Consciousness and Language

If the mind is computational, then there must be a language of thought. This is because computations are operations defined over symbols: thought processes must be symbolic processes.   The original conception of a Turing machine makes this very clear, since the operation of writing and erasing marks on a tape is obviously a process defined over symbols (ones and zeros, or numerals thereof). But the generalized notion of a computation, which involves any procedure of symbolic manipulation, will also entail that there is a language of thought, granted that thought processes are (or involve) computational processes. Thoughts will consist of, or be encoded in, sequences of word-like elements upon which operations are performed by something like a computer program. We have good reasons to believe that the mind is computational (which I won’t rehearse), so we have good reasons to believe in a language of thought. Moreover, the hypothesis of a language of thought helps deal with various conceptual and empirical problems.

            And yet there is a marked reluctance to accept it. Why? I think the main reason for this reluctance is that we do not perceive the language of thought (call it LT). We do not see or hear the words that compose it; neither do we introspect these words. By contrast, we do see and hear (and sometimes introspect) the words of our ordinary public language (call this LC for “language of communication”).    [1] That is, we are conscious of LC but we are notconscious of LT. I am now looking at words of my LC as I write, but as I think I am not in any analogous cognitive relation to the words of my LT. The language of thought is an unconscious language—not repressed, to be sure, but deeply hidden, inaccessible to consciousness. Why LT should be hidden is an interesting question, with no obvious answer, but it evidently is (granted its existence). It exists in the mind, entering into mental computations, but it does not exist in consciousness. It affects the conscious mind, leading us to form conscious thoughts we couldn’t have without it, but it never reveals itself to consciousness—it hovers in the background, invisibly and inaudibly. Thus we have no direct knowledge of its existence; its existence is inferred. We do not believe in a language of thought because of the manifest contents of consciousness, as we believe in thoughts, emotions, and sensations; we believe in it in the way we believe in atoms or genes. In fact, no one has ever caught a glimpse of LT and it remains elusive—we don’t even know what its lexicon looks like. It must contain symbols with an arbitrary connection to what they denote and with their own internal composition, but we don’t know what these symbols are—we don’t know the alphabet out of which the lexicon of LT is constructed. This epistemological situation is surely a large part of the reason that people doubt that a language of thought exists—no one has ever seen or heard the words that compose it. By contrast, we have no trouble accepting that LC exists, because we have all seen the words of LC: we are conscious of the words of LC in multiple ways—by seeing them written down, hearing them spoken, producing them in speech, and introspecting them in inner speech. It is not that we have merely theoretical reasons to believe in LC; we have direct evidence that LC exists. Because of this epistemic asymmetry we are apt to be skeptical of LTand accepting of LC.

            Now it is not that this asymmetry constitutes a good reason to believe in LC but not in LT, on pain of doubting every theoretical entity we have ever had reason to posit; and there is ample precedent for the notion of unconscious mental realities. But as a matter of human psychology people have a tendency to believe in what they can sense and to doubt what they cannot. To be conscious of something is always the strongest reason for believing in it. The point I want to make is that the epistemic asymmetry is entirely contingent and should not be interpreted as having ontological significance; it is merely epistemological. I aim to loosen the hold of the conviction that LT is suspect while LC is pucker. I shall do this by inverting the epistemology of the two languages: in a conceivable case we could be conscious of the language of thought but not conscious of the language of communication. And just as in such a case we should not interpret the epistemic asymmetry as having ontological significance, so we should not do so as things actually are with us. It merely happens that we have direct evidence of LC and not of LT; there is nothing metaphysically deep here. It might have been otherwise. It is not of the essence of LT to be unconscious and of the essence of LC to be conscious; the epistemic asymmetry is not built into their very nature.

Consider then the following imaginary case. First we suppose that LT is conscious for certain possible thinkers; for concreteness suppose that LT is Latin (once the language of learned European thought). Whenever these possible thinkers think, sentences of Latin surge conspicuously through their consciousness, encoding their thoughts. They have inherited this language from their ancestors and it is built into their genes (don’t ask me how). Mental computations are performed on Latin strings and the resulting thoughts contain words of Latin. Maybe these strings enter consciousness in some sensory mode—auditory images perhaps, or possibly visual. Just as we can introspect our conscious inner speech in our acquired language, so they can introspect their innate language of thought. This seems perfectly possible. It will then not be a difficult task to persuade them that they have a language of thought: that is evident from their everyday consciousness and everyone agrees about what this language is like. It is quite evident to them that they think in Latin; they have direct knowledge of this fact. There are no debates about whether such a language exists or what composes it; it is as clear as conscious thought itself. They can in good conscience avail themselves of the theoretical benefits of positing an LT—except that they don’t posit it, since they are acquainted with it. It might be that their spoken language is derived from their internal language—they speak Latin because they think Latin. They have molded their LC around their LT, quite deliberately; or perhaps evolution has taken a short cut to producing a communicative language by simply co-opting the language already coded in their genes. Given their epistemic relation to their language of thought, there will not be much controversy about invoking it in theoretical contexts.

            But what about LC and consciousness—how could its existence be a matter of conjecture, inference, and speculative positing? Aren’t our possible people bound to know it directly? Perhaps surprisingly, the answer is no. LC could be inaccessible to consciousness, just as LT is for us. For consider speakers with a kind of linguistic blindsight: when sentences are uttered in their presence they have no conscious impression of sound—it is as if they are deaf—but still the auditory input is processed by the brain, producing a belief about what was communicated. They know that the speaker said that p by producing certain sounds, but they have no consciousness of those sounds—they are processed unconsciously. Thus they have no consciousness of the words that compose their LC: all the work of hearing, parsing, and comprehension is done behind the scenes, with only the output reaching conscious awareness. We can suppose that no one has ever had a conscious percept of the words of their LC, their sound or shape, while they have all had conscious awareness of their LT. What they have in respect of LC is deafhearing: they hear (process) the words unconsciously, but they are deaf to them consciously. What they hear in this mode is inaccessible to consciousness. It is just like regular blindsight with respect to written words: someone might direct their eyes at written language and have no conscious impression of what is before them, yet the visual system might be able to process the stimuli in such a way as to produce knowledge of the inscription in question. They might engage in “blind-reading” through their eyes. Such a person could not tell you what the words they see (process) look like, since they are never conscious of these words, but the words exist and the person responds to them. Similarly, they might be able to produce spoken strings and yet not be conscious of their actions: it is all done behind the scenes with no awareness of what sounds are being produced. We are not normally conscious of the fine structure of our actions and many animals presumably lack consciousness of their motor activity (consider insects); well, these speakers lack any conscious knowledge of their speech acts—though they perform them perfectly adequately. In sum, they are not acquainted with their LC: it exists and they use it, but they are not conscious of its intrinsic character—its alphabet, phonetic properties, intonation patterns, etc. It is, as far as they are concerned, an unconscious language. We might compare it to the communication system used by bees: I don’t know whether bees are conscious (I rather think they are) but we can suppose that they are not, and then their language is an unconscious language—no bee is conscious of the words that compose its language. That doesn’t prevent bee language from existing. Bees just happen not have any conscious awareness of the language they employ. There is no necessity for a public language of communication to be conscious.    [2]

            Given this description of my hypothetical language users, what is their theoretical position vis-à-vis LT and LC? They invert our ontological prejudices. They have no scruples about believing in a language of thought, but they have their doubts about a language of communication. They may allow that an LC must exist, given that they communicate and understand each other, but the lack of direct evidence troubles them, and some may be openly skeptical. Some may make a living out of denying that there is a language of communication at all. After all, no one has ever directly observed LC, and the idea of an unconscious language strikes some as inherently problematic. They have no trouble with LT, however, since it is so evident to consciousness—as certain as thought itself (“I think in Latin” is indubitable for them). They don’t feel the need to prove the existence of LT, but they do think that the hypothesis of LC cries out for some sort of defense—perhaps arguing that it is entailed by a computational view of communication. But we can see from the outside that they are overly impressed by their own epistemological biases, reading far too much into them. We know that their LC exists, even if they doubt it. Just because they are not consciously aware of the words of LC doesn’t mean that there are none or that there is anything metaphorical or suspect about the notion of a language of communication. This is just a contingent fact about their epistemic powers. Similarly, it is just a contingent fact about our epistemic powers that we are conscious of our public language but not of our private language. There is a hidden language of thought—there has to be if the mind is computational—but we don’t happen to be conscious of it. Maybe we are indirectly conscious of it, because we are conscious of the work it does, but we are not conscious of it in its entirety. In the same way, my inverted speakers may have glimmerings of the public language they use because of their awareness of its effects in their consciousness, but they are (by hypothesis) not aware of the full nature of LC. We would think them irrational, or excessively cautious, if they were seriously to doubt the existence of a language of communication; but then why are we being commendably skeptical about the existence of a language of thought? If we have solid arguments that there must be an LT, why should its inaccessibility to consciousness count against its existence? Why should a language we use necessarily be a language of which we are conscious? Perhaps there are good biological reasons that we are not conscious of our LT  (it would clutter up our attention or be too costly metabolically), but these are not reasons to doubt its existence. After all, we are not conscious of a lot of what goes in with our ordinary spoken language too. In the case of bees it may well be that their communicative language has an internal counterpart that helps them navigate the world—a language of bee thought—and it too will be an unconscious language. Bees may have both an LC and an LT with neither of them conscious, simply because bees are not conscious–or whatever smidgen of consciousness they possess is not directed at their linguistic skills.

            I end with this thought: perhaps the reason we are not conscious of the language of thought is that it is just too abstract to enter our consciousness. The symbols in LT are universal to the human species, so they are less specific than the symbols of a spoken language (or a sign language). They are components of an abstract computational structure. Just as we cannot have in mind an image of a triangle in general, but only specific types of triangle, so we cannot have in mind a word belonging to an abstract universal language, but only words belonging to a particularized language. Consciousness will not admit elements of the requisite generality and abstractness: for how could such elements be contents of consciousness? We can have concepts of abstract things, but it is harder to see how the conscious linguistic medium of concepts could itself be abstract. Consciousness filters out the abstract, admitting only the concrete, while the unconscious tolerates the abstract. Why this should be I don’t know, but it seems like a possibility. Certainly it is a good question why the language of thought should be so removed from our consciousness.

    [1] I say this for the sake of argument; actually it is not clear that we do perceive our language of communication. We perceive various sounds and marks with the senses, but it is another question whether we perceive words and sentences. These are more abstract categories than sounds and marks, more deeply psychological; and it may be that we bring these categories to external stimuli rather than perceiving them there. Indeed, it may be that we never really perceive (with the senses) words and sentences—we infer them from what we do so perceive (or impose them). So the language of communication may not as accessible to consciousness as we think.

    [2] We need not suppose that bees engage in acts of speaker meaning, as characterized by Grice, which does require consciousness; it suffices for the point that they employ a system of communication whose nature is not given to them.

                                    Fixed and Variable Semantic Value

The orthodox view is that names are rigid designators and descriptions are (typically anyway) non-rigid designators: denotation varies from world to world in the latter case, but not in the former. What about connotation? The connotation of a description is fixed, even as it determines variable denotation from world to world. In the case of names, as Frege pointed out, connotation can vary from speaker to speaker and from time to time: speakers can have different individual concepts in mind, even though the same object is denoted. We use “London” to designate one thing, and the name is rigid across worlds, but what concepts are in our minds can differ, according to how we conceive of London (“the capital of England”, “the center of the music business”, “the place where parliament is”, etc). So the denotation is fixed (across worlds) while the connotation is variable (across minds): a name is a rigid designator but a non-rigid “expressor”. By contrast, a description has a fixed connotation, since the same individual concept is associated with the description from mind to mind: we all mean the same by “the tallest man”. The intension is fixed, while the extension varies–whereas for names the extension is fixed and the intension varies.  

            Thus names and descriptions semantically invert each other: what is fixed for one is variable for the other—yet each is fixed in one respect and variable in another. The name has a fixed reference and a variable sense, while the description has a fixed sense and a variable reference. This is the difference between names and descriptions. It is not merely that names are rigid and descriptions are not; names are also non-rigid (with respect to sense) and descriptions are rigid (with respect to sense). Each has a fixed semantic value and a variable semantic value, depending on whether we are considering denotation or connotation (extension and intension, reference and sense).

            If we consider Frege’s view of indirect discourse the inversion is complete. A name in indirect discourse will designate its ordinary sense, but this can vary from speaker to speaker, or from time to time. On the other hand, a description in indirect discourse will designate, not its variable reference, but its fixed sense. So in indirect discourse a name is a non-rigid designator and a description is a rigid designator, following Frege’s theory. Now if we take the semantic functioning of expressions to be basically determined by their behavior in indirect discourse, it will turn out that names are basically non-rigid and descriptions are basically rigid—with respect to their sense, that is. It is only if we take ordinary objects to be basic designations that the orthodox way of describing things holds. We could describe things in the alternative Fregean way too, thus inverting the fixed-variable distinction. The fact is that both sorts of expression can rightly be said to be fixed with respect to one thing and variable with respect to another.

            We can object to the description theory of names by arguing (as Kripke did) that descriptions are (typically) non-rigid while names are rigid; but we can also object that names are non-rigid with respect to sense while descriptions are rigid with respect to sense. We can argue, that is, that names could not be equivalent to descriptions, because names are non-rigid expressors, while descriptions are rigid expressors, which puts them in different semantic categories. The variability in the sense of a name is incompatible with the name being equivalent to a single description—just as the variability in the reference of a description is incompatible with it being equivalent to a name. The description theory of names fails both because of misplaced attributions of referential non-rigidity and because of misplaced attributions of expressive rigidity. Names and descriptions are rigid (fixed) and non-rigid (variable) in opposite ways, so they cannot be semantically equivalent.

            In addition to the description theory of names, there is the name theory of descriptions, i.e. the idea that a description functions like a name (Frege, Meinong, early Russell). Thus it may be said that a description can always be replaced by a name without change of meaning, since it is a name. It is a name with both sense and reference, but still essentially a name. But such a theory can be refuted by pointing out that descriptions don’t have variable sense in the way names do—they are rigid expressors. They are also non-rigid designators, which makes them different from names in another respect. So the name theory of descriptions suffers from two sorts of difficulty: first, it gets the relation to sense wrong, by implying a variability of sense that just isn’t there; and second, it gets the relation to reference wrong, by implying a fixity of reference that just isn’t there.

            The point about names being rigid designators and descriptions being non-rigid designators is thus part of a larger semantic picture, which undermines the alleged equivalence of names and descriptions from several angles. It is not wrong, just partial. The key point is that just as the denotation of a description can vary from world to world, so the connotation of a name can vary from mind to mind (or within the same mind over time). To put it simply, though somewhat inaccurately, names are inherently “ambiguous” while descriptions are not—that is, they have variable sense. The reason this is somewhat inaccurate is that the variability of sense or connotation is not incompatible with acknowledging a basic unity of meaning, since denotation is shared across differences of sense (and we don’t need to suppose that sense determines reference). What is important is that descriptive sense is uniquely fixed for descriptions, but not so for names. The underlying reason for this is that speakers need a way of conceiving their reference, and names themselves cannot supply such a way—so the speaker needs to come up with a way that suits her. The speaker’s information about the name’s bearer accordingly varies. But in the case of descriptions the information is right there in the description to begin with, needing no supplementation. For descriptions reference varies with world, while what is in the mind stays constant; for names sense varies with mind, while reference stays constant across worlds. Hence names and descriptions are fundamentally distinct semantic types.

Language and Reality

Consider the following thesis: objects are essentially nameable and properties are essentially predicable. That is, objects can only be named not predicated, while properties can only be predicated not named. To put it differently, objects can only be denoted from subject position, while properties can only be denoted from predicate position. This thesis asserts a strong connection between language and reality: namely, that an ontological distinction maps rigidly onto a linguistic distinction. It might be called “the mapping thesis”.

            Is the mapping thesis true? It is difficult to see how it could fail to be true for objects and names: how could an object be predicated of another object, or of a property? What would a predicate for an object be like? You might think that “is Aristotle” is such a predicate, but in fact it does not denote Aristotle but rather the property of identity with Aristotle. An invented Quinean predicate like “Aristotle-izes” likewise really means (if it means anything) something like “has the individual essence of Aristotle”, which stands for the property of having a certain individual essence. There are no predicates that denote particular objects, because there is no sense in the idea that an object could be ascribed to another object. Objects are not predicable things—any more than they are things that can be instantiated. We predicate properties of objects, but objects are not themselves predicable. This is why we don’t have sentences of the form “Plato(Aristotle)”.

            But the mapping thesis is not so obviously true for predicates and properties: for what is to stop us from naming a property and then predicating something of that property? Don’t we have sentences like “redness is a color” or “democracy is good” or “the property of being soluble is commonly instantiated”? These look every inch like subject-predicate sentences in which the subject term denotes a property, which is then made a subject of predication. So properties can be named, even if objects cannot be predicated: properties are not essentiallypredicable–they can be predicated or named.

One way to resist this breakdown of the mapping thesis would be to claim that all such sentences can be analyzed into sentences that do not name properties but only predicate them, as in “objects that are red are colored” or “it’s good for countries to be democratic” or “there are many soluble objects”. The ease of such paraphrases confirms the thesis that properties are always denoted by grammatical predicates in logical form. But still, there are sentences that contain nominative terms for properties, while there are none that contain predicative terms for objects. It is not then nonsense to form a sentence that names a property.

            A further response is to note that these alleged names contain predicates for the property in question, which occur in their usual ascriptive mode. Thus “redness is a color” means “the property of being red is a color”. Sometimes we produce a name for a property simply by using italics, as in “red is a color”, but this is just a conventional way to abbreviate the definite description “the property of being red” or “the property an object has in as much as it is red”. In these locutions “red” is functioning predicatively, not as a name. What we have are grammatically nominative expressions for properties that contain predicates that ascribe those properties in the usual predicative style. It is just that we seem to be able to name properties by using embedded predicates for those properties that are functioning predicatively. But these are really very odd names, being entirely derivative on predicative ways of denoting properties; they are not direct names of properties, capable of semantic independence from predicates used in the standard predicative way. They are not proper names of properties, and they cannot be understood without reference to the predicates from which they are built. They are actually descriptions, and hence akin to quantifiers: if we apply Russell’s theory, they are second-order quantifiers ranging over properties. Thus “the property of being red” means something like “there is a unique property P such that P is being red”.  That is an odd construction, with its peculiar use of “being red”: for what is being red except an object’s having the property of being red? We can’t get rid of the predicate from the putative name (description)—we can’t refer to the property without predicating it in some way. We can only name it by predicating it. The property resists being named in such a way that it breaks free from being predicated. So we might reformulate the mapping thesis as follows: properties are essentially predicable in the sense that there is no way to specify a property except by predicating it at some point. Properties can only be specified by at some point using expressions in predicate position, even if these occur inside syntactic singular terms, i.e. descriptions. There are no names for properties that work like ordinary names or descriptions, by referring to things without simultaneously predicating those things. It is as if the singular term is admitting: “Predication is really the only proper way to refer to a property—I am cheating by relying on that way”.  [1]

            The mapping thesis is therefore fundamentally correct: there is a tight correspondence between the ontological categories of object and property (particular and universal), on the one hand, and the logico-grammatical categories of subject and predicate, on the other. Objects must be named not predicated, and properties must be predicated not named (that is, the basic mode of reference to properties is via predicates). The link between language and reality is thus not arbitrary: the nature of reality (its metaphysical structure) is reflected in the nature of language (and thought). There could not be a language that named properties (primitively) and predicated objects. A property is the kind of thing that in its nature calls for expression by means of a predicate (it is essentially ascribable), while an object is the kind of thing that in its nature calls for expression by means of a name (it is essentially nameable). The way we speak of reality is dictated by reality in this basic respect. Grammar recapitulates ontology.

  [1] Someone might suggest that we can use a description like “the color of that cup” to name redness, and hence avoid using the predicate “red”. But that is not a rigid designator of the color red (unlike “the property of being red”) and can be understood without knowing that the color referred to is red. Only descriptions like “the property of being red” succeed in identifying their referent, and they employ a predicate of red. What there could not be is a language that contains rigidly designating directly referential identifying names for properties that are not parasitic on predicates for those properties.

                                    Blushing, Sneezing, Coughing, and Spitting

Blushing is involuntary: you can’t blush intentionally or intentionally suppress a blush (though you can of course undertake a course of action that will have such results). Blushing is not an action, as philosophers say; it is not “subject to the will”. Sneezing is similar in that you can’t sneeze voluntarily: you can’t decide to sneeze and do it as a result. No amount of effort of will can produce a sneeze (a real sneeze not a counterfeit sneeze). However, you can suppress a sneeze—you can stop yourself from sneezing. You can try not to sneeze when you feel a sneeze coming on, and sometimes you succeed (compare hiccupping). So sneezing is partly voluntary and partly not. Coughing is yet more voluntary, though not completely so. There are coughs you can’t suppress and coughs you can suppress, and you can cough intentionally. Coughing runs from the reflexive to the calculated: from the coughing fit to the polite ahem. Coughing is rather like laughing: they can both be involuntary or voluntary, uncontrollable or controllable. Spitting is fully intentional: all spitting is voluntary and can be suppressed if an urge to spit comes on (unlike salivating, which is like blushing). You can always hold a person responsible for spitting, but not for blushing, sneezing, and coughing. Spitting is like stealing or slapping—entirely subject to the will.

            What, then, of the sharp dichotomies that prevail in the philosophy of action? We are invited to determine whether an act (token or type) is voluntary or involuntary, intentional or unintentional, an action or a happening, willed or unwilled. To be sure, there are cases where the dichotomy works—say, blushing and spitting; but there are also cases where it doesn’t—say, sneezing and coughing. Even a particular cough or sneeze may not easily fall into one category or the other: could that sneeze have been prevented, was that cough partly intentional? There is a whole range of intermediate cases here: in addition to the intentional (and sub-intentional) we have the semi-intentional, the partially intentional, the somewhat intentional. We have the amplified spraying sneeze as well as the stifled minimal sneeze, as we have the loud ostentatious cough as well as the barely audible demur cough. Philosophers adore dichotomies, but here is a field in which they oversimplify considerably. The question should not be, “Was that intentional?” but “How intentional was that?” And the same point applies to the concept of intention itself: you can’t intend to blush or sneeze and you can intend to cough or spit, but you can also intend not to sneeze loudly (though not to blush deeply). You can intend to cough when you have a cough and the cough comes out more forcefully than you intend: how intentional was that cough? The idea of a sharp boundary between “actions” and “happenings” looks misguided in the light of the full range of cases.