The above minor arguments don’t play a central role in Davidson’s support of P1. The strongest support for the crucial premise is found in what I will call his Master Argument.[9] The Master Argument puts psychological restrictions on what it is to be an interpreter, and it supports the claim that one cannot have propositional attitudes without language. If the Master Argument succeeds, then Davidson’s arguments for denying that animals have propositional attitudes is compelling. But, as I shall argue, the Master Argument ultimately fails, and thus also fails to support the denial of propositional attitudes to animals.
According to Davidson’s interpretationism, having beliefs entails being an interpreter. The basic idea of the Master Argument is that possessing certain concepts is a prerequisite for being an interpreter, and that an organism must have language in order to have these concepts.[10] Davidson’s position differs from the more widely-held view that having some concepts is required for having propositional thought, by supposing that there are specific concepts that a creature must possess in order to have propositional thought. The Master Argument links thought to language by way of higher-order thoughts. Specifically, Davidson suggests that a concept of belief is a prerequisite for propositional attitudes, and that a concept of belief is unavailable without language. Here is Davidson’s Master Argument:
M1 If S has propositional attitudes, then S has beliefs.
M2 If S has beliefs, then S has a concept of belief.
M3 If S has a concept of belief, then S has language.
MC If S has propositional attitudes, then S has language.
The argument is clearly valid. But is it sound?
M1 is plausible; it just highlights Davidson’s view that beliefs are a fundamental propositional attitude, and that to have any propositional attitudes at all, a creature must have some beliefs. M2 and M3, the remaining premises, are interesting, but their meaning is unclear, for they contain a clause that needs to be unpacked: what exactly is a “concept of belief”? Let us distinguish three different conceptions of a “concept of belief”, varying in stringency. One conception of the concept of belief is robust, in which the concept of belief is the fully articulated belief-concept that is taken to be definitive of a mature theory of mind. On this robust view, having a concept of belief is an epistemologically-rich notion that entails having an ability to pass the “false belief test”. That is, it is criterial for having the concept of belief that one has the ability to attribute to others a mental representation of the world that may differ from the way the world is, as well as a recognition of the perceptual circumstances that would engender false representations. In contrast, a deflationary conception of what it is to have the concept of belief merely requires an understanding that the world is distinct from how it appears or how one takes it to be, or that belief can come apart from reality. On a deflationary view, then, having the concept of belief is rather like having the concept of an objective reality. Finally, we might consider an intermediate notion of the concept of belief that involves the ability to attribute representational mental states to oneself and others, without satisfying all the constraints that a robust conception must meet. Which, if any, of these conceptions of “concept of belief” is important for Davidson’s argument linking belief to language?
The robust conception of belief became important in developmental psychology in the context of concerns about Theory of Mind: having a notion of false belief was taken to be diagnostic of a mature TOM, and, according to many researchers in the field, only develops in humans at around four years of age (Saxe et al. 2004; Wellman et al. 2001; Wimmer & Perner 1983). However, requiring a concept of belief in the robust sense seems too demanding a condition for having propositional attitudes. While we might plausibly doubt whether prelinguistic infants really have propositional attitudes, it is hard to deny that young children who have already acquired a sophisticated facility with language have propositional attitudes. Children of two and three, for instance, clearly refer to objects in the world using language, and they readily express their desires (“I want the green monkey!”), beliefs (“I think the ball is under the bed”), as well as fears and other propositional attitudes. They understand others, refer to their own and others’ mental states, and communicate effectively. We typically and with great conviction attribute propositional attitudes to children of these ages. Nonetheless, according to most developmental psychologists (See e.g., Perner et al. 1987; Call et al. 1999; http://youtu.be/8hLubgpY2_w), until the age of four (two years after they develop considerable language abilities) children lack a concept of belief in the robust sense.[11] And if so, we make ordinary propositional attributions to children well before they possess the robust concept of belief. Thus, M2 is false.[12]
Not all psychologists agree that a robust concept of belief doesn’t develop until about four years of age. Some have argued that the methods used in many of the classic false belief studies rely too much on language or on inhibitory control, and that tests other than the classic false belief test are sufficient for demonstrating understanding of false beliefs. For instance, a recent study suggests that children have a concept of belief at far earlier ages than previously thought—earlier, in fact, than the development of language (Onishi & Baillargeon 2005; Baillargeon et al. 2010; Caron 2009). However, if this is so, then M3 is false, for the robust conception of belief does not depend on having language. This version of the Master Argument depends upon a tight connection between competence in the false-belief task and belief. On one conception of what evidence is sufficient to reflect performance on the false-belief task, M2 is false, and on another conception, M3 is false. Either way, the Master Argument is empirically refuted, and the robust conception of “the concept of belief” fails to link language possession and propositional attitudes.[13]
Davidson may well be unperturbed, for there is no textual evidence that he means to implicate the robust conception of belief when he claims the concept of belief is necessary for having beliefs. After all, from the standpoint of his radical interpreter, one can only be a believer in virtue of interpreting others, but it is unclear why the possibility of such interpretation should rest upon a grasp of others’ mental states being beliefs in this robust sense, rather then in some weaker sense. In “The Second Person” (1992), for instance, Davidson argues that for our mental states to have determinate content we must interact with another being in order to “triangulate” and thus make determinate the referents of our thoughts. Nothing in this picture requires that an interpreter have a robust concept of belief as opposed to a more deflationary one.
In line with the idea that Davidson has a more deflationary view in mind, in both “Thought and Talk” and “Rational Animals” he mentions a different criterion for having a belief, which he also thinks links the possession of language to the ability to have propositional attitudes. This is the criterion of possessing a concept of “objective truth.” Davidson’s argument for language via the criterion of objective truth is as follows:
O1 In order to have propositional attitudes, one must have beliefs.
O2 In order to have beliefs, one must have a concept of objective truth.
O3 In order to have a concept of objective truth, one must have language.
OC Propositional attitudes require language.
The logic here is again unproblematic, but unpacking the premises is not. At times Davidson seems to equate the concept of objective truth with that of belief. I take this as evidence that he intends “the concept of belief” in the Master Argument in its most deflationary interpretation: as an understanding that how the world is can come apart from how one takes the world to be. Given this interpretation one could believe that the concept of objective truth co-occurs with that of belief, or that the cognitive conditions that make possible the concept of belief are the same as those that make possible the concept of objective truth. In any case, Davidson sees a tight connection between the notions of belief and objectivity.
How are we to understand the “concept of objective truth” in O2 and O3? If Davidson means it to be a metasemantic concept, such as having a Tarskian definition of truth, or an understanding that truth applies to propositions, and so on, then it would be almost assured that one could not grasp the concept of truth without language. It would explain the prima facie plausibility of the Objective Truth version of the Master Argument. However, if we adopt that reading of objective truth, O2 would be false, for people certainly have propositional attitudes even if they never become philosophers, and even if they never have an inkling about metasemantic notions.
Another clue about what Davidson means by objective truth comes from his emphasis on triangulation. Davidson thinks we need to interact with another person in order to come to see the world as external to us—in order to develop a notion of objectivity. By linguistically triangulating on an object with another, we are forced to recognize that object as part of an objective reality. Davidson illustrates this view in “The second person”:
Belief, intention, and the other propositional attitudes are all social in that they are states a creature cannot be in without having the concept of intersubjective truth, and this is a concept one cannot have without sharing, and knowing that one shares, a world, and a way of thinking about the world, with someone else. (2001b, p. 121)
However, there are two fundamental problems with using triangulation as an argument for the necessity of language for thought. First, there is nothing apparent about triangulation that requires spoken language as opposed to some other sort of joint interaction or non-linguistic communication. It is, indeed, difficult to see why language as opposed to action would be operative in developing a notion of a world external to ourselves. So triangulation fails to show that language is necessary for thought. Second, it is difficult to see how triangulation could itself suffice for a notion of objectivity. In order for me to triangulate with another, I must first see the other as part of the external world, as opposed to an element in my mentality. As long as the other is merely a part of the way I take things to be, it cannot fulfil the role of the second person (see for example, Roskies 2011). So triangulation also fails as a mechanism for constructing the concept of objectivity. Nonetheless, Davidson’s emphasis on triangulation strongly suggests that by “objective truth” he is referring to the appearance/reality distinction.
This interpretation is further strengthened by taking seriously the fact that Davidson thinks the concepts of belief and truth are closely linked (1984). As mentioned earlier, having the concept of objective truth is nothing other than understanding that how the world is can come apart from how one takes the world to be. What evidence do we have that language is required for this?
As further evidence that Davidson intends a deflationary view of the concepts of belief and objective truth, we can turn to another formulation of the Master Argument. In his most forthright explication of what he means by “concept of belief”, he suggests that there is a behavioral mark that is coextensive with having such a concept: surprise.
In order to have any propositional attitude at all, it is necessary to have the concept of a belief, to have a belief about some belief. But what is required in order to have the concept of a belief? Here I turn for help to the phenomenon of surprise, since I think that surprise requires the concept of belief. (Davidson 2001a, p. 104)
The willingness to consider some sort of non-linguistic behavior as relevant to the question of whether a creature has propositional attitudes is a methodological breakthrough, for it provides an avenue independent of language for assessing whether an animal has the requisite cognitive machinery to be a believer. Davidson maintains that the ability to be surprised is diagnostic of having the concept of belief. It indicates recognition that one’s own mental representation fails to conform to that which it represents, and as such it constitutes necessary and sufficient evidence of the concept of belief.
Following this intuition, we can amend Davidson’s Master Argument to incorporate this insight:
S1 If S has propositional attitudes, then S has beliefs.
S2 If S has beliefs, S has a concept of belief.
S3 S has a concept of belief iff S has the capacity for surprise.
S4 If S has the capacity for surprise, S has language.
SC Propositional attitudes require language.
The idea that surprise goes hand-in-hand with the concept of belief is not implausible: if surprise issues from the recognition that one’s belief about how the world is fails to correspond with the way the world is, then surprise is good evidence for the concept of belief. Moreover, because this idea does not have implications for the ability to attribute propositional attitudes to others in an operative sense, it suggests that the interpretation of “concept of belief” that Davidson favors is a deflationary interpretation: one that involves appreciation of the appearance/reality distinction, or, as discussed above, the concept of objective truth. Thus, S2 takes the deflationary interpretation of the concept of belief, and for the argument to be valid, S3 must also take that interpretation.
Unfortunately for this version of the argument, S4 is false. There is clear and abundant empirical evidence that the ability to be surprised at the mismatch between the world and one’s own representation of the world is independent of language (Dupoux 2001; Feigenson et al. 2002; Hauser & Carey 1998; Santos et al. 2002; Wynn 1992). Take, for example, an invaluable tool in the developmental psychologist’s toolkit: the violation of expectancy looking method (V) for testing infants. Many studies performed on pre-linguistic human infants employ this paradigm in order to explore what an infant knows. The idea is simple: infants look longer at stimuli that fail to correspond with their expectations. This method has been used to determine, among other things, that infants have an innate (or very early developing) concept of number. In now classic experiments, Wynn and colleagues demonstrated that infants can do simple arithmetic (Wynn 1992). She showed infants as young as five months a toy, and placed it behind a screen. Then she showed them another toy and also placed it behind the screen. The screen was then lowered, revealing either two toys (the expected outcome), or only one toy. Infants looked longer at the unexpected outcome. The same paradigm was used with different numerical combinations, demonstrating that for numerosity up to three, infants can do simple addition and subtraction, and are surprised when what is revealed behind the screen does not comply with their expectations. Significantly, this robust effect, which is due to surprise, precedes the development of language by more than a year.
Davidson might reply that it is not actually possessing language, but rather possessing the capacity for language that is important for surprise, and thus for the concept of belief. Maybe, even though they cannot yet speak, infants possess a language faculty, which, immature as it may be, is sufficient to support surprise. However, this attempt to patch the argument also fails. The VELM is used frequently in studies with nonhuman primates, and while they never develop language nor seem to have a capacity for natural language, they too exhibit surprise when their expectations are violated (Hauser 2000; Hauser et al. 1996). So, it seems, language is not a requirement for surprise, nor is surprise evidence for the presence of or capacity for language.
The empirical studies of developmental psychologists and primatologists undermine the Surprise version of the Master Argument: surprise does not depend upon having language. Moreover, if premises S2 and S3 are true—if the capacity for surprise is evidence of the concept of belief, and if propositional attitudes depend upon possession of the concept of belief—then propositional attitudes do not depend upon language.
Let us briefly revisit the Objective Truth version of the Master Argument. I have argued that only a deflationary notion of objective truth is a candidate interpretation for the argument. I have also suggested that this is the only notion of “the concept of objective truth” that meshes with the arguments Davidson raises regarding belief and surprise. Thus, having a concept of objective truth is having a concept that the way the world is can come apart from how one takes it to be. If this is correct, then the Objective Truth version of the Master Argument is false.
In Wynn’s looking-time studies discussed above, the child has clearly developed expectations of what lies behind the screen, and must somehow represent this to herself. When the screen is lowered and the child sees what is behind the screen, there must be some sense in which correspondence with the expectation or lack of correspondence is noted, and in which the data coming in from the senses is privileged over the internal representation. This is, in essence, what it is to recognize that beliefs about the world can come apart from the way the world is. Clearly this sort of grasp of reality does not depend upon language: pre-linguistic infants and non-linguistic animals possess it. One can easily imagine how violation of expectation can be instantiated in a system with imagistic thought. The languageless child need only conjure up an image of the objects behind the screen and compare this with the visual scene before him. As long as the child privileges the sensory information over the mental representation, we might say that he has a concept of reality and of the belief/reality distinction. In summary, then, language is not required for a concept of objective truth.
We have ruled out both the robust and weakest notions of “concept of belief” as candidate notions for a successful interpretation of Davidson’s argument linking belief to language. Perhaps an intermediate notion can do the job. This notion involves the ability to attribute representational states to oneself and others; it is less sophisticated than that required to pass the false-belief task, but more complex than the recognition of an appearance/reality distinction.
One potential reason why representational-state attribution may be important for having beliefs involves self-reflection: perhaps being a believer requires being able to think of oneself as a believer, and thus requires the concept of belief. This amounts to the claim that beliefs cannot be held non-reflectively. Since we clearly do have beliefs that we do not have beliefs about, what is at issue is not the actuality of having beliefs about beliefs, but the possibility or capacity to do so. However, while there are arguments that the ability to think about oneself as a believer is required for a rich construal of theoretical rationality (see Bermúdez 2003, Ch. 7), there is no clear argument why such reflective ability should be constitutive of having beliefs. Indeed, it seems like the ability to believe things about one’s beliefs would require that one could believe things, so that belief is conceptually prior to self-reflection. In any case, self-reflection is not Davidson’s stated reason for thinking that the concept of belief is important for having beliefs.
The other reason to hold that having belief requires having a concept of belief under the intermediate conception links the ability to attribute mental states to others with having the concept of belief. Thus there are two different strengths of intermediate interpretations to consider. According to the less demanding interpretation, a concept of belief is required in order to attribute contentful states to other creatures; whereas the more demanding interpretation holds that a concept of belief is required to attribute propositional attitudes to others: one must be an interpreter, not just an interpretee.
We can discount the less demanding of these interpretations for the purpose of this argument linking thought to language,[14] because if M2 (“If S has beliefs, then S has a concept of belief”) is interpreted in this way, then M3, the claim that language is required for a concept of belief, read in this way, is false. There is growing evidence that non-language-using animals are able to attribute representational states to other animals. One compelling illustration of this comes from (Hare et al. 2000), who show that subordinate rhesus monkeys only approach food in the presence of a dominant male when they know that the male is unable to see the food (interestingly, dominant males appear not to care whether or not a subordinate male sees food, pointing to yet a further level of sophistication in the cognitive processes of non-linguistic animals). Thus, if it is the case that to believe requires having the ability to attribute contentful mental states to others, then it is not the case that believing requires language. Indeed, recent work on non-human primate theory of mind suggests that monkeys and chimpanzees have a theory of mind that represents goal states and distinguishes between knowledge and ignorance of other agents (the presence and absence of contentful mental representations), even if it fails to account for misrepresentation (Call & Tomasello 2008; Kaminski et al. 2008; Marticorena et al. 2011). Although they may have a less articulated theory of mind than we do, we may nonetheless adequately characterize their representational system with mental-state terms (Butterfill & Apperly 2013; Marticorena et al. 2011).
What remains is the notion that the ability to attribute beliefs qua propositional attitudes to others is necessary for having beliefs. That is, not only must they attribute mental states to others, but those mental states must possess the characteristics of beliefs. Remember that we have already discounted the robust notion of belief as too demanding, so what is necessary is not that animals have a notion of false belief per se, but rather that they have a notion of a belief as a representational mental state that can play a role in behavioral explanation or prediction. So far there is no compelling evidence that nonhuman animals have this, consistent with the possibility that such a representational ability as this may indeed require language, or at least some sophisticated ability to symbolize abstractions and predicate them of objects. Whether this is so is ultimately an empirical question. However, at least some philosophers think monkeys may be able to do this. As Lurz characterizes the above studies, animals do have the ability to represent propositional mental states in others—not as attitudes aimed at representing objective truth, but instead as attitudes with propositional contents that provide information regarding motivation to act (2011a). Lurz characterizes this as a kind of belief–desire attribution. Baillargeon’s data proves relevant here too, for her results are best explained by taking the infants in her study as postulating representational mental states of the actor in order to predict her behavior; violation of their expectation causes them to look longer. Thus, without imputing these infants some understanding of others’ representational mental states, we would be unable to account for this data. However, in this case M3 would then be false, for the linguistic abilities of fifteen-month-old infants typically are minimal—certainly not of the sophistication we would expect would be necessary to linguistically encode a belief-concept. While the evidence that bears on this case is perhaps the least well-established, and this study involves infants at an age when they are poised to develop language, the burden of proof is shifted to the person who wants to argue that language is necessary for a concept of belief. That burden is not discharged: Davidson lacks a positive argument for why this relatively demanding notion of attributing content to others is the one required for an organism to be a believer.