably a distinct mental module. As we noted in section 1.7, Chomsky has presented powerful and convincing arguments for the innateness of the language faculty (see his 1988; see also Curtiss, 1989, and Pinker, 1994). And the operation of that faculty, like the operation of other modules, is both mandatory and extremely fast. But language differs from many other input and output modules in at least two respects.
First, language always takes its input from other modules (at least, this is so in the sense that it takes its input from the sensory transducers of other modules; in fact it may be that language can draw on this sensory information before the stage at which it has been processed by the sensory module in question). When an in-coming sentence is processed and understood, it is always either heard (speech), seen (writing, sign-language), or felt (Braille). Other input modules, in contrast, have their own unique sensory transducers, and operate largely independently of one another. (Input modules do appear to interact, to some degree, at least at higher levels of processing. What you see is, at least sometimes, a function of what you hear or feel.)
Second, language is both an input and an output module. It not only processes, but also generates, speech (or Sign) and writing. Here, too, the language faculty may be contrasted with other output modules, in that its output always has to be routed through the operation of other motor-control modules. Roughly speaking, the language faculty can only achieve public output through movements of the larynx and hands.
The RT-based model. Let us now put all of the above components together, by means of an inference to the best explanation. What is the best explanation of the introspective datum that much of our conscious thinking appears to take place in natural language? When taken in the context of the RT theory of consciousness, the answer, I believe, is this: that human cognition comes to instantiate such a reflexive structure by accessing, and activating and manipulating, some of the higher levels of representation within our perceptual (and/or motor) faculties specifically, in the form of images of natural language sentences. It is then imagination, together with common-sense psychology and knowledge of language, which enables human thinking to be conscious.
The best explanation of the available introspective data is that we mostly think (when our thinking is conscious) by imaging sentences of natural language, and trains of thought consist of manipulations and sequences of such images. We have access to the forms of our thoughts because a record of each imaged sentence is briefly held in short-term memory, so that we can recall what we have just imaged. We can then think about what we have just thought because our language contains the conceptual resources to do so in particular, because we are masters of a common-sense psychology. Having imaged the sentence (and hence thought to myself), 'The world is getting warmer', I can then image the sentence (and so think), 'That thought may have been too hasty', and so on. Human thinking thus becomes conscious in virtue of our reflexive thinking faculty having the capacity to access knowledge of natural language, and then by deploying resources from perception so as to activate that knowledge in imagination.
Merely imagining a sentence is not yet to think the content of that sentence, of course, in at least two respects. First, it is possible to entertain an image of a sentence from a language one does not understand, without (of course) thereby thinking any content. If I have listened to a fair bit of French with only minimal understanding, for example, then it may be possible for me to entertain French sentences in inner speech, without knowing what they mean. What this shows is that the object which occupies the causal role of a thought, on the proposal being developed here, is not only a phonological representation of a sentence. Rather, in the normal case that representation will carry with it its syntactic structure and associated semantics. When I think, 'The world is getting warmer', by imagining that sentence, my image is immediately and non-inferentially imbued with content, just as if I had heard that sentence uttered aloud. Sentences in a language one understands are always heard under a particular interpretation, in such a way that we seem to hear the meaning of the utterance. The same is true of the images which constitute our thoughts the content of the thought is heard in its form.
Second, it is possible to imagine a sentence without taking it to be true. When I obey the instruction, 'Imagine someone saying, "Grass is purple", in a squeaky voice', by forming an image of the sentence in question, this is not the same as judging that someone has said that, let alone the same as believing, myself, that grass is purple. The difference in question is a difference in causal role. What constitutes an image of the sentence, 'Grass is purple', as an occurrent judgement that grass is purple is a battery of characteristic causes and effects judgements are characteristically made in response to questions, whether internally or externally generated; judgements tend both to be caused by, and to further cause, long-term storage of their associated contents; and so on. What it is about any given imaged sentence which is determinative of one or other of the available causal roles judgement, fantasy, 'idle thinking', and so on I do not know. But whatever it is, I doubt whether it is transparently accessible to consciousness. That is, I doubt whether we can always tell, merely by introspection without self-interpretation whether an imaged sentence is expressive of a judgement, or has some other cognitive role. (Thus I have to concede that even my RT-based theory must provide some limited scope for self-interpretation.)
We might, then, represent the architecture of human consciousness in the form of another box-diagram, as in Figure 3. Much of the structure here is just as it was before, the same as the general (non-species-specific) account of consciousness presented in Chapter 7. But now there is a triple arrow down to C, to represent that it is not just a record of what is thought, and of the fact that it has been thought, which is passed back down to C to be available to further thought, but also the way in which it was thought it not just the content and occurrence, but also the form, of our thoughts which is available to further thought. (Remember that the triple-arrow, here, does not represent three different channels of access. Rather, it represents that the thoughts in reflexive thinking are available to three different kinds of further thought thoughts about their content, thoughts about their occurrence, and thoughts about their form.)
****** Insert Figure 3 RT theory and language about here *****
A language box has also been added, which takes input from perception and passes on its results to C, to be made available to thought. (The language box will also, of course, take input from reflexive thinking and transform this into instructions for motor output, but I have left this out of the picture, for simplicity.) The dotted arrows leading from the language and percept boxes to reflexive thinking are there to represent that the latter operates by accessing structures inherent in the former two. (There should also be dotted arrows from the language box to non-conscious thinking, if such thinking, too, makes use of natural language representations, but non-imagistically, as I shall suggest in section 8.7 that it does.) So human conscious (and properly propositional) thinkings achieve their status as such by virtue of consisting of deployments of natural language sentences in imagination, which are then made available in short-term memory to be thought about in turn.
Note that, while I have drawn a separate box for language in the model above, there does not have to be any commitment in the account to the thesis that we possess a distinct, innately structured, natural-language module. This is, in fact, a thesis that I believe to be true. I believe that Chomsky and others have made an overwhelmingly plausible case for the existence of an innately structured, and largely modularised, language faculty. But for present purposes it might not matter very much whether natural language is largely innate or entirely learned. That is, it might not matter whether we think of our knowledge of language as incorporated in the structure of a separate input (and output) module, or, rather, as a subset of central belief. Either way, we could still maintain that reflexive thinking makes use of that knowledge, representing natural language sentences in imagination. In what follows, however, I shall assume that much of our knowledge of language is in fact innate, embodied in the structure of a specialised language faculty. Besides having a powerful claim to truth, this assumption will greatly facilitate the evolutionary story that I shall shortly begin to tell (in 8.2).
Similarly, although I have described the reflexive thinking faculty in my model as having access to an innate theory of mind (perhaps embodied in a separate module), this is not strictly necessary either. Even if common-sense psychology has to be learned or otherwise acquired in childhood, we could still maintain that reflexive thinking makes use of the concepts embedded in that theory in making judgements about its own occurrent thoughts. But I shall continue to assume, in what follows, that common-sense psychology is largely innate, both because I believe this to be true (see section 1.7 above), and because, again, it will facilitate the evolutionary story to be told in the next section.
8.2 An evolutionary story
So far in this chapter I have sketched a model of human conscious cognition in which natural language would play a central part, based on the RT theory of consciousness defended in Chapter 7. I shall now consider how the proposed cognitive architecture might have evolved. In fact it is relatively easy to see how the evolution of the structures involved in my RT-based model might have taken place, putting together, and perhaps further developing, elements which would have had independent value in survival.
Imagination would, of course, have had an important role in problem solving, as it seems to with chimpanzees today (see Kohler, 1925), and would probably have made its appearance quite early on in our ancestors' evolution. Some form of (innate) common-sense psychology, too, would have been extremely valuable to highly social creatures, such as our ancestors almost certainly were. And, of course, the arrival of natural language would then have hugely facilitated both social co-operation and the acquisition of knowledge.
In fact, as I indicated in the last section, it seems to me highly plausible that human beings have a natural language faculty which is innately structured. (In which case the language box, in the picture of human cognition presented in Figure 3, should be thought of as a distinct input and output module, not as a subset of central belief.) Moreover, it is easy to see why such an innate module should have evolved, if not exactly how. (See Bickerton, 1990, and Pinker, 1994, ch. 11, for some useful speculations on the latter.) For its arrival would have made possible the detailed exchange of information, as well as the intricate but indefinitely flexible co-ordination of activity, which underlies much of the success of our species. Just how much would then have been necessary for our ancestors to begin entertaining conscious thoughts in natural language would have depended upon what sort of role was accorded to early forms of language within their cognition.
Language for thought. Suppose, first, that the public use of signs was, from the beginning, accorded a direct role in the determination of action, in such a way that those early uses of signs could quite naturally be described as a kind of 'thinking out loud'. On this model, early forms of language would have been useful, not only for purposes of communication, but would already have had some sort of executive role in cognition. This suggestion is not at all implausible. Think of the advantages which might have accrued to our ancestors if a statement like, 'If you see a snake, don't try to pick it up', (as directed from adult to child) were (other things being equal) immediately to have the cognitive role of a belief in its content. Equally, when early hominids debated with one another about the best way to hunt an elephant, for example, and one of them said, 'We should attack from both flanks', there would have been some advantage in this statement figuring immediately as an item in practical reasoning, to be considered, manipulated, accepted, or rejected, without having to be translated into a wholly different system of representation (Mentalese).
All that would then have been necessary was for our ancestors to start using their imaginations to represent sentences of natural language, and for a reflexive feed-back loop to be added (if it was not already present in the form of short-term memory), so that they could then make use of their folk-psychological concepts in thoughts about their own thoughts and experiences. This final step, too, would have had great survival value. For it would have enabled them to keep their thoughts to themselves, where necessary, and also to maintain a continual stream of reflexive thinking without the necessity for continuous overt verbalisation.
According to this first suggestion, then, the arrival of language was (more or less) the final condition needed to facilitate conscious and hence indefinitely improvable thinking. For it is reasonable to suppose that both the power to form and manipulate mental images, as well as at least the beginnings of an innate model of human psychology, would already have been in place prior to the evolution of the language faculty, just as they seem to be with chimpanzees today. On this account, in fact, it is to the evolution of an innate language faculty that we owe our distinctive status as beings with a conscious mental life.
Of course, the language faculty is unlikely to have evolved all at once, in the space of a single mutation. And it is quite possible that its emergence was interwoven with the evolution of the theory-of-mind module. It may be, for example, that the latter evolved through a series of stages, corresponding to the theory-stages postulated by developmental psychologists for example, simple desire psychology, perception-desire psychology, and then belief-desire psychology (see Wellman, 1990, and Segal, 1995). And some psychologists have suggested that the evolution of the final stage might have presupposed some form of linguistic communication (see Harris, 1995, and Smith, 1995). So it may be more accurate to say that it is to the evolution both of an innate language faculty and of an innate theory-of-mind module that we owe our status as conscious thinkers.
(Let me stress, here, a point which has been implicit in some of the discussion of the last two chapters: namely, that the notion of modularity involved in the above speculations concerning the evolution of language and of common-sense psychology is a relatively weak one; certainly weaker than Fodor's 1983 notion. For although the operations of these modules may be fast, mandatory, and largely innately specified, they are certainly not fully encapsulated. It is plain that we have at least partial access to the contents of our theory-of-mind module, since its concepts are consciously accessible, and since we may be capable of articulating some of its main principles, such as the practical reasoning syllogism. And if the model of human cognition represented in Figure 3 is correct, central cognition will also have access to in the sense of being able to make use of some of the representations of the language module. Moreover, nothing that I say here is intended to rule out the notion of modularity defended by Karmiloff-Smith, 1992, according to which knowledge of language and of theory of mind becomes modularised in development, starting from an innate basis of attention-biases and domain-specific learning principles.)
Language for communication. Now suppose, second, that early forms of language evolved only because of the role of language in communication, as many people have suggested. On this account, language was selected for in evolution because, and only because, of its role in facilitating the acquisition of knowledge and the co-ordination of collective action. So on this model, early forms of language would not have had any part in the executive functions of human cognition, and a good deal more would then have had to change in order for human cognition to exemplify the cognitive architecture depicted in Figure 3.
In particular, merely using imagination to represent natural language sentences, in such a way that those imaginings were then made available to the same imaging faculty, would not have been enough to give those imagined sentences the status of thoughts. In order for this to happen, those imaginings would need to have usurped (at least partly) the higher cognitive functions of planning and decision making previously undertaken at some other level of the mind, using some other mode of representation. But it is by no means implausible that such a transformation should have taken place, given the obvious evolutionary advantage which would attend the capacity for reflexive thinking. On this model, then, the acquisition of language would have been the penultimate, rather than the ultimate, step necessary for the evolution of consciousness. The final step would have been the harnessing of language to the executive functions of thought.
8.3 The argument from introspection revisited
With my RT-based model of human conscious cognition in place, let us return, now, for a final evaluation of the argument from introspection of Chapter 2. That argument took as its starting point a datum of introspection which is not (or should not be) seriously disputed namely, that much of the stream of human consciousness is occupied with inner speech, or with imaged sentences (spoken or heard) of natural language. These sentences mostly have the forms (if not necessarily the causal roles) of thoughts including assertoric judgement, question, self-directed imperative, and expressions of intention and desire.
The thesis then advanced in Chapter 2, and again in section 8.1 above, was the one embodied in our common-sense psychology that these streams of imaged sentences are our conscious thoughts, and have the sort of executive role in our cognition which we believe to be distinctive of conscious thinking. So it is because I think the words, 'The world is getting warmer', that I then go on to think, 'I must use less fuel'; and it is (in part) because of this that I may later be found walking rather than driving to work. I have just now (in section 8.1) sketched the sort of cognitive architecture which could enable this to happen, and have then (in section 8.2) suggested how such an architecture might have evolved.
The upshot of these recent discussions is this. When taken together with my defence of the RT theory of consciousness against Dennett's multiple drafts model (mounted in Chapters 6 and 7), and when taken also with our consideration of Dennett's alternative explanations of inner verbalisation (in section 6.8), the above discussion shows convincingly that the common-sense position survives comparison with the Dennettian alternatives. I claim, that is, that the RT-based model of the role of inner verbalisation presented in section 8.1 is certainly to be preferred, both to Dennett's hypothesis of 'the Joycean machine', and also to his hypothesis that inner verbalisation serves as an inner bulletin board, enabling different cognitive systems to communicate with one another. But what, then, are the remaining alternatives? Are there any further competing explanations of the role of inner speech, which might be consistent with Fodor's thesis that all thought is conducted in Mentalese?
Recall the Fodor/Searle account of 'thinking aloud' sketched briefly section 3.5. In answer to the question why we go in for this activity (of uttering sentences aloud and imposing our thought-contents on them), if those sentences are in no way constitutive of the acts of thinking in question, it was suggested that such objectification of thought may serve as an aid to memory, and also as an aid to self-reflection and self-criticism. In fact there were two distinct suggestions here, which will need to be considered separately. The point of translating our Mentalese thoughts into the public-language medium might be to enable us to remember them, and/or to think about and criticise them more easily. Plainly, each of these explanations may find an analogue in the case of private not publicly expressed verbalisation.
Before setting out and discussing these suggestions in some detail, let me first stress that I believe the common-sense, RT-based, explanation to be the default one. Other things being equal, we should believe what we already believe, namely that when we engage in inner verbalisation we are thinking. So my proposed underpinning of the cognitive conception of language by claiming, in section 8.1, that much of our conscious thinking is conducted in natural language gets to win by default in the absence of any stronger competitor. In effect, I claim that a draw is all that I need in order to win.
This claim is an implication of what I take to be the only really defensible, coherentist, epistemology: in forming rational, or justified, beliefs, you should start from where you already are. Rational epistemology does not begin from sure foundations (as foundationalists would claim) but from whatever beliefs you already find yourself possessing. You should then seek to weld those beliefs into a coherent whole: constructing explanatory theories, testing theories, seeking new data, constructing alternative theories, and so on. In the course of this process a great many of your original beliefs may come to be rejected. But, being the starting-point, they also form part of the default end-point: other things being equal, those beliefs will survive, and will continue to be rationally believed. (For further discussion of various forms of coherentism, see my 1992a, and Bonjour, 1985.)
The memory explanation. The first suggestion made in section 3.5 was that overt (publicly expressed) verbalisation of a (Mentalese) thought might help us to remember it, by fixing it more firmly before the mind as a content to interact with yet further thoughts, so serving in various ways to make our thinking more efficient. (Recall from section 7.3 above, that all thinking, whether conscious or non-conscious, requires memory. Since thinking takes time, thoughts must be held in memory for long enough to achieve their effects. Anything which serves to keep them in memory for longer may therefore lead to better thinking.)
This explanation seems especially plausible in connection with thoughts expressed in writing. What the written sentence does is give us a permanent, or semi-permanent, record of the thought, to which we can return attention at will, allowing it to interact with a wide variety of beliefs and other thoughts. The suggestion was then that spoken utterances may serve a similar function, somehow fixing, or helping to fix, the thoughts in question in memory. It is easy to see how this story might go, since a thought which is uttered, and so heard by the speaker, will be entertained twice-over first prior to encoding in public language, and then second when the heard sentence is decoded back into a Mentalese thought again. And two exposures, or repetitions, of the thought may very well serve to make it more memorable.
The suggestion might then be that inner verbalisation would have a similar function by translating each occurrent thought into an imaged natural language sentence, in such a way that it is then automatically decoded back into the same thought once again, we might ensure that the thought is held in memory for long enough to interact as it should with further thoughts in processes of thinking and reasoning.
Now we might wonder, to begin with, whether this can really be a convincing explanation of the stream of inner speech, since it appears to go all round the houses to achieve the desired effect. If the proposal is that memory-enhancement is achieved by rehearsal of the thought, then it would be a great deal simpler to repeat each occurrent thought to think it twice over. Since the proposed explanation of the role of verbalisation has us thinking each thought twice over anyway, the route through public language might seem to be just an unnecessary dog-leg. In fact, however, it is a well-established fact about human memory systems that memory can be greatly enhanced by association (see Baddeley, 1988).
If asked to memorise a list of items, for example, it will be more efficient to associate them with something else, rather than simply repeating the names to yourself (even repeating them many times over). Thus, you might imagine walking around the rooms of your house, placing a distinct item in each room. This then gives you an independent fix on those items in memory you can either recall them directly, or you can recall the rooms, from which you might extract the associated item. Something similar might very well take place in the case of inner verbalisation. By translating a Mentalese thought into its imaged natural language equivalent, we get an independent fix on that thought, so making it more likely that it will be available to enter into our reasoning processes as and when the need arises.
This suggestion that inner speech functions as an aid to memory is then certainly a possible one. But to what extent can it really explain the data? In particular, can it explain the extent of inner verbalisation in our conscious lives whether we are day-dreaming, engaged in some sort of simple reasoning, or employed in a complex reasoning task? For why would we need to remember our thoughts if those thoughts are merely idle ones, as in day-dreaming or fantasy? And why would we bother to adopt the inner-verbalisation strategy as an aid to memory if our thoughts are simple ones, or are particularly easily memorable? Surely the prediction would be that inner verbalisation would only occur when we are engaged in complex practical or theoretical reasoning tasks that matter. For verbalisation takes time and energy. We surely would not do it idly.
Here my RT-based explanation can fare better. Since linguistic thinking is, by hypothesis, one of the normal modes of functioning of a special-purpose executive system (the reflexive thinking faculty), it is no surprise that inner verbalisation should continue to occur when the system is idling. For although the gains of having such a system only accrue when it is working seriously (flexibility, adaptability, etc.), of course it will continue to function in the same language-involving way when idling, or when taken 'off line'. (Compare: the carburettor in a car continues to inject a fuel-air mixture into the ignition-chamber even when the gears are not engaged, and the engine is idling.) According to the memory-explanation canvassed above, in contrast, inner verbalisation is just a dodge a strategy exploited to make thinking more efficient by making thoughts more memorable. And it is then something of a puzzle why we would continue to exploit that (costly) strategy when thinking is easy, or done to no serious purpose.
There is some reason, then, to prefer the RT-based model (according to which inner verbalisation is a kind of thinking) to the suggestion that inner speech functions as a mere memory-aid for the real thought-processes which are conducted in Mentalese. The main argument which I want to stress, however, and develop at some length, is as follows.
Given the correctness of the RT theory of consciousness defended in Chapter 7, then only if the RT-based model of section 8.1 above is correct, and inner verbalisation is a kind of thinking, do we do any conscious (propositional) thinking at all.
In fact I shall claim that it is a deep presupposition of our common-sense belief that we do sometimes engage in conscious propositional (as opposed to purely imagistic) thinking, that inner speech is such thinking. And note that if this argument succeeds, then it will not only rule out the memory-aid proposal currently under consideration, but will equally count against any theory of the role of inner speech (including Dennett's bulletin-board hypothesis) which does not accord such speech the role of thought.
According to the RT-based model developed in section 8.1, we have immediate non-inferential access to our own inner verbalisations, because a phonological representation is stored in C to be made available to further thought, just as if the sentence had been spoken or heard. And, moreover, those inner verbalisations are thoughts. So we have non-inferential access to some of our occurrent thoughts, which therefore count as conscious ones. For RT theory maintains that conscious thoughts are those to which we have immediate and non-inferential access.
According to the memory-aid proposal, on the other hand, the inner verbalisations to which we have non-inferential access are not our thoughts themselves, but rather events which are caused by those thoughts in order to improve memory. Our access to our own thoughts is therefore not immediate, but inferential to know what we have thought, we have to decode back again to reach the appropriate sentence of Mentalese which is (according to this proposal) the thought itself. So the thoughts which get verbalised in inner speech are not conscious ones, even though they are thoughts which we may reliably know ourselves to have, much as I may reliably know of the thoughts of other people on hearing them speak.
Now, so far this is just the default-argument canvassed at the outset of this section the RT-based model, and only the RT-based model, preserves for us our common-sense belief that conscious inner verbalisation is a form of conscious thinking. But when we add to this the claim that we have no other mode of engaging in conscious propositional thinking except through inner verbalisation, then we get the new argument, that only the RT-based model can preserve for us the belief that we engage in any such conscious thinking. So if we want to hang on to our common-sense belief that we do sometimes entertain conscious thoughts, and we accept the RT characterisation of consciousness, then we had better also believe that inner verbalisation is a form of thinking.
(I should stress again that by thinking, here, I should be understood as meaning cognitive processes which are properly propositional, or fully conceptual. Of course one can also treat manipulations of visual and other images as a form of thinking, and such manipulations can often have a serious purpose. Moreover, such images are characteristically conscious, occurring in such a way as to be available to further thought. So on any account there exists something one might call 'conscious thinking' which does not involve natural language. But a visual image does not express a proposition. In the present argument I am confining attention to those thinkings which are properly propositional.)
Against purely-propositional consciousness. In order for this new argument to work, I now need to show that we have no immediate and non-inferential access to our own propositional thoughts except where those thoughts are expressed in inner speech. In effect, I need to establish that there is (for us) no such phenomenon as conscious purely-propositional thinking. I require an argument to show that in those cases where subjects are prepared to self-ascribe a thought while denying that they entertained that thought verbally or in the form of any visual or other image (which they sometimes do see Hurlburt, 1990 and 1993), the thought in question was not, in fact, a conscious one. So I need to establish that human beings do not have the sort of semantic-ascent architecture that we imagined for the Stalnaker Martians in section 7.3, which might give us non-inferential access to the contents and occurrences of our Mentalese thoughts, thus rendering the latter conscious according to RT theory.
As I have already mentioned briefly in section 7.3, this has, to my mind, been convincingly demonstrated by a rich body of data coming out of the social psychology literature. Here it has been found that there are a wide variety of circumstances in which subjects will confabulate self-explanations which are manifestly false. (See Nisbett and Wilson, 1977, Nisbett and Ross, 1980, Wilson et al, 1981, Wilson, 1985, and Wilson and Stone, 1985.) What follows are just a few salient examples.
(1) When asked to select from a range of identical items, people show a marked preference for items on the right hand side of the array; but their explanations of their own choices never advert to position, but rather mention superior quality, appearance, and so on. And note that the explanations, here, can be offered within seconds of the original choice.
(2) People's willingness to help someone in distress is inversely correlated with the number of other observers the more people there are present, the less willing they are to help but subjects never mention this in explanation of their own behaviour. (This finding may now be out of date. Some of the ways in which the presence of other people can have effects upon beneficence-behaviour are now widely known, and have become absorbed into common-sense psychology. The callousness of passers-by in crowded city streets is now legendary.)
(3) People who have been paid to play with a puzzle report less intrinsic interest in it than those who do so purely voluntarily; but these reports do not correlate with the extent to which they are observed to play with it in their free time.
(4) People are very poor at knowing which factors in a situation influence their evaluations or decisions, such as which aspects of someone's behaviour influenced their evaluation of his physical characteristics (appearance, etc.), or which aspects of a job-applicant's portfolio influenced their decision to call her for interview; and interestingly, observers merely told about these studies make exactly the same erroneous judgements as do the subjects in them. Moreover, both groups (participants and observers) tend to make correct judgements when, and only when, the influential factor is recognised as such within common-sense psychology.
The best explanation of these data (and the explanation offered by Nisbett and Wilson), is that subjects in such cases lack any form of conscious access to their true thought-processes. Rather, lacking immediate access to their reasons, what they do is engage in a swift bit of retrospective self-interpretation, attributing to themselves the thoughts and feelings which they think they should have in the circumstances, or in such a way as to make sense of their own behaviour.
Looking across the full range of the experimental data available, the one factor which seems to stand out as being common to all those cases where individuals confabulate false self-explanations, is simply that in such cases the true causes of the thoughts, feelings, or behaviours in question are unknown to common-sense psychology. The best explanation of the errors, then, must be that they occur in cases where individuals are actually employing common-sense psychology, relying on its principles and generalisations to attribute mental states to themselves. And this means that in such cases the access which they have to their own mental states is inferential and interpretative (in so far as they do have access; that is, in cases where they get it right). In particular, they do not have the sort of non-inferential access to the states ascribed which would be necessary for the latter to count as conscious ones, according to the RT account of consciousness.
I propose, then, that what are often described as purely-propositional (non-verbal) thoughts, available to introspection (and hence conscious), are really the results of swift self-interpretation. So even where the self-interpretation happens to be correct, the thoughts in question are not conscious ones. For RT theory tells us that a conscious thought is one which must be available to the subject non-inferentially, not as a result of self-interpretation.
Such self-interpretations need not operate exclusively on overt behaviour, of course. Often the thoughts self-attributed may provide the best explanation of other conscious (verbalised) thoughts, or of the conscious images one was entertaining at the time, or of the way in which one was directing one's conscious attention, and so on. But I would have to predict, on this proposal, that subjects would not be prepared to self-ascribe purely-propositional thoughts in cases where there was nothing in their behaviour, circumstances, or in their other conscious mental states which might warrant the self-ascription that is, where there was nothing which might lead an observer who knew of the relevant behaviour and/or mental states to ascribe the very same thought by interpretation.
I believe that this prediction is borne out by (or is consistent with, at any rate) the introspection-sampling data provided by Russ Hurlburt (see his 1990 and 1993). Since many of his examples of purely-propositional (or 'unsymbolised') thought are not described in any great detail, it is not always possible to determine whether or not such a thought could have been attributed by a fully-knowledgeable observer. And in any case, of course, many of the relevant beliefs and intentions which might have played a part in the subject's self-interpretation may not have been conscious at the time of the consciousness-freezing beep, and so would not have been reported. For no one maintains that everything which figures in an interpretation must be conscious at the time. Nevertheless, let me work through a couple of examples by way of illustration of my approach.
In one example, the subject had been drawing a picture of her room with the radio playing, unattended, in the background just prior to the beep, but the words 'brothers in arms' had just penetrated her awareness; and at the moment of the beep, she reported that she was wondering why she had heard those particular words and not others; but this act of wondering did not involve any introspectively-accessible words or other images (see Hurlburt, 1990, p. 106). It seems not at all implausible that she should really have interpreted herself to be so wondering, in fact. For this is just what one might, very naturally, predict that she would have been doing in the circumstances.
In another example, the subject was looking at a box of breakfast cereal on the shelf of a supermarket at the time of the beep. She reported that she was wondering wordlessly whether to buy the box; and that she was thinking again wordlessly that she did not normally eat breakfast, and that the cereal might therefore be wasted; and that she was also considering the expense involved (see Hurlburt, 1993, p. 94). Again, it seems reasonable that these thoughts might have been ascribed inferentially, as a result of self-interpretation, rather than occurring in such a way as to render them properly conscious. For these are just the thoughts which an observer might naturally attribute to her, who knew what she knew: namely, that she was attending to the price-label on a cereal packet; that she did not normally eat breakfast; and that she was generally careful in matters of expense.
It may be objected against my account of these examples that it does not feel to the subjects in question as if they are interpreting themselves. On the contrary, they report their purely-propositional thoughts as having the same sort of phenomenological immediacy as any other conscious state only without the phenomenology, as it were. But the reply to this objection is easy: nor does it feel as if we are interpreting other agents much of the time, either rather, we just see much of their behaviour as intentional, and as imbued with certain thoughts. Indeed, our theory-of-mind faculty appears to have a number of the properties of a Fodorean module: besides being at least partly innately specified (see section 1.7), its operation is both mandatory and fast. We often just cannot help seeing the behaviour of an actor on the stage as displaying anger, or fear, or whatever, despite the fact that we know him to be acting. And much of the time we are not aware of ourselves as having to interpret his behaviour, either, as being deceitful, or conciliatory, or whatever; rather, we just see it that way, immediately. So it is only to be expected that when people engage in self-interpretation, this will often take place extremely swiftly, and without self-awareness of what they are doing.
(What about the puzzling data, also provided by Hurlburt, that subjects sometimes report thoughts which are definitely linguistic, and not purely-propositional, but where subjects were not aware of any determinate words, or of any imaged sentence? How can this finding be fitted into the story I am telling in this chapter about the role of natural language in conscious thinking? For example, one subject reported that at the time of the beep she was looking at a picture of Napoleon's hat, and it was exactly as if she were speaking to herself (in Dutch) the words, 'What's so special about it?', except that no words were actually present in consciousness (see Hurlburt, 1990, p. 101). Here is one possible explanation, consistent with my account: perhaps inner verbalisation, like outer verbalisation, begins with some sort of conceptual representation of the message to be expressed in LF, as it might be; see section 8.4 below which is then used to construct a phonological representation of the sentence; and this process takes time. Then perhaps, in the example, the beep interrupted this process before a phonological representation of the sentence, 'What's so special about it?', had been constructed and placed in C, available to consciousness. But still the LF representation was in existence, already formulated; and following the beep it could have been used to generate just such a phonological representation, which would then feel exactly right to the subject she would naturally have accepted it as precisely what she had been thinking at the time.)
It might be objected against the argument that I have been developing in this part of the section, that the fact that subjects sometimes confabulate as a result of engaging in self-interpretation, does not show that they do not also entertain conscious purely-propositional thoughts, perhaps exploiting some form of semantic-ascent architecture. Granted, there is no entailment. But I think that there is a sound inference to the best explanation to be made, in the absence of any particular proposal about the range of cases in which one might entertain conscious thoughts by semantic ascent. Until that proposal is forthcoming, the most reasonable explanation is that we employ a self-interpretation strategy in all cases where purely-propositional thoughts are self-ascribed. For we can then explain the heterogeneous range of examples in which people get themselves badly wrong what these cases have in common is simply that the true mental causes are here unknown to common-sense psychology, and so are not available to an interpreter, self- or otherwise. (See also Gopnik, 1993.)
If the above explanation of the social psychology false-explanation data is allowed to stand, then we face a choice. Either we can maintain that inner verbalisation is simply an aid to memory (or functions as an inner bulletin board, or whatever), and is not itself a kind of thinking; in which case we must say that we never have access to our own thoughts which is non-inferential; and so we never really have thoughts which are conscious, if the RT account of consciousness is correct. Or we can allow that the inner verbalisations to which we do have non-inferential access are actually, themselves, occurrent thoughts. In which case these thoughts, at least, can count as conscious. The latter alternative is surely the more reasonable, other things being equal preserving for us our common-sense belief that we do sometimes engage in thinking which is conscious.
The access explanation. I have argued that the memory-enhancement explanation of the stream of inner verbalisation is unsuccessful. The alternative Searle/Fodor proposal is that verbalisation of our Mentalese thoughts is designed to give us conscious access to the contents and occurrences of the latter. Not that those thoughts would thereby be conscious, according to RT theory, since our access to them would be inferential. But such a system would nevertheless give many of the advantages of conscious thinking, since one would thereby be able to think about one's own sequences of thinking and reasoning, gaining much of the flexibility and adaptability distinctive of RT theory.
So here is the proposal: we entertain our thoughts in Mentalese and not in natural language; these thoughts remain non-conscious, but we regularly translate them into inner-verbalised sentences to which we do have conscious access, so that we can mimic some of the advantages of conscious thinking. (Note that this proposal is, in effect, a variant on Dennett's inner-bulletin-board explanation of inner verbalisation, only embedded, now, in a more orthodox picture of mental architecture, containing some sort of central executive in which chains of practical and theoretical reasoning take place.)
This explanation can be dealt with quite swiftly, since it is subject to just the same weaknesses as the previous one. In particular, it will still force us to conclude that we do not engage in conscious (propositional) thinking at all. Indeed, here the counter-intuitive conclusion is forced on us even more powerfully, since it is no longer an option, even in principle, to appeal to purely-propositional conscious thinking. For if we could engage in such thinking, then we would not need to engage in inner verbalisation, on the current proposal. If the whole point of inner verbalisation is to give us indirect access to our own thought-processes, then this must be because we do not have any other form of access to them in particular, it must be because we do not have the sort of semantic-ascent architecture envisaged earlier. So on this proposal we really would have no option but to conclude that we never entertain thoughts in such a way that we have immediate and non-inferential access to their contents and occurrences. That is to say, we should have to conclude (in the presence of RT theory) that we never have any conscious thoughts at all.
The present proposal also faces similar problems to before, in explaining the range of cases where we engage in inner speech (including simple thoughts, idle thinking, day-dreaming, etc.). For why would we bother to translate our thoughts into natural language sentences in all these cases, given that we have nothing to gain by achieving second-order access to our idle thoughts? As before, the RT-based model experiences no difficulty here, since the proposal is that our reflexive thinking faculty standardly operates by manipulating natural language sentences in imagination it will therefore continue to do this even when that faculty is doing no useful work. The problem for the Mentalese-based access-explanation is to find some corresponding kind of thinking such that all and only thoughts of that kind would routinely be translated into natural language sentences so as to give us meta-access to their contents and occurrences, even when those thoughts are merely idle ones. I do not say that this cannot be done, but I cannot see, myself, how to do it.
I conclude this section, then, by claiming that the introspective argument of Chapter 2 may be allowed to stand. The best explanation for the stream of inner speech is that it is constitutive of conscious thinking. So, much of our conscious thinking does indeed take place in natural language, and Fodor is wrong to claim that all thinking is conducted in Mentalese.
8.4 Working memory and the central executive
In this section I shall compare the model of human cognitive architecture sketched in section 8.1 above with the account of the working-memory system developed over a number of years by Alan Baddeley (see Baddeley and Hitch, 1974, Baddeley, 1986, 1988, and 1993, and Gathercole and Baddeley, 1993). Both theories postulate special-purpose short-term memory stores intimately linked to such cognitive functions as planning, reasoning, and conscious awareness; and both assign a role to imagistic representations of language within the systems described. I shall then consider an objection to my position arising out of this comparison.
Comparison with the working-memory model. Baddeley has proposed that the working-memory system consists of a central executive and two specialised slave-systems, the visuo-spatial sketchpad and the phonological loop. The relationships between them are represented in Figure 4 (adapted from Gathercole and Baddeley, 1993, p. 4). The central executive controls the flow of information within the system as a whole, and is charged with such functions as action-planning, retrieval of information from long-term memory, and conscious control of action. The executive also allocates inputs to the visuo-spatial sketchpad and phonological loop, which are employed for spatial reasoning tasks and language-related tasks respectively. Since the central executive must presumably have access to linguistic knowledge, if it is to be able to generate linguistic inputs to the phonological loop, this model could easily be presented in such a way as to resemble quite closely the RT-based model of Figure 3.
***** Insert Figure 4 working memory about here *****
One difference from my RT-based model concerns the special-purpose nature of the phonological loop. In particular, Baddeley seems to think of it as essentially a phonological system. (In fact, Gathercole and Baddeley consider whether the system has an articulatory rather than a phonological basis, and leave the issue unresolved. But they do appear to believe that it must be either the one or the other see their 1993, p. 16.) In contrast, the RT-based model proposes that we can, in principle, entertain linguistically-formulated thoughts through the imaginative use of any language-related sense-modality. In normal individuals, no doubt, such thinking involves auditory, or perhaps articulatory (kinaesthetic) imagination (or both). But in the case of those whose only native language is some form of Sign the theory predicts that their linguistic thinking will involve the manipulation of visual (or kinaesthetic) images. And perhaps some ordinary thinkers, too, sometimes employ visual images (in this case of written language) in their thinking.
One empirical prediction of the RT-based model, then, is that exactly the sorts of interference-effects which have been used to explore the properties of the phonological loop in normal subjects would be found in the visual (or perhaps the kinaesthetic, gestural) modality for deaf subjects whose native language is a form of Sign. Another prediction is that aphasics or other brain-damaged patients who have lost the phonological component of working memory should be able to recover their capacity for language-based thinking by employing the resources of some other form of imagination either kinaesthetic, developing an articulatory loop, or visual, imaging written sentences. For according to the RT-based model, the exact form in which linguistic information is represented in reflexive thinking is plastic, and may vary from individual to individual, and within individuals over time.
Another difference between Baddeley's model of working-memory on the one hand, and my RT-based model of cognition on the other, concerns the function of the phonological loop its causal role in the activity of the cognitive system as a whole. In Baddeley's model, as I understand it, the phonological loop is employed only for language-based tasks that is, only for tasks which are explicitly about language, or explicitly involve language. Thus the phonological loop is said to be involved in such tasks as: memorising sequences of letters; vocabulary acquisition; reading development; and language comprehension. But there is no suggestion that it is also involved in the planning of action, or in other forms of reasoning about the world (rather than about language). These tasks are allotted, rather, to the central executive. (In fact Gathercole and Baddeley note in passing, following Hitch 1980, that the phonological loop may be implicated in mental arithmetic; see their 1993, p. 234. But nothing further is made of this point.)
In the RT-based model, in contrast, the phonological loop (and/or its equivalent in other sense-modalities) is involved in many forms of conscious thinking and reasoning about the world. My hypothesis is that it is by formulating some of our occurrent thinkings in the form of images of natural language sentences that our cognitive system is able to gain access to (in such a way as to render conscious) its own processes of thought. The function of the phonological loop is thus much more than just to enable the system to engage in language-involving processing tasks. It is also to enable the system to gain access to its own occurrent thoughts, thus facilitating the sort of indefinite self-improvement that comes with self-awareness. And according to RT theory the sentences represented in the phonological loop often are the acts of thinking the thoughts which are expressed by those sentences. Thus when, in context, I entertain in auditory imagination the sentence, 'The world is getting warmer', this is, according to the RT-based account, my act of thinking that the world is getting warmer, and is the realiser of the latter's causal role.
Of course I do have to concede that there is also a need, within my RT-based theory, for something resembling Baddeley's central executive. For something must be responsible for selecting and manipulating the imaged sentences in the phonological loop, which therefore become the system's conscious occurrent thoughts (in virtue of the reflexive availability of the contents of the loop). But the contents and modes of operation of this executive will themselves be non-conscious. It is also true that there is a need within the RT-based theory for a visuo-spatial sketchpad, in which spatial reasoning tasks can be undertaken in non-language-involving mode. Indeed, this has been implicit in the discussion of RT theory throughout, both here and in the previous chapter. So there are still a number of respects in which Baddeley might find my RT-based account congenial.
An objection and three replies. It may be objected that for me to concede to Baddeley the existence of a central executive charged with selecting and formulating the sentences to be imaged within the phonological loop, is for me to concede to Fodor the case against the role of language in thinking. For on standard models of sentence production, this process proceeds through a number of different stages, beginning with a conceptual representation of the message to be conveyed, and concluding with detailed instructions to the muscles in the mouth and larynx to express that message in speech. (See, for example, Garrett, 1975, and Levelt, 1989.) But what would 'a conceptual representation of the message' be, except a thought whose content is that message? Since this thought could not (on pain of vicious circularity) itself be expressed in natural language, it would presumably have to be formulated in Mentalese. And then Fodor gets his conclusion that the real thought-processes underlying inner verbalisation do not constitutively involve natural language, but rather employ Mentalese.
I have three possible replies to this objection, each one more controversial than the last. My fall-back position is to endorse at least the first. This would give me the thesis I shall later call 'NNw'. I shall present arguments later in this chapter in support of the second. This would give me the thesis I shall call 'NNs'. And I am strongly tempted by the third, but will not pursue it any further in this book.
(1) The least controversial reply is to concede the standard model of speech generation, and to concede that the conceptual representation of the message is expressed in Mentalese, but to claim, firstly, that this representation can only attain the status of a conscious thought by being expressed in natural language, in the form of inner speech; and secondly, that it is the latter imaged sentence which occupies the causal role distinctive of conscious thinking.
The first of these claims has been partially defended above, and will be returned to again in sections 8.5 and 8.6 below, in the form of thesis NNw. I do not believe that we have the capacity for non-inferential access to our own occurrent thoughts except when they are formulated in inner speech (and/or other images). The second claim commits me to the truth of such counterfactuals as this: if my conceptual representation of the message, [the Earth is getting warmer] had never been realised in the imaged sentence, 'The Earth is getting warmer', then I would never have gone on to draw the conscious consequences that I did draw, nor go on to act as I did. We are warranted in believing such counterfactuals, I maintain, because the RT-based model of inner verbalisation provides a better account of the phenomena than the alternatives, as I argued in section 8.3 above.
(2) The second, more controversial, reply, again concedes the standard model of speech generation, allowing that speech processes (whether inner or outer) begin with some sort of conceptual representation of the thought to be expressed; but it denies that this representation can itself have the causal role of a thought (whether conscious or not), independent of its effects in sentence-production. Or rather, it insists that these conceptual representations are themselves natural language ones. It may be, that is, that such representations are not central-process thoughts expressed in Mentalese, but are internal to the language faculty, having no independent causal role outside of it. I can then concede that the cognitive processes involved in the production of a sentence may include multiple levels of representation (phonological, syntactic, and crucially semantic), while claiming that all of these representations are themselves natural language ones.
For example, Chomsky has maintained that there is a level of linguistic representation which he calls 'logical form' (LF), which is where the language faculty interfaces with central cognitive systems (see his 1995). I can then claim that all conceptual, propositional, thinking consists in the formation and manipulation of these LF representations. The hypothesis can be that central-process thinking operates by accessing and manipulating the representations of the language faculty. Where these representations are only in LF, the thoughts in question will be non-conscious ones. But where the LF representation is used to generate a full-blown phonological representation (an imaged sentence), the thought will be conscious. I shall return to this suggestion in section 8.7 below. Notice that if this proposal is to be much more than a mere variant on the access-explanation canvassed in section 8.3, then I must claim that there is a distinctive executive system which can only operate on fully-represented phonological sentences; indeed, my claim is that this system is the reflexive thinking faculty.
(Philosophers and logicians should note that Chomsky's LF is very different from what they are apt to mean by 'logical form'. In particular, sentences of LF do not just contain logical constants and quantifiers, variables, and dummy names. Rather, they consist of lexical items drawn from the natural language in question, syntactically structured, but regimented in such a way that all scope-ambiguities and such like are resolved, and with pronouns cross-indexed to their referents and so on. And the lexical items will be semantically interpreted, linked to whatever structures in the knowledge-base secure their meanings. The important point for these purposes is that sentences of LF are not sentences of Mentalese they are not central-process representations, but are rather internal to the language faculty; and they are not universal to all thinkers, but are always drawn from one or another natural language.)
What would be the motivation behind this second, more controversial, reply? It would be this: we shall need to make such a reply if we come to believe, not just that inner verbalisation is often the only way in which the system can have non-inferential access to its own occurrent thoughts (and hence render them conscious), but also that there are some thoughts which cannot be entertained independently of language. That is, if there are some thoughts (considered as types rather than tokens) which constitutively involve natural language, then we shall need to deny that any conceptual representations of those thoughts, employed in the processes which generate the sentences in question, can themselves exist independent of the language-faculty. This is an issue that I shall return to in section 8.7 below, in the form of thesis NNs.
(3) Thirdly, and much more controversially, I could reply to the objection by rejecting the standard models of speech production. By way of motivation, notice that exactly the same question can be asked about the supposed conceptual representation of the message, as it might be expressed in Mentalese, as is asked about the inner verbalisation of that message in natural language namely, 'How did it come to be formulated as it was?'. In explaining how a sentence of Mentalese is put together there are surely just the same sorts of reasons which apply in the case of natural language, for proposing that the process starts with a conceptual representation of the thought to be expressed, before proceeding to the selection of lexical items and grammatical forms to express that thought. But plainly this then would, in this case, lead to a vicious regress.
Now, if it is allowed, as it must be, that Mentalese thoughts are formulated without a prior conceptual representation of their content, then why should the same concession not be made for natural language? We could then adopt, for example, some variant on Dennett's 'pandemonium model' for speech production instead (see his 1991, ch. 8). According to this model, sentences are constructed in cognition out of relative chaos (and certainly without any prior representation of the message to be expressed), by multiple sub-units proposing sentence parts to a set of assessor units (presumably operating in a context-sensitive way), until the whole system settles upon a sentence which it finds acceptable.
It is sometimes objected against the hypothesis of Mentalese (for example, by Dummett, 1989), that it can bring no advantage over the hypothesis that thought is conducted in natural language, without generating a vicious regress. This is a misunderstanding if it is directed at the Mentalese-based explanation of the content of natural language sentences. As we saw in Chapter 3, it may very well be possible to reduce the content of a natural language sentence to the content of a thought, and then to offer some independent semantics for the latter (e.g. in terms of causal co-variance). While I have argued that such semantic theories are by no means mandatory, and face a number of difficulties, there is no reason to think that they generate any vicious regress.
Matters may be different, however, when we turn our attention from explanations of the semantics of natural language, to explanations of the status of natural language utterances as intentional actions. Here there really is a threat of a regress if we explain the intentional status of a natural language utterance in terms of an intentionally formulated representation of the message to be expressed. For then we shall need yet another representation to explain the intentional status of the latter, and so on. And if we can explain how the construction of a Mentalese sentence can be intentional without appealing to any prior intention, then it is not clear why we cannot adopt just the same sort of explanation of the intentional construction of a natural language sentence.
To put the point somewhat differently: at some level, plainly, sentence-based cognition (whether involving Mentalese or natural language) must be realised in processes which are not sentential perhaps involving connectionist networks of various sorts. And if such processes are appealed to in the explanation Mentalese-based thinking, then it might be possible to adopt a similar form of explanation for natural-language-based thinking, cutting out Mentalese from cognition altogether. This is, at any rate, a research avenue well worth exploring.
8.5 The thesis of natural necessity (weak)
Let us now take final stock of our position. I have responded to all of Fodor's arguments attempting to show that Mentalese is the language in which all thinking is conducted. Some of these proved easy to reply to, while some particularly the semantic ones proved more difficult. They required a long detour to consider the semantics of natural language and of thought, conducted in Chapters 3 and 4. Then the provision of an adequate theory of consciousness, in Chapter 7, has finally put us in position to reply to the one remaining argument, from the thoughts of animals and infants. So we are now able to conclude that Fodor has given us no good reason for believing that all thinking takes place in Mentalese.
Yet we still have in place, however, the introspective argument of Chapter 2, for the conclusion that some thinking takes place in natural language (specifically, human conscious thinking which is genuinely propositional). Indeed, this has now been vindicated by our discussion of theories of consciousness in Chapters 5 through to the present. For although it is possible to construct theories of inner verbalisation specifically, the Dennett/Fodor hybrid introduced in section 6.8, or the memory-enhancement explanation discussed in section 8.3 which predict that we are systematically deceived in making the introspective claim, these theories are by no means the best available. On the contrary, the RT-based model of human conscious cognition which I have finally endorsed gives us every reason to believe that introspection is reliable on this matter.
So the conclusion is (at least tentatively) established: some thinking constitutively involves natural language. This is already enough to secure at least a weak version of the cognitive (as opposed to the communicative) conception of natural language, since we have shown that language is not just an input and output module of the mind, but is implicated in central processes of believing, desiring, and reasoning. However, even this conclusion is, as yet, a bare claim of fact. No modal claims are yet warranted. So this must be our final task. I propose to argue that there is a natural necessity attaching to the fact that human conscious thinking involves public or natural language. (The sense in which I intend the term 'public' will be elucidated in section 8.6 below.)
I have sketched, in outline, a model of how human conscious thinking does in fact take place, and of how such cognitive structures might have evolved. But for all that has so far been said, it might seem to remain entirely accidental that humans should ever entertain their conscious thoughts in natural language. Is there no element of necessity attaching to the model? I believe that there is. The first (weaker) version of the claim that I wish to defend is the following.
NNw: Some human conscious thinking (viz. conscious propositional thinking) is such that, of natural necessity, it involves public language, in virtue of the given architecture of human cognition together with causal laws.
In the remainder of this section I shall first comment on the scope of the above claim, and then present arguments in its defence. In the section that follows I shall then say something further about its proper interpretation.
The scope of NNw. The claim expressed in NNw is relatively weak in a number of different respects. First, it does not yet follow from it that there are any thought-contents (considered as types) which can only be grasped through their expression in some public language. In fact, for all that NNw has to say, it may be that every thought which we can entertain consciously, through the medium of public language, we can also entertain non-consciously, in the medium of some sort of Mentalese. A stronger version of the thesis, which implies that there are some thought-types which can only be entertained at all in public language NNs will be explored in the final two sections of this chapter.
The second way in which NNw is relatively weak, is that it does not claim that all conscious human thinking involves public language. It is conceded that some conscious thinking can be purely imagistic, without involving language, as we have been allowing since near the outset of this book, in section 1.8. Admittedly, (to reiterate a point from section 2.2), it may be that in many of those cases where a thought is carried by an image of an object, an embedding in a linguistic context may be necessary to confer on that image a determinate content. In which case such thoughts, too, will turn out to involve public language. So when I think, while at home, of how to arrange the furniture in my office, employing an image to express my thought, what makes that image into an image of my office will be an implicit or explicit sentential embedding. My full thought ought properly to be expressed thus: 'I shall arrange the furniture in my office like this [insert image]'.
Moreover, it should also be pointed out again, that if our thoughts are to have anything like the expressive powers they actually possess, even amongst primitive peoples (setting aside thoughts about the sub-atomic particles of physics, for example), then they would have to employ some sort of symbolic language. For as we noted in section 1.8, it appears quite impossible for images, as such, to carry the content of any but the very simplest thoughts about the immediate environment. (Of course, as we noted, it would always be possible to teach someone to use images as symbols, somewhat like an ideographic script, and so gain all of the expressive powers of natural language. But then this would be a public language, in the sense that I intend. Public languages do not have to be speakable. An entirely visual language is still a language.) It is easy to understand how an image of a stack of perceptually presented boxes might, in context, constitute the thought of a solution to some practical problem. But it is very hard indeed to see how any image by itself, not used conventionally in the manner of an ideographic script, could carry the content of even a perfectly ordinary thought like, 'Ripe apples can be either red or green'.
Despite the above points, it has to be admitted that sometimes our conscious thoughts can consist entirely of images of objects, and not of images used as conventional symbols. As we said, the thoughts of a composer may consist entirely of auditory images, the thoughts of an engineer may sometimes consist entirely of visual images of arrangements of objects, and my thoughts as I find may way across a darkened room might consist simply in an image of its layout in egocentric space. NNw is not intended to deny any of this.
Note, however, that a combination of the views being defended here will together entail that the status of an image as conscious depends upon the thinker's possession of a natural language. For if the points made in section 1.8 were sound, concerning the severe limitations on the representative powers of images, then it is almost certainly the case that no image by itself can represent, or carry a thought about, another image as such. Yet RT theory tells us that an image will only count as conscious if it is available to conscious thought. And if the thesis NNw is true, these conscious thoughts will then have to be entertained linguistically. So the capacity to entertain a conscious thought about an image, necessary for the latter to be conscious according to RT theory, depends upon language. But this is only so for us. Other species (the Stalnaker Martians, and perhaps chimpanzees) may be capable of conscious thinking without employing language.
But what, then, is the intended scope of NNw? Precisely which types of conscious thinking does NNw claim (in the human case) to involve public language as a matter of natural necessity? The answer is: all those types of conscious thinking which, as a matter of fact, constitutively involve such language (in the manner already established in section 8.3). Wherever a conscious thought is constituted by a natural language sentence, my hypothesis is that a thought of that type (individuated by content) can only be entertained consciously, by us, in its public-language form. The arguments for this view will be set out in the next part of this section, where I shall reply to a number of objections.
The argument for NNw. In assessing the truth of NNw, much may depend upon the correct surroundings for the model of human cognition depicted in Figure 3, and on the powers inherent in other aspects of the system. For as it stands, if we focus only on Figure 3, then the claim embodied in NNw is almost trivially true. For removal of the language box, or of the dotted lines of access between language and reflexive thinking, would, by hypothesis, destroy the functioning of the system as a whole. The question is whether there is any alternative way in which a (properly propositional) reflexive thinking structure could be supported in human cognition without employing public language. In this part of the section I shall consider four alternative possibilities. These are, so far as I can see, there are only four alternatives available. If none of them is viable, then NNw will be established as true by default.
(1) The first possibility is that some of those thought-types which are, as a matter of fact, tokened in us in the form of natural language sentences, could have been tokened in the form of mental images instead. But this possibility was, in effect, ruled out in section 1.8, where I argued that purely-imagistic and properly-propositional thoughts constitute a disjoint set that is to say, that any thought which one actually entertains in language one could not entertain purely imagistically, and vice versa.
Recall how this point was established in section 1.8, in connection with thoughts whose contents do not go beyond what can be perceptually manifest (where one might have expected there to be the strongest case for the content to be thinkable in the form of an image). We considered, in particular, the thought, [that a cat is sitting on a mat]. If this thought is tokened in the form of the English sentence, 'A cat is sitting on a mat', then according to the view defended in section 8.3, this token of that thought is constituted by the sentence in question. But might one not also entertain a token of that very thought in the form of a visual image of a cat sitting on a mat? And would this then not be an alternative manner in which the thought could be entertained consciously, provided that the image in question was itself available to reflexive thinking?
In section 1.8 I answered these questions in the negative, drawing on the claim that content is, at least partly, individuated by functional or conceptual role to make the point. Since any image always has excess content in comparison with a sentence, the contents of the two can never really be the same. So in entertaining an image of a cat on a mat one would not really be entertaining a thought of the very same type (viz. a properly propositional thought), even though it might concern the same subject matter, and might on occasion play a somewhat similar role in cognition. So the point is established: any thought-type which is in fact tokened consciously in the form of a natural language sentence, could not have been entertained in the absence of language in the form of an image.
Of course, as we also noted in section 1.8, it is also possible to group thought-contents into types by subject matter, as well as by functional role. By this criterion, an image of a cat on a mat can express the same content as the sentence, 'A cat is on a mat'. And then we shall have to concede that some types of conscious thought (viz. thoughts whose subject-matters can be represented purely imagistically) do not constitutively involve public language. But it seems certain that the range of such thoughts will be extremely limited, probably being confined to representations of perceptible objects and events in our immediate spatio-temporal environment. So NNw will still claim that conscious thoughts which concern subject-matters that are any more complex or sophisticated than this will have to involve public language.
(2) The second possible way in which someone might try to deny the thesis of natural necessity, NNw, would be to maintain that human beings do not need to access knowledge of public language in order to entertain conscious propositional thoughts, because they already have an innate language namely, Mentalese whose forms, structures, and vocabulary they could access instead. The claim here, is that everything in the cognitive architecture depicted in Figure 3 could remain as it is, in the absence of public language, but with reflexive access to sentences of Mentalese replacing the sort of access we actually have to inner deployments of natural language sentences.
This proposal seems plainly false, however. Even if we do possess some sort of Mentalese, we have no access, surely, to its forms and structures we are incapable of thinking a thought in Mentalese, and then thinking (non-inferentially) about the Mentalese sentence we have just entertained. So without a public language we should be incapable of conscious (propositional) thinking, if the RT-based account depicted in Figure 3 is correct.
I suppose it might possibly be replied that we were each of us, as pre-linguistic children, capable of entertaining conscious thoughts in Mentalese, but that Mentalese was later supplanted by natural language. According to this suggestion, even if we had never learned a public language we would still have been capable of thinking about our own thoughts on a regular basis, and would then have had access to their Mentalese mode of expression. But why would learning a public language supplant Mentalese, rather than come to exist alongside it? After all, when people learn a second natural language, and become so proficient that they can do their thinking in that language, this does not deprive them of the ability to think in their first language. Moreover, it would surely be strange indeed that not one of us should have any idea what the sentences of Mentalese might be like, if we had each of us had access to their forms and structures as young children!
(3) The third way in which someone might try to deny the thesis of natural necessity, NNw, would be by claiming that we do in fact have the capacity for immediate knowledge of what we have just thought without knowledge of how we have thought it, which might be underpinned by some sort of semantic-ascent mechanism in Mentalese. That is, it might be claimed that while we do employ public language in some of our conscious thinking, in the sort of way depicted in Figure 3, we also have the kind of 'semantic-ascent architecture' we imagined for the Stalnaker Martians in section 7.3. Support for this view might be sought in those many familiar cases where we know, immediately, why we did something, or what we were thinking or believing, without knowing in what form those thoughts were expressed. So, according to this suggestion, even if we lacked any public language, we might still be capable of the sort of immediate knowledge of what we have just thought which is necessary for conscious thinking.
The main problem with this suggestion was pointed out in section 8.3. It is that all the evidence suggests that our knowledge of our own recent thinkings and reasonings, where those reasons have not figured in consciousness in sentential (or imagistic) form, is not immediate, but inferential in which case those reasonings were, in fact, non-conscious rather than conscious, if the RT account of consciousness is correct. I argued that this is manifest from the many cases where subjects will claim knowledge of their reasons which is wildly at variance with the facts for example, claiming to have selected a garment for its colour, when the evidence shows that it was selected, rather, because it was on the right hand side of an array of such garments. I claimed that what we actually do in such cases is engage in a sort of self-interpretation which is not essentially different in form from other-interpretation, but done so swiftly and smoothly that we hardly notice what we are about. In which case I am left free to claim that it is only where our reasonings occur in the form of occurrent sententially or imagistically expressed judgements, that we have immediate (reflexive) knowledge of them. And it is only in such cases that our reasonings should be counted as conscious, if my RT account is correct.
Notice, too, that according to this third suggested alternative to NNw, the reflexive thinking architecture depicted in Figure 2 would have to exist alongside, and in addition to, the language-involving architecture depicted in Figure 3. For the proposal being made, is that even if we did not have the sort of reflexive access to the forms of our (public-language) thoughts, required for the functioning of the Figure 3 architecture, we could still have had reflexive access to the contents and occurrences of our occurrent thinkings. This then appears wildly extravagant. Why should we have evolved two such reflexive-access structures? And why would we bother to run our conscious thoughts through formulation in public language if we could think them, and know that we were thinking them, without it?
(4) Finally, the only other way in which someone could try to deny the natural necessity thesis outlined above, would be by maintaining that a human being who lacked a natural language would still be capable of entertaining conscious propositional thoughts by inventing a symbolic system to deploy in imagination, to whose forms the thinker would then have access. Now there are two ways of taking this proposal. And taken in one way, it seems likely to be true; but taken in the other, it seems likely to be false. Yet neither alternative genuinely threatens NNw. Let me explain.
As we saw in section 2.1, there is evidence that our innate language faculty will seize upon any quasi-linguistic data, no matter how degenerate, to begin constructing properly linguistic representations. Thus Susan Goldin-Meadow and colleagues have found in their studies of congenitally deaf children deprived of any language experience, that the latter will pick up the gestures employed by their parents, extending, regularising, and transforming them into a simple form of natural sign-language (see Goldin-Meadow and Mylander, 1990, Butcher et al, 1991, and Goldin-Meadow et al, 1994). So it may be that even in the absence of linguistic exemplars, normal human beings will employ their innate language-building capacities to construct their own symbolic systems, with which they can then think, and which they could then entertain in imagination in such a way as to render their propositional thoughts conscious. But this proposal is not inconsistent with NNw, since the resulting symbolic system would still be a natural language. So it would not be the case that the thoughts in question were ever entertained consciously in the absence of such language.
The other possible way to take this fourth suggestion, is to propose that language-deprived humans would be capable of using central reasoning processes to invent their own language (which would then almost certainly not be a natural one by Chomsky's criteria, since it would fail to satisfy the principles and parameters of the language faculty). In its literal sense, though, invention plainly requires thought not necessarily conscious thought, perhaps, but certainly thought. For in order to invent something you have to be able to represent it to yourself at least partially and schematically in advance. So the question is whether our non-conscious, pre-linguistic, thoughts could have the conceptual resources and sophistication required to invent something as complex as a language. This seems highly unlikely. Is it only lack brain-power that prevents chimpanzees from inventing a language, for example? I suggest not. It is rather that the conceptual resources available to them are simply not up to the task. Moreover, this suggestion would predict that a 'wolf-child', or a pre-signing deaf child, should be able to invent and use its own language, even if it suffered from some form of severe specific language-impairment. But not only is there no evidence that this happens, but it is difficult to see how it could, given the kinds of thoughts which seem available to such children.
The main point to notice, however, is that even if language-deprived children could use central reasoning processes to invent their own symbolic system to deploy in imagination, this would still not be inconsistent with NNw. For although the resulting language would not be a natural one, it would, almost certainly, be a public one, in the sense to be discussed in the next section. That is, the thinker's conscious thoughts would still consist of images of movements, or shapes, or sounds, which could in principle be externalised in the form of actual movements, or shapes, or sounds. So, again, at no point would these children have been able to employ conscious thoughts in the absence of a public language, which is what would be necessary for this proposal to be inconsistent with NNw.
So I conclude, then, that there are no real alternatives to the weak thesis of natural necessity, NNw. Since human beings can neither access nor re-use the sentences of any form of Mentalese, in order to entertain thoughts about their own thoughts, nor employ some other sort of symbolic system for the same purpose, we must be dependent upon the acquisition of a public language in order to be capable of conscious propositional thinking.
8.6 Objections and elucidations
In this section I shall try to cast some further light on the thesis NNw, first discussing the sense in which it is claimed to be naturally necessary, and then commenting on the sense in which the language employed in conscious thought has to be a public one (this will enable me to respond to an objection to NNw made from a Chomskian perspective).
Natural necessity. I commented earlier on the various qualifications to the scope of NNw. Here let me make some remarks on its status as a claim of natural necessity. One might initially think that it has an exactly parallel form and status to that of the following: 'It is naturally necessary that bodies on Earth fall with an acceleration of 32 feet per second per second'. (A restricted version of the law of gravity). This latter statement is true in virtue of the given mass of the Earth, together with the universal law of gravity. While it has to be acknowledged that the mass of the Earth is contingent, and might have been otherwise consistent with the laws of nature remaining as they are, things would had to have gone very differently in the past in order for it to be different. And if we treat the mass of the Earth as a constant, subserving a stable local sub-system of nature, then what we get is something that, given that constant, has to be the case.
In the same way, then, I can claim that while the architecture of human cognition is contingent (depending, as it does, on the accidents of evolution), things would had to have gone very differently in the course of our evolution in order for it to have been different. And if we treat that architecture as a constant, subserving a stable local sub-system of nature (viz. human psychology), then what we get is something that has to be the case. That is, my claim can be that human beings would had to have evolved a different cognitive architecture in order for them not to have entertained conscious thoughts in public language. Or equivalently: given that human beings have evolved as they have, they cannot now entertain conscious propositional thoughts in any other medium.
The analogy between thesis NNw and the restricted version of the law of gravity is by no means perfect, however. For what, in the case of NNw, corresponds to the universal law of gravity? What law of nature is such that, when combined with the description of human cognitive architecture depicted in Figure 3, will yield the conclusion that human conscious thinking must involve public language? No candidates suggest themselves. While the restricted law of gravity is useful in illustrating how there can be natural necessities which nevertheless depend upon entirely contingent initial conditions, it does not provide the best model for understanding NNw.
The model I actually prefer it this: 'You cannot make a motor-car engine out of a bicycle'. Or, more strictly and fully: 'You cannot, using only such tools as spanner and screwdriver, reassemble the parts of a bicycle into a functioning motor-car engine'. This statement is, surely, naturally necessary. Moreover, it is a necessity which, as with the previous example, depends upon the given (and contingent) facts of the respective structures of a bicycle and a car engine, and the shapes of their parts. Yet there is no single law which, when combined with descriptions of those structures, will entail the statement above. Rather, the necessity of the statement depends upon many different laws of physics and mechanics, operating at a number of distinct levels.
Similarly, then, in the case of NNw: I suggest that the given structure and parts of human cognition (as depicted in some suitably expanded and enhanced version of Figure 3) are such that they cannot be re-combined or re-used in such a way as to enable conscious (reflexive) thinking to take place without employing public language, at least not if that thinking is to be genuinely propositional in nature. And then this is exactly the claim which was defended in the latter part of the previous section.
Public versus natural language. Notice that many of the claims made above have been expressed in terms of public, rather than of natural, language. This is so because, as I have already indicated, I endorse the existence of an innately structured language faculty of the sort defended by Noam Chomsky. For the term natural language has come to be synonymous, amongst many of Chomsky's followers, with the idea of a language which is permitted by the innate structures the principles and parameters of our language faculty. But Chomsky's thesis only predicts that the learning of an artificial language (that is, one which is non-natural in the above sense) would be very slow and laborious, not that it would be altogether impossible.
So it may remain true, consistent with the existence of an innately structured language module, that a human child could be brought up from birth, with much effort and training, to speak a non-natural language as their only language. In which case it might, in fact, be possible for an individual human being to entertain conscious thoughts by using a symbolic system that did not depend upon their prior possession of a natural language, and that failed to satisfy the innate principles and parameters of the language faculty. But this would still count as a public language, in at least two respects. First, it would have been learned though some public process of teaching, reward, and correction. Second, its signs, in their use in private thought, would nevertheless admit of a straightforward public expression, since they would, presumably, consist of images of patterns of shape, or sound, or movement, which could be re-produced readily in the public domain should the situation so demand.
So it might, in fact, be possible for human beings to entertain conscious thoughts without having mastery of a natural language, if they had been brought up and trained to employ some other suitably structured system of representations. But this language would still count as a public one, in the sense that I intend. Moreover, their use of the signs and symbols in question would still, in fact, be parasitic upon someone's possession and use of a natural language. For it would surely be impossible for in individual to invent such a system for themselves, relying only upon non-conscious, non-language-involving, thoughts. It seems highly plausible that anyone capable of inventing a non-natural language would already have to be master of some natural language. In which case, we can now claim that it is impossible for human beings in general to entertain conscious thoughts except through their possession of natural language. If there exist individuals who have conscious thoughts without having mastery of a natural language, then this will be parasitic upon the conscious thoughts of those who do have such mastery.
I have been claiming that human beings, both in general and as individuals, are incapable of entertaining conscious thoughts except through their mastery of some public language. Note that it is entirely consistent with this, that there should be individual human beings who do their thinking in some sort of private code, as did Samuel Pepys in his diaries. For such codes are only 'private' in the sense that, as a contingent matter of fact, no one else is in a position to interpret and understand them. The private thoughts of such people would still consist of images of potentially public symbols, which the thinker could express publicly (as Pepys in fact did, by writing them down), and which could be explained to others if the thinker should choose to do so.
True enough, such codes would only satisfy one of the two criteria for being public, set out above. For, while they would be publicly expressible, they would not have been acquired through any public process of teaching and learning. Even so, such codes would be parasitic upon some public (almost certainly natural) language for their existence, since they would have been invented through the use of the latter. Moreover, they would presumably be structurally isomorphic to some public language, at least semantically, since their invention would be largely a matter of selecting new signs to express old (publicly expressible) concepts. So I would still want to claim that such people do their conscious thinkings through the use of a public language.
One further reason for expressing thesis NNw in terms of public rather than of natural language is this: I want to allow that thoughts expressed in mathematical, scientific, or musical notation, for example, should count as linguistic. For such notations are, of course, not natural, in the sense in which English and French are natural, but rather invented. Yet I would certainly want to allow that they make it possible for us to entertain, and are partially constitutive of, new kinds of conscious thinking. Mathematicians, scientists, and composers certainly engage in a kind of thinking, and thinking which is genuinely propositional; but this thinking need not be mediated by natural language. All the same, these are notations which have to be learned, and which have a straightforward public expression. Indeed, they are best thought of as extensions of natural language, rather than as wholly disjoint symbolic systems. They are therefore public, and I can thus continue to claim that it is naturally necessary that human conscious thinking should involve public language.
It is important to stress here, however, that my talk of public language in articulating and defending thesis NNw, need not carry any commitment to the real existence of public languages as objects of scientific study. In particular, I do not have to accept that there really is such a language as English, in any sense other than an abstraction over a vast range (only vaguely specifiable) of slightly different individual idiolects. In fact, I do not have to be committed to what Chomsky calls E-language as opposed to I-language (see his 1986). On the contrary, I can agree with Chomsky that the proper object of scientific study is the state of the individual's language system (their I-language). I merely maintain that the mature state of that system will always be a language which is public in the sense that it is capable of public expression. In contrast with Mentalese, the languages we employ in conscious thought can always be publicly expressed, or manifested, in principle (abstracting from such possibilities as paralysis and motor aphasia). There is nothing here that Chomsky need object to, in my view.
I should also stress that I do not intend to be endorsing, here, any sort of rejection of private language in Wittgenstein's sense (see his 1953, §§ 243ff). This would be a language whose terms refer to inner private experiences. On the contrary, in section 7.6 above I conceded that it may be possible for us to deploy purely recognitional concepts of inner experience. But if these concepts are to figure in conscious propositional thoughts, then according to NNw they will have to be expressed by terms which are public in the sense that they are images of (potentially) publicly expressible signs. The contrast I intend between public and private language is not a semantic one it does not concern the subject-matter of the language but rather lexical/syntactic. The intended contrast is between public language and Mentalese, in fact.
8.7 The thesis of natural necessity (strong)
I have defended the weaker version of NN, according to which our cognition is so structured that we can only entertain conscious propositional thoughts by formulating them in public (normally natural) language. But this does not yet show that such language is constitutive of the thought-types so formulated. What needs to be investigated, now, is the sort of case that can be made out in support of the stronger version of NN, as follows.
NNs: Some human conscious thinking is such that, of natural necessity, it involves public language (in virtue of the given architecture of human cognition, together with causal laws); and, necessarily, some of these propositional thoughts belong to types which (for us at least) constitutively involve such language.
As I noted in section 4.6, it is only if this stronger thesis can be established that it will follow that public language is constitutive of (many of) our conscious thoughts as types (individuated by content). And it would seem that the full significance for philosophy and psychology which we advertised in section 1.2 will only be realised if this stronger thesis can be established (I shall return to this question in section 8.8).
What NNs implies is that many of our conscious thoughts, as tokens, constitutively involve public language, and that thoughts of those types could not have been tokened consciously in the absence of such language (this is thesis NNw, which NNs entails). The thesis NNs also implies, in addition, that many of those conscious thought-tokens belong to types which constitutively involve public language, so that one could not entertain any token thought of that type (whether consciously or non-consciously) in the absence of such language.
Note that while NNs tells us that many of our conscious thoughts, as types, must involve public language, it does not have to imply that there are any types of thought which can only be entertained consciously. Indeed, I can see no good reason why each of our public-language-involving thoughts should not be able to occur non-consciously as well as consciously. And in section 8.4 we saw in outline how this might be possible, in fact propositional thinking, in us, may take place by means of the central systems accessing and deploying representations drawn from the language faculty; when these representations are in LF only, the thoughts will be non-conscious ones; but where the representations in LF are used to generate a phonological representation as well, which can then be held in the short-term memory store C in such a way as to be made available to further thought, then the thoughts in question will be conscious ones. I shall return to this proposal shortly.
The thesis NNs, itself, therefore leaves it open whether there are some thought-types which can only be entertained consciously, or whether every thought which we can entertain consciously, in public language, we can also entertain non-consciously in the same medium. But in fact I am strongly inclined to believe the latter, on the grounds that most genuinely innovative and creative thinking appears to take place non-consciously. If it is possible to entertain non-conscious thoughts about the relativity of space-time, or about the Benzene ring, or about a chess end-game (as the evidence from both creative thinkers and ordinary experience suggests) then it seems very likely that there are no real limits on the subject-matters of our non-conscious thoughts. (See Ghiselin, 1952.)
How can NNs be defended? I have two lines of argument. One is to compare NNs directly with its Mentalese-involving competitors, arguing that the former is preferable on grounds of simplicity (among others). The second involves returning to the arguments from developmental psychology and aphasia of section 2.1, reassessing their strength in the light of NNw. For if we can take it as already established that public language is necessarily employed in our conscious thinking, as NNw maintains, then the simplest explanation of the data may be that language is essentially involved in is constitutive of many of the types of thought which are available to us, whether those thoughts are entertained consciously or non-consciously. I shall elaborate each argument in turn.
The argument against the competition. To deny NNs, while continuing to accept NNw, one has to deny that any of the thought-types which actually involve public or natural language (when entertained consciously) necessarily involve such language. That is, one must maintain that it is naturally possible for us to entertain (non-consciously) the full range of thoughts which are available to us in public language, but without employing such language. There are just two conceivable ways in which this might be done, the first of which can be dismissed almost immediately.
The first way of denying NNs is developmental. We could allow that, where our conscious thoughts involve language, all tokens of those thoughts which we ever actually entertain consciously (in the normal case) are similarly language-involving (this is NNw). But we could maintain, nevertheless, that if we had never acquired public language, then thoughts of all those types would still have been available to us non-consciously but in that case expressed in Mentalese rather than a public language. Such a claim would obviously be extravagant, however, and would conflict with what I have taken the lovers of Mentalese to be conceding from near the outset of this book: namely, that possession of a public language is at least a necessary condition for us to entertain many types of thought, even if it is not always thereafter involved in those thoughts.
The second way in which NNs might be denied (while accepting NNw) would be to claim that our (adult) non-conscious thinking faculty has available to it all of the conceptual resources of conscious (public-language-involving) thinking, but expressed in Mentalese rather than public language. On this account, while possession of a public language may have been a necessary condition for us to acquire certain concepts and so entertain certain thoughts; and while such language may be constitutively involved in the conscious occurrence of those thoughts (as NNw maintains); still thoughts of those types can be activated (non-consciously) without the involvement of public language, and it would still be possible for us so to entertain them if the capacity for such language were lost. According to NNs, in contrast, many of our conscious language-involving thoughts belong to types which (for us at least) constitutively involve public language. So if it is possible for us to entertain tokens of those thoughts non-consciously, those thoughts, too, will be expressed in public language.
Which of these two accounts is the better? The main point to note is that the cognitive resources and structures postulated by each of them in the normal case are different, and that considerations of simplicity favour NNs. For the Mentalese-involving explanation must suppose that conceptual resources are duplicated in normal cognition, with all our concepts being expressed both by public language terms and by lexical items of Mentalese. (Remember, it is presupposed here that NNw is correct, and thus that when we do entertain a thought consciously in public language, the use of those public language signs is the thinking, and also that tokens of the thought in question cannot be entertained consciously in the absence of public language.) NNs, in contrast, maintains that many of our concepts are tied to their public language expression only.
Recall that we have been taking it for granted, since section 1.8, that propositional thoughts are relations to sentences. So it is not open to the opponent of NNs to claim that there is but a single conceptual store, which may receive expression either in terms of Mentalese or in terms of public language. For according to our sententialist assumption, concepts are lexical items in some or other system of signs, which are caused, stored, and processed in characteristic ways. So in claiming that our thinking can be conducted either in Mentalese or in public language, the opponent of NNs (who nevertheless accepts NNw) is committed to the claim that we each of us possess two distinct, but semantically equivalent, conceptual systems. This claim appears thoroughly extravagant. Is there any way in which it might nevertheless be motivated?
I can conceive of just one possibility. It might be claimed that public-language thinking can, for some reason, only take place consciously. In which case it would certainly make good sense that all of our public-language concepts should be routinely copied over into Mentalese, so that we could enjoy the advantages of non-conscious as well as conscious thinking across all domains. If there exist non-conscious hypothesis-generators and problem-solving systems (as there appear to see the remarks about creative thinking earlier in this section), then there would surely be distinct advantages in ensuring that the full range of concepts, and hence thoughts, was made available to such systems.
The trouble with this proposal lies in its initial assumption, however namely, that language-involving thoughts can only be entertained consciously. For I can see no good reason why we should believe this. On the contrary, the hypothesis that natural-language representations can be processed and manipulated non-consciously as well as consciously would appear to be well-motivated. To maintain the opposite, one would have to insist that a sentence of natural language can only figure in cognition as a sentence with its full phonological properties represented in imagination, in such a way as to be available to reflexive thinking. But why should this be so? It is generally accepted amongst linguists working within a Chomskian framework, at least, that there is a level of language processing at which phonological properties are stripped away, leaving just syntax and semantics (this is the level of logical form, or LF). In which case, why should not those very representations be made available for use in central cognition, for purposes of non-conscious thought? Indeed, the anecdotal evidence suggests that they are. For creative non-conscious problem-solving can deal with matters which are overtly linguistic. If poets, novelists, and headline-writers can come up with their best turns of phrase, or their best language-involving ideas, in the absence of conscious reflection, as many reports suggest, then this must imply that natural language expressions are being tokened and manipulated non-consciously.
I conclude, then, that when we compare NNs with the Mentalese opposition, the former is seen to be much the more believable. In the end, it is more reasonable to believe that public language should be employed both for conscious and for non-conscious thinking, than it is to believe that we should employ two distinct, but semantically equivalent, representational systems.
The argument from the psychological evidence. A further set of arguments in support of NNs can be obtained by returning to the developmental and aphasic evidence of section 2.1, but armed, now, with acceptance of NNw. The first point to note is that the opponent of NNs must advance a variety of explanations of different aspects of the data, as follows.
(1) The fact that cognitive and linguistic abilities normally advance together is to be explained by claiming that public language, in human beings, is the channel of communication through which we acquire many of our belief systems, together with their embedded concepts.
(2) Ildefonso's difficulties with temporal discourse, and the fact that the play of the language-deficient twins studied by Luria and Yudovich became immensely more structured and creative within a few months of acquiring significant amounts of language, is to be explained by claiming that certain concepts are initially tied to the deployment of public language signs, but become available in Mentalese thereafter.
(3) The fact that global aphasics appear to have difficulties with conceptual thinking (e.g. difficulties in recognising that the activities of frogs and kangaroos are alike, in that both hop) is to be explained by supposing that both public language and Mentalese-involving concepts are stored in the same region of the brain, almost inevitably being damaged together.
The explanation advanced by NNs, in contrast, is the same in each case namely, that the phenomena in question arise because our cognition is so structured that certain kinds of concept, and certain types of thinking, constitutively require and involve public language. Now other things being equal, one explanation which unifies a diverse range of phenomena is to be preferred to a number of distinct explanations. So there is good reason to prefer NNs. Now consider the explanations (2) and (3) above in a little more detail.
(2) What the opponent of NNs has to say about the acquisition of temporal concepts, and of the logical concepts necessary for hypothetical thinking and for complex planning, is that these concepts are initially tied to the public language expressions, through competence in whose use they are acquired; but that they are later copied onto lexical items of Mentalese, in such a way that such thoughts can thereafter be entertained in the absence of public language. But this account is difficult to understand or make sense of. If we already have one set of signs whose use constitutively expresses for us a certain range of thoughts (as NNw maintains), then what would be the purpose of copying these into another set of signs to entertain the very same range of thoughts? What would be the point? What would be the gain? Again, to make sense of this one would have to maintain that natural language representations cannot be deployed non-consciously. It is surely more reasonable to believe that the capacity for entertaining those thoughts (whether consciously or non-consciously) should remain tied to their public language expressions.
(3) Consider, too, what the opponent of NNs has to say about the effects of global aphasia on conceptual thinking. Here the claim has to be that public language lexical items and their Mentalese equivalents are stored in the same regions of the brain. This now looks much less plausible in the light of acceptance of NNw. If public language had been only a medium of communication, then it would not have been entirely surprising, perhaps, that language should be stored in the same region of the brain as the bearers of the thoughts (in Mentalese) which it is standardly used to communicate. But if such language is also constitutive of the occurrence of many of our conscious thoughts, as NNw maintains, then the situation would be strange indeed. Why would the lexical items constitutive of our conscious thoughts be stored alongside, and in the same place as, the lexical items of Mentalese supposedly constitutive of our non-conscious ones, given that their overall roles in cognition would be so different? It is surely more reasonable to accept that public language sentences are constitutive of the occurrence of many types of thought (whether consciously or non-consciously), just as NNs maintains.
Taking together all the arguments in this section, then, I conclude that, with NNw already established, it is much more reasonable to believe NNs than to deny it. We should therefore accept that our cognition is so structured that many types of thought can only be entertained through the medium of public language, whether consciously or not.
8.8 The scope and significance of NN
What does seem to have been definitely established by the RT-based model of human cognition presented in this chapter (at least if that model is correct) is that much of human conscious thinking involves public language; and that this is so out of natural necessity. This is the thesis NNw. But I have now argued that, in the light of the likely truth of NNw, NNs may also be taken as established. I have argued, that is, that many types of thought (individuated by content) can only be entertained, by us, through the medium of public language (whether consciously or non-consciously). In this final section I shall comment on the significance of these results, as well as raising, somewhat inconclusively, the question of their scope.
The significance of NNw. The truth of NNw is already sufficient to show, firstly, that language is not just an isolated module of the mind, but is directly implicated in the distinctively human (conscious) aspects of the central functions of believing, desiring, and conscious reasoning. This is a decisive vindication of the cognitive conception of language, as against the communicative conception. And we can claim, moreover, that it is no mere accident that our conscious propositional thoughts make use of public language. Rather, this is determined, of natural necessity, by the very structure of human cognition. So the study of natural language is the study of a faculty which is essentially implicated in the central functions of the (human) mind, just as Chomsky has sometimes maintained. However, the truth of NNw alone is not sufficient to show that the language-faculty is constitutively involved in our systems of thought and belief as such. For that, we would need NNs.
As for the question of significance for philosophy, here, too, the truth of NNw seems already sufficient to gain for us much of what was advertised in section 1.2. For if philosophers choose to focus their attention on the expression of thoughts in public language, then it seems that they can be confident (in the light of NNw) that the structure and use of the linguistic expressions will mirror the structures of the thoughts (which are, of course, the true objects of interest). For those thoughts are public language sentences used and processed in characteristic ways. So the analysis of problematic concepts can proceed via consideration of their expression in language, and the 'linguistic turn' of analytical philosophy is established.
It might be argued, however, that if philosophers are concerned with the concepts and conceptual structures involved in any given thought [that P] as such, then there is scope for doubt about the appropriateness of the linguistic turn. For according to NNw, this thought, as a type, need not involve public language, and may be thinkable, non-consciously, without it. It is only those occurrent tokens of the thought which are conscious which essentially involve public language, if NNw is the most that can be established. So it might be wondered whether the thought as such could have an underlying structure which is only misleadingly captured by its dual modes of expression its expression, namely, non-consciously in Mentalese, and consciously in public language.
But this is a confusion. For according to the sententialist assumption adopted in section 1.8, thoughts are sentences in some or other system of representation. So if one thought can be expressed by sentences from two different systems (Mentalese and natural language), then this just means that those sentences are identical in respect of their functional or conceptual role, and also in respect of their truth-conditions. There is simply no room for the possibility that the thought in question may not be properly expressed in one system of representation or the other. This is not to say, of course, that a sentence cannot express its own content non-perspicuously. But this just means that features of the use of the sentence cannot easily be read off from its form. It does not mean that the content is something which can exist, and which can be investigated, independently of any sentence.
The scope of NNs. What is the scope of NNs? To what thought-types does it apply? Which types of thought are such that they can only be entertained at all, by us (whether consciously or non-consciously) through their expression in a public (normally natural) language? Note that there is no question of us answering, here, that NNs applies to all propositional thought, in the way that we did in connection with NNw above. For public-language thinking is not now to be contrasted just with thoughts expressed in private images, as it was when it was only conscious thoughts which were in question. Rather, there may well be types of thought which can be entertained non-consciously in a form of Mentalese; and then these would count as properly propositional.
Recall that the basic argument in support of NNs turned on the implausibility of supposing that all concepts and conceptual systems are duplicated in cognition, finding expression both in public language and in Mentalese. We suggested, rather, that public language would probably do double duty for both conscious and non-conscious thinking. But from the fact that not all concepts are duplicated it does not follow that none are. So we cannot yet conclude that all non-imagistic thought-types which are available to us in public language will be essentially public-language involving. There may well be particular domains in which, for one reason or another, our concepts are expressed both in natural language and in Mentalese. These would be domains which human beings are capable of representing, and entertaining genuinely propositional thoughts about, before the arrival of public language.
The most plausible general hypothesis, I think, is that public language is essentially implicated in all thoughts whose constituent concepts are dependent upon language for their acquisition. We can conclude, in fact, that all concepts which require public language must also involve public language. For if certain concepts require public language, in the sense that they can only be acquired through language-learning and/or the acquisition of information and new beliefs from other people, then by NNw the conscious tokens of those concepts will necessarily involve public language. And then our general argument for NNs would suggest that the non-conscious tokening of those concepts will also be undertaken in public language, rather than being duplicated into Mentalese. And so we get the conclusion that those concepts can only be tokened at all in public language, which means that they fall within the scope of NNs those concepts determine a class of thought-types which (for us) necessarily involve public language; and if we were to lose public language, then we would lose our capacity to entertain thoughts of those types.
There is surely no doubt that the class of such language-dependent concepts is very large indeed, including concepts from mathematics and the natural sciences (prime number, gene, electron), concepts from the social sciences (inflation, nation-state), as well as a whole range of perfectly ordinary concepts, including marriage, law, permission, school, writing, story, drama, employer, wage, holiday, bus-stop, ticket, generosity, and many, many, more. (And note, here, that to say that a concept is language-dependent is not necessarily to deny that it is innate see below.) In fact, it is no exaggeration to say that most human concepts and thoughts will turn out to involve public language essentially, if the arguments of this chapter are sound.
Roughly speaking, then, to look for evidence of the exact scope of NNs is to look for evidence that certain thoughts have constituent concepts which are dependent upon public language for their acquisition and/or for their continued existence. Thus if the acquisition of temporal concepts is dependent upon language, as the evidence provided by Schaller's pre-signing deaf man Ildefonso would suggest, then NNs will imply that such concepts are necessarily language-involving. And if the sorts of general beliefs which go to make up our concepts of natural kinds (e.g. 'Kangaroos hop', 'Kiwis cannot fly', and so on) are dependent upon language, as some of the evidence from global aphasia suggests, then these concepts, too, will be essentially language-involving.
Conversely, evidence that certain types of thought fall outside the scope of NNs will be provided wherever there is evidence that the constituent concepts of those thoughts might have evolved, or might be constructable, prior to language, and so be independent of it. Thoughts about the motions and interactions of middle-sized physical objects (folk-physics), and thoughts about animal and plant kinds (folk-biology) might be plausible enough examples. (See Carey, 1985.) These thoughts, in going beyond what could be represented purely imagistically, would require some sort of Mentalese to be in place prior to language. But when language is then acquired, the constituent concepts would presumably be copied into the public-language medium, and the very same thoughts would then become available for expression (consciously) in public-language form. Since conscious thinking is useful, it would be easy enough to understand why such copying-across should take place.
In fact, there is now evidence that infants and pre-linguistic children have at least some understanding of a quite remarkable range of subject-matters, including causality and an intuitive mechanics, action and agency, and number (see Karmiloff-Smith, 1992, for reviews). But here we have to be careful to ensure that what is being displayed is genuinely thought about these subject-matters, and not merely discriminations of, or sensitivities to, them. We need to be mindful, in fact, of the distinction Annette Karmiloff-Smith draws between knowledge which is merely implicit, perhaps embedded in some practical procedure, and knowledge which is explicit, available for general use in cognition, and for interactions with centrally-stored information outside of that procedure. With this distinction in place, the evidence of genuine (explicitly represented) thought about the above subject-matters, prior to language, is extremely meagre.
Even evidence that a certain body of beliefs and concepts is innate does not show that the thoughts in question are independent of language. For the evolution of those beliefs may have been closely inter-woven with that of language. Our folk-psychology may serve as an example here, since I have already argued (in section 1.7) that it is probably innate. For there is also evidence, briefly cited in section 2.1, that language-ability and competence in theory of mind tasks develop together. And some researchers have begun to speculate that the crucial concepts of belief, and of false belief, can only be underpinned by language (see Harris, 1995, and Smith, 1995). Put in modular terms, the hypothesis is that a primitive form of desire-perception psychology may pre-date the arrival of language, both in evolution and in child development, but that fully-intentional belief-desire psychology may have had to await the beginnings of language and linguistic communication.
It may well be, then, that the development of explicit (genuinely thoughtful) knowledge about many domains is dependent upon language even where that knowledge is innate or has a substantial innate basis. So the evidence, cited above, that humans have domain-specific and innately constrained knowledge of such matters as folk-physics, folk-biology, and folk-psychology, does not show that it is possible for us to entertain the thoughts in question independently of language. This remains an open question. It may be, in particular, that some relevant language-experience is a crucial trigger for the innate knowledge to develop beyond the merely implicit level. Perhaps such cognitive modules can only grow and operate if they can co-opt some public signs to fill the innately-determined concept slots within the module.
A research-programme for NNs. Of course the question of the exact scope of NNs is, ultimately, an empirical issue; and a detailed answer to it would require us to consider, and gather, a variety of kinds of developmental and neuro-psychological evidence. Here just let me mention some of the kinds of evidence which appear to me to be most germane, and also say something briefly about the difficulties which stand in the way of us gathering it.
First, it would be particularly worthwhile to study congenitally deaf children who have had no exposure to conventional language, to see the range of thoughts which might nevertheless be available to them. For although, as we have seen, such children are probably not entirely languageless, we can be fairly confident that they have no language abilities beyond those that they overtly express. (Since they construct their languages for themselves, they are unlikely to be in the position of young normal children who can understand more than they can say.) So wherever we find evidence that they possess concepts which do not find expression in their invented sign-languages, there we may conclude that the concept in question is independent of language. It would be interesting to see what explicit knowledge such children have of the domains of folk-physics, folk-biology, and folk-psychology, in particular.
There are two main difficulties which stand in the way of such studies, however (aside from the practical problem that such children are becoming increasingly rare, as the almost total ineffectiveness of lip-reading for congenitally deaf children with no prior knowledge of language becomes generally recognised). The first is the difficulty of experimentally discriminating implicit from explicit knowledge. Only the latter will show the existence of language-independent thought. The second is the difficulty of devising non-linguistic tests of knowledge and understanding. As has become clear to those working in the area of theory of mind, who have been attempting to develop non-language-dependent tests of false-belief understanding for use with apes and very young children, this can be an extremely difficult thing to achieve.
Second, it would be worthwhile to study what happens to such pre-signing deaf children if they are then immersed in a signing environment, at age 6 or 7, say. Since such children would be developmentally fully-ready for language, the prediction will be that they would acquire Sign extremely fast. And then it would be interesting to observe the changes which take place in their cognition as a result. If there are kinds of thought which did not appear to be available to them before, but which become available consequent upon acquisition of full-blown language, then this will be evidence that thoughts of those kinds constitutively require public language. (In effect, this would be to re-run the Luria and Yudovich twin-experiment.)
Third, we need to continue to study individuals with global aphasia, to see what kinds of thought can be spared, and which are inevitably lost with the loss of language. One problem here is that of knowing whether the aphasia is genuinely global. Perhaps in the future it may be possible to use a combination of knowledge of neural anatomy and advanced brain-imaging techniques to ascertain that all the language systems have been completely destroyed, and not just the input and/or output systems. Another major obstacle is, again, to develop tests for the various types of thought which are independent of language. But provided these problems can be overcome, we might discover that there are kinds of thought which can be spared in global aphasia, and so which are not necessarily language-involving, and kinds which are inevitably lost, in which case we can conclude that they are language-involving.
Let me stress again, however, the importance of the implicit/explicit distinction here. It is not sufficient, if we wish to establish that certain kinds of thought can be entertained non-consciously independently of language (hence restricting the scope of NNs), to show merely that patients whose aphasia is genuinely global can engage competently in certain practical activities. For the most that this would show is that they retain an implicit knowledge of the domains those activities concern knowledge which may be embedded in their practical procedures. It would not show that they are genuinely capable of thinking (explicitly) about those domains. So once again we face the problem of distinguishing empirically between implicit and explicit knowledge.
In fact, I would predict, public language may well be required for any kinds of thought which are distinctively human, which are not available to other species of animal. Indeed, since the cognition of the average human child in its second year seems rather less sophisticated than that of many species of animal, it may be that human cognition is so structured that even relatively simple types of thought which in other species receive expression in some form of Mentalese require public language in the case of human beings. But I have to admit that I have no account ready to hand of the kind of complexity that marks the divide between those types of human thought which are essentially language-involving, and those which are not.
The significance of NNs. I have argued that, in the light of the likely truth of NNw, NNs may also be taken as established. I have argued, that is, that many thought-types (presumably those, in particular, of the more complex and abstract variety) can only be entertained, by us, through the medium of public language (whether consciously or non-consciously). This would then provide philosophers with some further guarantee of the methodological soundness of the linguistic turn. Since many of the thought-types which we entertain consciously in public language are essentially language-involving, it must follow that in studying the use of the public signs we are studying the thoughts themselves. Indeed, the concepts which tend to attract the attention of philosophers, and which tend to be philosophically especially problematic, are ones which are distinctively human, and would presumably count as 'complex and sophisticated' and so essentially language-involving according to the above account. (Think here of concepts like causal necessity, personal identity, objective truth, and so on but presumably not indexical concepts like this, or here.)
As for significance for scientific psychology, the truth of NNs would mean that the language faculty is yet even more deeply embedded in, and partially constitutive of, central cognition. Not only would conscious propositional thinking essentially involve natural language sentences, as NNw maintains, but the thoughts so entertained (together with their embedded concepts) would consist in relations to natural language expressions. Thus, not only would the language faculty be employed in central cognition, but it would bring with it many of the concepts and conceptual structures which are distinctively employed in human conscious thinking. On this picture, it really would be only an exaggeration to say that the study of language is the study of the mind.
In this final chapter I have outlined an architecture for human conscious thinking which assigns a central role to language, and have argued that this provides a better account of the place of public language in our cognition than do the alternatives. I have also argued, on the basis of this account, for two types of natural necessity thesis. The weaker version of this thesis maintains that propositional forms of human conscious thinking necessarily involve public language. The stronger version claims that many of the thought-types so entertained are themselves essentially language-involving. I have thus suggested that it may be naturally necessary that distinctively-human (complex) thinking should involve public language. But as for the exact scope of this strong form of natural necessity thesis, I have had to leave this largely undetermined.
(Note, however, that both forms of thesis NN are contingent on a creature's possession of the sort of human cognitive architecture depicted in Figure 3, and so do not extend beyond it. Thus Stalnaker's intelligent and conscious Martians remain a possibility.)
Wittgenstein famously wrote, in the preface to the only book he was to publish in his lifetime (the Tractatus Logico-Philosophicus, 1921), 'May others come and do it better'. I have to say, much more modestly, 'May others come and do it'. For the real work is yet to be done. I can only claim to have been pointing in the right direction.