If one asks about the relation between thought and language, people expect the issue to concern such matters as whether we think in language, whether creatures without language can "think", and the way language shapes our concepts. In my opinion, there is a much deeper question, which concerns the nature of linguistic communication. Philosophers and linguists standardly conceive of language as basically a means by which speakers convey the content of their thoughts to others. The question is whether that is a correct picture of linguistic communication. This is a question about the relation between thought and language because this standard picture of communication gives propositional thought a certain priority over language. If, as I intend to show, there are reasons to doubt the standard picture, then we cannot expect to make much progress with the more superficial questions without thinking about the nature of linguistic communication.
My plan in this article is as follows. In section 2, I will characterize the
standard picture in more detail. In section 3, I will highlight some of the
issues that can be raised within the standard picture among its adherents. In
sections 4 and 5, I will detail what I take to be the primary liabilities for the
standard picture. In section 6, I will return to the shallower issues usually
associated with our topic. Finally, in section 7, I will characterize a
neglected conception of the dependency of thought on language.
2. The standard picture: The expressive theory of communication
My label for the standard picture will be expressivism. Briefly, the expressivist holds that the primary function of language is to enable speakers to convey the content of their thoughts to hearers. Less briefly, when language is used in the normal way, the speaker has a thought with a certain content and chooses words such that on the basis of those words the hearer will be able to recognize that the speaker has a thought with that content.
For example, suppose that I see you are about to walk out behind the house and I know that there is poison ivy back there. I ask myself, "How can I get you to believe that there is poison ivy behind the house?" I reason that if you believe that I believe that there is poison ivy back there, then you might believe that too (since you might trust me). I reason, further, that if I say, "There is poison ivy behind the house", then you will recognize that I believe that there is poison ivy behind the house. Why? Because I know that you have the ability to interpret my words, that is, to infer my thought on the basis of my choice of words.
It is not at all plausible that we are regularly conscious of such thought
processes, and the assumption that they are conscious is no part of the
expressivist's thesis. Further, the expressivist does not have to hold that the
expression of thought in this way is the only possible use of words or even
that it is the most common. The expressivist might treat commands,
requests, questions, lies, jokes, and poems quite differently. The idea would
be only that the use of words to express thoughts in this way is normal or
fundamental in the sense that this is what we must look at if we want to
understand such things as: the structure of sentences, broadly speaking; the
possibility of language learning; the possibility of the evolution of language.
In order for a speaker to express a thought in this way, it is necessary for the speaker and hearer to share an understanding of language. There has to be a kind of knowledge of language that the hearer will apply in inferring the content of the speaker's thought from the speaker's choice of words and that the speaker can count on the hearer to apply. There is a tacit consensus among expressivists that the process of inferring the content of the speaker's thought has three aspects or phases: 1. The identification of the meaning of the words used. 2. The identification of the proposition expressed in light of the meaning and the rest of the situation in which the utterance takes place. 3. The identification of further implicatures over and above the proposition expressed.
For example, suppose that a teacher enters a classroom, looks around and declares, "Everyone is present". Taken out of context, this sentence does not express any particular proposition, because, taken out of context, there is no particular domain of discourse relative to which we may interpret "everyone" and no particular time and place that "present" might refer to. Nonetheless, the sentence, as a sentence of English, carries a certain potential for expressing propositions, and this potential is, in one sense, its meaning. The meaning of this sentence is such that in this particular setting, an utterance of it might express the proposition that all of the students enrolled in the course are at the time of utterance located in the classroom where the utterance takes place; whereas there is no setting in which it might express the proposition that there will be no lecture on that day. Beyond the proposition that the utterance literally expresses in light of the meaning of the sentence and the setting in which it is uttered, there may be certain other propositions, called implicatures,that the teacher intends to convey by means of conveying this first proposition. For instance, the teacher might intend to convey the fact that he will commence lecturing.
Implicit in this conception of communication is a distinction between sentence expression and speaker expression. Relative to a setting, a sentence expresses a proposition in virtue of its meaning and certain features of the setting. Which proposition a sentence expresses relative to a setting is a matter of the semantic rules of the language that interlocutors must normally know in common. In contrast, the proposition that the speaker expresses will be the content of the thought that the speaker intends the hearer to recognize on the basis of the speaker's choice of words and shared understanding of semantic rules. If the speaker neglects the rules or is mistaken about what they require, then it may happen that the proposition that the speaker expresses is not the proposition that the speaker's sentence expresses. A source of confusion about this distinction is that the features of the setting that determine which proposition a sentence expresses may sometimes be features of the speaker's state of mind. For example, it might be thought that what determines the domain of discourse relative to which we should interpret the quantifiers in a sentence is just the class of things that the speaker has in mind in speaking the sentence.
The concept of thought content is integral to the expressive theory of linguistic communication inasmuch as it is the content of the speaker's thought that the speaker intends the hearer to grasp on the basis of the speaker's choice of words. Apart from the conception of content as something shareable between speaker and hearer, the expressive theory of communication would amount to little more than the thesis that something happens in the speaker, which causes the speaker to speak, and as a result of the speaker's speaking something happens in the hearer. The expressive theory is distinguished from this completely vacuous theory primarily by the idea that in successful communication there must be a certain relation between what happens in the mind of the speaker andwhat happens in the mind of the hearer, and that relation is a relation of common content (although the hearer's attitude toward that common content may be different from the speaker's attitude toward it).
I am stressing this because it sometimes happens that a theorist advances a theory of communication that in various ways is committed to the expressivist framework but then declares that it is not to be expected, even in cases of successful communication, that the content that the speaker expresses will match the content that the hearer ends up grasping (e.g., Bezuidenhout 1997). Such a theory is liable to be vacuous unless the theorist can tell us what relation has to obtain between the content expressed and the content grasped, and if the theorist tells us that, then we will be able to use that answer to define a level of sameness of content such that we may say that the content expressed must be identical, at that level, to the content grasped.
Contents conceived as something shared in communication must be distinguished from contents of various other kinds represented in the philosophical literature. There are epistemological, folk psychological, and various semantic conceptions of content, and it cannot be taken for granted that any of these others is just the kind of content required by the expressive theory of communication.
Expressivism. as I have described it here, cannot be directly attributed to any particular philosopher. The reason for that is that it is not very easy to find explicit statements of what almost everybody takes for granted. Nonetheless, I believe that expressivism represents a common tendency among many authors including: Grice (1989), Stalnaker (1973, 1976, 1991/1974, 1998), Kaplan (1989), Davidson (1990), Fodor (1975), Loar (1981), Bach (1987), Jackendoff (1994), Lewis (1975), Bennett (1976), and Sperber and Wilson (1995). I would be very surprised if any these authors (other than Grice, who is deceased) would not affirm that expressivism, as I have described it here, is essentially correct. If your name is on this list, and you do not consider yourself an expressivist in my sense, please tell me why not.
3. Issues internal to expressivism
Issues that divide expressivists fall into two main categories: First, there are issues concerning the place of thought in the theory of semantics. Second there are issues concerning the nature of the underlying thoughts and their contents.
One basic issue concerns the prospects for intention-based semantics. Again, the expressivist will allow that it is by virtue of a common knowledge of the semantic properties of words that the speaker can expect the hearer to grasp the content of his or her underlying thought. In his paper, "Utterer's Meaning, Sentence-meaning and Word-meaning" (first published in 1968; reprinted in Grice 1989 and Davis 1991) Grice had proposed to explain the semantic properties of words in terms of speaker's intentions. Roughly, the timeless meaning of a sentence was to be the sort of thing that speakers "have it in their repertoire" to mean by it. So not only are we to understand the speaker's meaning on a particular occasion as the content of a thought that motivates the act of speech, but in addition the semantic properties of words that speakers exploit in this way on particular occasions are to be explained in terms of what speakers of the language generally mean by such forms of words. The project of intention-based semantics was pursued as well by Bennett (1976) and early Schiffer (1972).
However, it is quite possible to be an expressivist in my sense without believing in intention-based semantics. For instance, Lewis (1975) thinks of a language, including the semantic properties of the language, as a conventional choice among the members of a community. In Davidson's later writings (such as 1986, 1990), although not perhaps in his earlier writings (such as 1975), Davidson seems to qualify as an expressivist, but one who thinks of semantic properties not as a matter of speaker's intention but as a matter of radical interpretation. The program of intention-based semantics has been criticized by an apostate Schiffer (1987). A particular stumbling block was how to explain the possibility of generating meaningful, but novel sentences on the basis of meaningful subsentential components. It seems to me that there is no one working in the philosophy of language today who believes in intention-based semantics.
Even among expressivists who reject intention-based semantics an issue can arise about the place of thought content in semantics. As I have explained, a sentence will not generally express a proposition apart from the setting in which it is uttered. A question among and between expressivists is whether or to what extent the pertinent features of the setting include features of the speaker's state of mind. For example, if on a given occasion the demonstrative "that" refers to a cat, then one might say that what makes it the case that "that" refers to that is that that is what the speaker had in mind; or, alternatively, one might maintain that there are certain semantic rules that determine the reference of a demonstrative in light of the setting irrespective of what the speaker has in mind, so that if the speaker does not have in mind the object determined by these rules, then what the speaker expresses will not be what the speaker's sentence expresses in the setting. (See Wettstein 1984, Reimer 1991a, 1991b, and Bach's replies to Reimer, 1992a, 1992b. My own 1997 is relevant here, but it is aimed at criticism of expressivism rather than taking sides on an issue within expressivism.)
Another issue in this vicinity is in what way we should think of language as conventional. The very fact that there are many languages gives some sense to the claim that we might have used different words in place of those we do use. But there may be some doubt about whether language is conventional in any stronger sense. For instance, Lewis (1969, 1975) holds that language may be conceived as a solution to a coordination problem (which is a problem in which the best choice for each of several parties depends on the choice that the other parties make), and someone might doubt that. It might even be doubted that there is any sense in which languages must be shared. For instance, this has been doubted by Davidson (1984b).
The second class of issues among expressivists concerns the nature of the thoughts that underlie language use and the nature of their contents. One issue in this class concerns the structure of these thoughts. Should we think of them as subpersonal particulars localizable in the brain (the majority view), or should we think of them as somehow states of the whole brain or whole organism (Stalnaker's view, 1984)? If we think of them as subpersonal particulars, should we think of them as having a structure very similar to that of a spoken sentence, so that they are subject to a division into word-like components classifiable as nouns, verbs, etc. (Fodor 1975)? And if we think of them that way, can they contain demonstratives and indexicals and be elliptical or in other ways be incomplete so that their contentfulness may be as much a matter of the context in which they occur as it is a matter of their internal structure (Perry 1993/1986)? Or must they instead be something like eternal sentences, whose meaning is entirely determined by their structure and components (as Pinker 1994 and Levinson 1997 seem to think)?
A second issue in this class concerns the kind of content that might be communicated. The contemporary concept of mental content has its roots in the late 19th/early 20th century writings of Gottlob Frege (especially 1892 and 1918). A great deal of recent writing has revealed that Frege's various conceptions of content do not all amount to the same thing (e.g., Taylor 1995). Quite apart from the problem of sorting out the various conceptions of content, there is a problem about what kind of content might be said to be expressed in linguistic communication. In particular, there is the following sort of problem (discussed by Heck 1995 and Paul, unpublished ms.): Suppose a speaker communicates a thought by means of a proper name. Is the person or thing named literally part of that content, or does the content contain, in place of the thing named, just some description or conceptualization of that object? If the thing named is literally part of the content, then we may find that we have to say that communication has been successful, because the hearer has grasped the right content, even though intuitively communication has not succeeded, because the hearer has grasped this content in the wrong way. If, on the other hand, we build some characterization of the object named into the content, then it may be unreasonable to expect hearers to grasp the content solely on the basis of the speaker's words and the setting.
Finally, there is the question, "What makes it the case that a particular thought has a particular content?" What makes it the case that a particular belief, in some particular person's head, is, say, the belief that the President is a Democrat? Why is that same thing not, instead, a belief that the President is a Republican or a belief that yesterday was a rhinoceros? One point of view is that the content of a thought is first of all a matter of its functional role within the thinker who has it (Block 1986). Another point of view is that content can be explained in terms of biological function (Millikan 1986, 1989). Yet another is that content can be explained in terms of correlations between occurrences of the thought type and occurrences of that which it represents (Fodor 1987). Still another idea, closely related to functionalism, is that there will be a general psychological theory, formulated in terms of content-bearing states, such as belief and desire, and that a creature has such contentful states just insofar as it is a model, in the logician's sense, of the theory. (Loar 1981 can be taken as an illustration of this idea, although he does not formulate his thesis in terms of models.) Perhaps the most common idea is that the meaningfulness of thoughts may be understood, on analogy with cartographical representation, as a kind of isomorphism between the elements in a system of mental representation and the world (Cummins 1989, 1996).
4. Criticism of the expressive theory of interpretation
One category of criticisms of the expressive theory of communication concerns the nature of interpretation. The question is whether it really is necessary to suppose that in normal cases of communication the hearer in some way contemplates the content of the speaker's thought and that the speaker intends the hearer to do that. Many people have been persuaded that this is so by Grice's famous paper "Meaning" (1989, originally published 1957). In that paper, Grice claimed that cases of someone's meaning something by something are distinguished from other acts in that the speaker intends to bring about some effect and intends to bring it about by means of the hearer's recognition of the speaker's intention. Originally he thought that in the case of declarative utterances if the speaker means that p, then the intended effect would be that the hearer believe that p. In later work, however, the intended effect was to be that the hearer recognizes that the speaker believes that p.
The trouble is that the only sorts of arguments that have ever been offered for this analysis are patently fallacious. Grice himself tends to reason as follows: He describes a case of not meaning anything by anything; he observes that it lacks some feature, and then he infers that cases of meaning something by something must possess that feature. For example, he asks us to consider a case in which someone leaves B's handkerchief at the scene of a crime intending the detective to believe that B was the murderer. According to Grice, this person does not mean anything by doing so. Presumably because in this case of not meaning something by something the agent does not intend the detective to recognize that he intends the detective to believe that B committed the crime, Grice infers that in general if someone means something by something, then he or she intends that the hearer will recognize his or her intention to get the audience to form a certain belief. But this is a fallacy. It is as though we were trying to define "mammal", observed that an alligator is not a mammal and that an alligator lacks wings and then inferred that all mammals have wings.
In reply it might be said that Grice expects us to see not only that some of the intentions that he says are characteristic of meaning something by something are absent in cases of not meaning anything by anything but also that they are always present in cases of meaning something by something. But what reason is there to believe that? Suppose two people are waiting for a bus and A observes that B is impatient for the bus to arrive. (B checks her watch, steps out in the road to look for the bus, etc.) A says to B, "The bus will be here within five minutes". Presumably, A meant by that that the bus would arrive within five minutes. Does A intend that B will recognize that A believes that the bus will arrive in five minutes on the basis of B's recognition that A intends B to recognize this? I see no reason to think so. It is not even obvious that A intends B to believe that the bus will arrive within five minutes or to believe that A believes this. Maybe A has no idea whether B think such things but hopes that in any case saying so will put B at ease.
Whether or not we need a Gricean analysis of meaning something by something, it might be said that communication is clearly a matter of a hearer's recognizing the content of the thought that motivates the speaker's act of speech. What makes this so clear, it might be said, is the phenomenon of meaning more than we say. The best way to understand what is going on in these cases, it might be said, is to suppose that the hearer infers what the speaker has in mind somehow on the basis of what the speaker literally says. Grice had an influential theory of this too, which he presented in his paper, "Logic and Conversation" (1989, originally published 1975). According to Grice's theory, hearers may determine what more the speaker has in mind by supposing that the speaker was trying to be cooperative in saying what he or she did say. In one of Grice's examples, A is standing by his car at the side of the road, and is approached by B. A says, "I am out of petrol". B replies, "There is a garage around the corner". What B means by this is not just what he literally says, namely, that there is a garage around the corner, but also that A can buy petrol at the garage around the corner. According to Grice, the way A will understand this is by inferring that this must be what B has in mind given that what he literally says is supposed to be cooperative; and this is what B will intend A to do.
But such examples could be treated very differently if we were not antecedently committed to expressivism. In the petrol example, it is not obvious that A has to contemplate what B has in mind at all. On the contrary, he might simply make an inference from the setting and what B literally says to the conclusion that he will be able to buy petrol at the garage around the corner. Perhaps they are at a busy intersection in the middle of the day; so if there is a garage around the corner it is liable to be a thriving concern, open for business with petrol for sale. If the conversation takes place in the middle of the night in the middle of nowhere, A will not conclude that the garage is open and has gas to sell. He will first ask for more information. If B leaves without giving more information (perhaps he is driving by, shouting out the window), then A might still, in desperation, go around the corner hoping for petrol, but he will not presume that the gas station is open with petrol for sale just on account of what B said. Generalizing from this example, we might infer that even when speakers mean more than they say, what happens may not be that hearers infer what speakers have in mind but that hearers make an inference from what the speaker literally says and the setting.
There is also a more subtle question to be raised about the expressivist's conception of semantics. As I explained in the previous section, an expressivist will need a theory of semantics, conceived as what interlocutors know in common about their language (or their respective languages) that enables them to infer the speaker's state of mind on the basis of the speaker's choice of words. The question is whether the expressivist can give a workable theory of the ways in which the proposition expressed depends on the situation in which the utterance takes place. For instance, how does the situation determine the reference of demonstratives? How does the situation determine the domain of discourse relative to which we interpret quantified sentences? On the one hand, it can be doubted whether the pertinent facts about the situation are what the speaker has in mind. For instance, it can be doubted whether the domain of discourse relative to which an utterance of a quantified sentence ought to be evaluated is just the domain of things that the speaker has in mind. After all, the speaker has an obligation to speak in ways that are understandable; and so one might expect that the pertinent features of the situation have to be accessible to the hearer without the hearer's having to do anything so difficult as infer what the speaker has in mind. On the other hand, it is doubtful whether there is available to the expressivist any other general theory. (For a development of this more subtle critique, see my 1997.)
5. Criticism of the expressive theory of thought
Expressivism is not overtly a theory of the nature of thought, but it does put certain constraints on a theory of thought, and we may ask whether these constraints are tolerable. In particular, expressivism entails that our theory of the nature of thought must treat thought as sufficiently language-independent that we might without circularity explain language in terms of thought in the manner of expressivism.
In considering this question, it is important to understand that what is at issue is specifically the kind of thought having the kinds of contents that, according to expressivism, words have the function of conveying. Call this conceptual thought. We could attempt an independent characterization of this conceptual thought. For instance, we might define it recursively, thus: a conceptual thought is either a thought to the effect that some particular belongs to some category, or a thought based on an inference, broadly speaking, from some other conceptual thoughts. But the main thing to understand is that it is the kind of thought whose contents are supposed to be shared in communication. Not every kind of thought does qualify as conceptual thought in this sense. For instance, imagistic thinking does not. Moreover, there may be many other sorts of mental process that deserve to be called thinking although they cannot be grasped by analogy to words or pictures but can be understood only in terms of neurology or in terms of a terminology invented just for the purpose of explaining these kinds of thinking.
The expressivist certainly need not deny that conceptual thought is in every way independent of language. Of course, what we think depends largely on what people tell us. Some of the things we think about, such as words and books and even Wednesdays and marriages, depend on language in various ways. There may be thoughts that a person could not very easily hold in mind without the help of the notations that public languages pr ovide. Most importantly, the child may form certain concepts only because those are the concepts expressed by the words in the language that the child has to learn. Thus, it is perhaps only because the child observes that a number of objects are called "chair" and a number of other, in some ways similar objects are not called "chair" that the child forms a concept comprising all and only chairs rather than a concept comprising both chairs and stools but excluding other things. What might challenge expressivism is only the claim that the very contentfulness of thought is understandable only in terms of the meaningfulness of the words that people speak.
One way to challenge expressivism along these lines would be simply to criticize all of the various theories that attempt to explicate thought content in a language-independent manner. That would be an undertaking beyond the scope of this article, but I do want to say something simple about what I take to be the predominant conception of mental representation, namely, the isomorphism idea. The idea is that mental representations represent the world insofar as they stand in an isomorphism to the world. This idea is encouraged by an analogy between mental representations and actual maps, but not much can rest on this analogy since it is not obvious that the relation between maps and what they represent does not depend on the fact that maps are understood, that is, represented in certain ways. It is actually not very easy to understand how the technical definition of isomorphism (roughly, a structure preserving mapping) could be put to use in a theory of mental representation. (For example, what relation in the world is the image of the relation that obtains between two sentences when a symbol for disjunction is written between them?) When all is said and done, what the idea comes to is that if we think of a set of mental representations as a theory (in the logician's sense), then an interpretation (in the logician's sense) of the language of the theory in terms of objects and actually obtaining relations between them must be a model (in the logician's sense) of this theory.
When the isomorphism theory is shown to amount to this, then several basic problems plainly emerge. One is that the world constitutes a model of a theory only if the theory really is true, and we cannot assume that every contentful, sentence-like mental representation is true. Another is that whenever there is one model there are many. Given one model we can always get another one by substituting individuals for individuals in accordance with a one-one mapping of individuals into individuals (making the substitutions both in the interpretation of singular terms and in the extensions assigned to predicates). So even in a case where we could suppose that all of the mental representations in the system really were true, the isomorphism theory would do essentially nothing to identify the representational content of those mental representations. So the isomorphism theory has to be supplemented with an account of the sorts of structures of objects and relations from which the intended model might be constructed. Any attempt to do so is liable to founder on the following objection: In order to adequately narrow the range of acceptable models, we will have to very strictly limit the sorts of relations that can be represented. Indeed, we will have to restrict them to a narrower class of relations than can be represented in fact. In other words, if the acceptable models can be built from all of those relations that can be represented in fact, then there will not be just one model.
Quite apart from such questions concerning the nature of thought content, the expressivist can be challenged to give an adequate theory of how concepts arise in the mind in the first place. Here I am thinking of concepts as components of thoughts that stand to individual words as whole sentences stand to whole thoughts. If thoughts are subpersonal particulars that bear a content, then concepts are likewise subpersonal particulars that bear a content (and, confusingly, this content might also be called a concept). The question how concepts arise in the mind is the question how there arises in the mind a thought-component with that kind of content. Here again there are many theories and each of them might be evaluated on its own terms, but there is also a general challenge that can be put to any theory that attempts to answer this question in a manner compatible with expressivism, as I will now explain.
The expressivist will probably hold that the concepts words express are formed as a result of observing the use of words. Several objects are called "chair"; several otherwise similar objects are not called "chair". The child may try calling something a "chair" and things may go smoothly. The child may try calling something else a "chair" and things may not go so smoothly. On the basis of these observations the child somehow abstracts the general concept chair (which we are thinking of as like a meaningful word in a language-like system of mental representation). The hard question is: Why is it not a miracle that different people, abstracting from different exemplars, wind up abstracting roughly the same concept, so that they will mostly agree on whether some novel object that comes along ought to be called "chair"? In outline the answer has to be that the process of abstraction is constrained or guided in some way, and it is constrained or guided in the same way in all human beings. The problem is that no one has any very helpful ideas about what these constraints and guidelines might look like. The proposals in the literature, such as Rosch's theory of correlational structure (Rosch and Mervis 1975), are usually so vaguely and imprecisely, or confusedly, formulated that it is impossible to put them to use. The few clear proposals in the literature (such as Markman's mutual exclusivity assumption, 1989) do little to narrow down the possible abstractions.
A philosophical argument against the notion that the nature of thought content can be explicated in a manner compatible with expressivism can be constructed on the basis of social externalism. Social externalism, which grew out of Tyler Burge's seminal paper, "Individualism and the Mental" (1979), is roughly the thesis that the very content of a person's thought depends on the way words are used in the surrounding linguistic community. For instance, suppose that in Art's world "arthritis" applies to specifically inflammations of the joints; whereas in Bart's world, that same word applies to a broader class of rheumatoid ailments, including such as might occur in the thigh. However these differences between their worlds have not affected them, and Art and Bart are microstructurally identical. Still, it can be argued, when Art says, "I have arthritis in my hands", he is expressing a different thought from the thought that Bart expresses when he says those same words. (For a defense of social externalism, see my 1991 or my 1994, ch. 3).
If social externalism is true then it would seem that there is some kind of
circularity in the expressive theory of communication. On the one hand, the
expressivist wants to explain the way words are used in terms of the
contents of underlying thoughts. On the other hand, social externalism tells
us that we can explain the contents of underlying thoughts only in terms of
the way words are used. Thus social externalism appears to lead
expressivism in a circle. Of course, even granting the truth of social
externalism, various questions about this argument would have to be
answered: Is it really the way words are used to which content is relative
according to social externalism? Will the appearance of circularity remain if
we are careful to distinguish between various concepts of the use of words?
I will not try to answer those objections here. (For a full discussion, see my
forthcoming.) I should note, however, that insofar as the concept of content
at issue is proprietary to expressivism, the repudiation of expressivism is
equally a repudiation of social externalism formulated as a theory of content.
6. The superficial questions
Now I am in a position to say something about the questions whether we think in language, whether nonhuman animals can think and how language shapes our concepts.
As for the question of whether we think in language, this has sometimes been discussed at some length without ever raising the question of the nature of linguistic communication (for example, by Carruthers 1996), but I do not see how this can be very profitable. So long as we accept the expressive theory of communication, then the answer has to be basically, no, thinking in language is not very fundamental. Discussion with other people can be an aid to thinking not just because we can gain information from them, but also because it challenges us to defend our views and organize our thoughts and think things through. The expressivist can allow that in all these ways talking to ourselves might be an impetus to real thought. Nonetheless, from the expressivist point of view, the real thinking has to be conducted in the sort of conceptual thought that we would express to others if we were really talking.
If, on the other hand, we are willing to deny the expressive theory of communication, then we might conceive of thinking in a language as making possible a kind of problem solving that would not be possible otherwise; but even then we have to acknowledge processes of thought that are more fundamental than this. For instance, if we think of language on the model of the use of a tool, a tool for manipulating other people's behavior in ways that would not be possible otherwise, then it might be conceivable that once we have learned to use this tool to manipulate the behavior of others we can use it to control ourselves as well. But however we propose to escape from expressivism, we will have to suppose that some kind of thinking, other than thinking in language, makes language learning possible and continues to underlie our choice of words. We should be prepared to find that the heart of cognition lies in this other kind of thinking.
As for the question whether nonhuman animals think, no one should doubt that they do think in the sense that mental processes take place in them that result in a limited kind of problem solving. The only question is the nature of those mental processes, and in the context of discussions of language and thought the question is specifically whether they think the kind of conceptual thoughts that human beings are able to express in words. This question is bound up with the nature of linguistic communication as follows: If the expressive theory of communication is correct, then there is room for such a hypothesis. We do not have to suppose that nonhuman animals have any concepts quite like those that human beings have or that we speak the literal truth if we say something like, "The dog thinks the squirrel is up the tree". (Can a dog have the concept squirrel, as opposed to pesky critter?) Still, some animals might possess a kind of thought of which human's conceptual thought is an in some way more congealed form. But if the expressive theory of communication is mistaken, then the question becomes unclear. We know best what we mean by "conceptual thought" in the context of expressivism: It is the kind of thought the content of which words may be used to convey. Apart from a commitment to expressivism it is unclear what sort of thought is at issue.
Concerning the idea that the meanings of words determine the concepts we think in, the only point I want to make is that sometimes authors aim to give language a major role here but do not really succeed in escaping from the expressive theory. An example is Andy Clark, who writes:
In principle, the idea that words mark situations and by so doing alter the similarity relations between them might serve us well in a nonexpressive theory of how language works to facilitate cooperation between people. But all Clark is saying is that words form representations of individuals and, as he might have added, kinds where, apart from a need to talk, there would have been no need to treat various experiences as in some way the same. Clark's own expressivism becomes explicit when, in commenting on the "mundane observation that language allows ideas to be preserved and to migrate between individuals", he explains:
To see how little is actually credited to language in the claim that words compress sensory patterns into objects and kinds, consider this question: Where do the words come from in the first place? It cannot be just because the child has observed that a number of very different things (Great Danes, Pekingese) are all called "dog" that the child forms the general idea of dog. Someone or some group of people must have had some reason for calling all these different things "dog" in the first place. So in principle it must be possible for the mind to form this grouping in a language-independent way. Words can at most speed the process, promote uniformity, and perhaps preserve certain arbitrary divisions. (For another example of an attempt to give a major role to language without ever leaving the safe harbor of expressivism, see Dennett 1991, ch. 8.)
This is not to say that expressivism is inevitable. There is not much point in thinking of words as denoting categories unless we suppose that thoughts of those categories are shared between interlocutors. But we do not have to think of words as doing their work only insofar as they denote categories. Insofar as we can talk, we can of course talk of categories. For instance, in using the word "dog" we may talk about the class of things comprising all and only dogs. But it does not follow that our fundamental theoretical account of how language facilitates cooperation between people should explain it in terms of such a denotation relation between words and categories, and so likewise we might conceivably explain how language works without invoking a relation of expression between words and representations of such categories.
7. Another model
An alternative to expressivism would have to be a comprehensive theory of language, including both a theory of the mental processes that underlie the use of language and a theory of the norms of discourse that guide conversation. Such a thing is obviously beyond the scope of this article. But in closing I do want to say something about how one might conceive of specifically the relation between language and intentional states such as belief and desire if one does not suppose that the function of language is to express such intentional states.
One simple idea, which we have already considered, is that we think in language. Indeed, it might be useful to think of certain occurrent thoughts as like silent assertions or commands in a language that can also be spoken aloud. What distinguishes occurrent thought of this kind from overt speech might be not so much that it is silent but that one directs it toward oneself and not toward others. In any case, that does not yet tell us how we should think about standing intentional states such as beliefs and desires. There is no particular reason that I know of to think of them on the model of a file cabinet: as sentences of a spoken language stored somewhere in a "belief box" or "desire box". From the point of view of the language of thought hypothesis, according to which conceptual thought is realized in sentence like structures in an innate language of thought, it might make sense to think of beliefs and desires as sentences stored in files; but there is no reason to preserve this model of belief and desire once we have abandoned the language of thought hypothesis.
An alternative manner of conceiving of beliefs and desires as dependent on language is to suppose that what we have to understand about beliefs and desires is first of all the distinctive role of talk of such things. Somehow talk of beliefs and desires adds something to the power of language to facilitate cooperative activity between people. If we could explain how talk of beliefs and desires does this, then there would be the option of supposing that all that can be said about the nature of beliefs and desires is that talk of them plays such a role. That there is nothing more to say about what beliefs and desires really are presumes that our account of our talk of them would not invite some further reduction. If we supposed that the sole reason to talk of them were that doing so enabled us to explain and predict behavior and that, moreover, this method of explaining and predicting behavior really did work, then we would have to expect that somehow beliefs and desires could be located among the arrangements of matter that at a deeper level really do produce the behaviors thus explained and predicted.
But there is another way of thinking about the attribution of beliefs and desires that does not in this way invite further reduction. When a person makes an assertion, he or she usually expects that act to have some kind of impact on other people (which is not to say that the act of speech is the product of some kind of deliberation over how best to produce that effect). In attributing a belief that p to a person in his or her absence we may expect our act to have much the same impact as that person's own assertion that p might have had. In other words, an attribution of belief to a person may serve as an assertion on that other person's behalf. Similarly, an attribution of a desire to a person may serve as a command or request on that other person's behalf. If along these lines we could develop a general theory of the function of attributions of belief and desire, then that might justify the following claim about the relation between thought and language: Public language is the medium of a certain kind of thought not because, or not only because, one kind of thinking is identical to talking to oneself, but because in general intentional states have their essence in the role that talk of intentional states plays in the conduct of productive conversation.