1. Does the Brain Think?

Consider a picture of a domino with an arrangement of apparently concave and convex circles. The same picture rotated through 180 degrees makes the concavity and the convexity appear reversed. Why does this happen? Why does the very same picture, the same arrangement of pixels or ink on a page, appear so different when turned upside down? Chris Frith gives the following answer in Making Up The Mind:

Frith claims that the brain has a rule built into it and uses this rule to makes decisions about how things seem. Taken literally, saying that the brain ‘uses’ a rule, as opposed to merely behaving in a rule-governed or law-like way, implies that the brain somehow represents the content of the rule. And saying that the brain makes decisions implies that the brain is something like a thinker; in short, the brain thinks.

For those like Frith, the idea that the brain represents the world (or ‘thinks’), should be accepted as part of the orthodox ideology of cognitive science or cognitive neuroscience. Others say that the question is totally confused.Footnote ² For them, this kind of talk embodies a mistake; or even worse, a fallacy (and not all mistakes are fallacies). M.K. Bennett and P.M.S. Hacker have argued that it is an instance of what they call the ‘mereological fallacy’: the ‘mistake of ascribing to the constituent parts of an animal attributes that logically apply only to the whole animal’.Footnote ³ In this they take themselves to be following Wittgenstein, who famously said that ‘only of a human being and what resembles (behaves like) a living human being can one say: it has sensation; it sees, is blind; hears, is deaf; is conscious or unconscious’.Footnote ⁴ A brain doesn't resemble a living human being; it doesn't even resemble something that resembles a living human being. A chimpanzee at least resembles something that has thoughts; the brain does not even resemble a chimpanzee.

Bennett and Hacker think it's not empirically or straightforwardly false that the brain represents the world. Rather, it is a conceptual truth ‘that perception, thoughts and feelings are attributes of human beings, not of their parts – in particular not of their brains’.Footnote ⁵ So it is a fallacy to say something which is incompatible with this conceptual truth. But the supposed fallacy cannot derive from any conceptual principle that you cannot, in general, attribute things to the parts of a system that you would also attribute to the whole. There are many cases where you can do this (e.g. weight, colour etc.) which of course Bennett and Hacker will not deny. So if there is a fallacy here, it must be to do with the use of the terms ‘thought’ or ‘sensation’ or ‘consciousness’ or ‘thinking’ or ‘deciding’: mental terms in general.

It is true that the paradigm applications of the concepts of thought, decision, sensation and so on are to organisms: things like human beings and those animals which it makes sense to describe as conscious or thinking. But often we extend the use of words creatively beyond their paradigm applications, to illuminate or illustrate some significant feature of the thing described. Stephen Mulhall makes this point in a recent discussion of the concept of a picture. In considering what he calls the ‘projectability of language’, Mulhall writes,

He continues,

As the use of the word develops across time, we can extend its use by applying it intelligibly to other things. We find it very natural to extend the use of a word beyond some original or initial contexts, without the word changing its meaning in any strict sense. Another simple example is the concept of flying. Suppose a child asks, can a person fly? Of course people can't fly, we might reply; Superman can't really fly, it's only a story. So what is it to fly? Perhaps we might say something like this: to be propelled through the air, by using the motion of wings (as a bird flies) or some other kind of mechanism attached to the flying object (as a rocket flies).

But what if I tell you that I am flying to Turin in a few weeks' time? Then we should give another answer to the question ‘can people fly?’. Of course people can fly: travelling in an aeroplane is flying. It would be at best a bad joke to respond to the question ‘did you fly here, or did you come by train?’ by saying ‘no, I can't have flown here, because to fly is to propel yourself through the air using wings or some other kind of mechanism’. And yet, as Mulhall says about pictures, there should be no temptation to say that the word has one meaning in the first conversation and a different meaning in the second. Just as developments in painting led to the natural extension of the word ‘picture’ to Pollock's work, so developments in aeronautical technology led to a natural extension of the word ‘fly’ to what people do when they go somewhere by plane: in this sense, people can fly. When we focus on the historical development of the use of the word, any puzzlement we felt about giving different answers in the two scenarios should vanish.

As Mulhall says, this possibility of words having uses which extend across many different kinds of case, without being ambiguous or polysemous, is something to which Wittgenstein drew attention:

The relationship between the different uses or meanings of the words ‘picture’ and ‘fly’ in our scenarios seems to be a kind of family resemblance of the sort Wittgenstein talks about. To say this is not to propose or defend any particular theory or account of meaning; it is simply to draw attention to a phenomenon. But once we have recognised this phenomenon, we thereby open up the possibility of making sense of the idea that the brain thinks or represents the world.

However, some may object at this point that the meaninglessness of saying that the brain thinks is underwritten by the well-established distinction between the personal and sub-personal levels of psychological explanation and description. The concepts of thought, decision, and so on belong to the personal level, and we should not extend them to the sub-personal level. Before explaining what I think it means to say that the brain thinks, I need to put this objection to one side.

Daniel Dennett introduced the personal/sub-personal distinction in Content and Consciousness in the context of a discussion of pain:

As Dennett's remark about ‘alternative modes of explanation’ indicates, the personal and sub-personal are two modes of explanation of the one cognitive system. The personal mode of explanation appeals to concepts like sensation and then behaviour, location and reaction; and the alternative explanatory level of the sub-personal appeals to brains, nerves, neurones, synapses and events in the nervous system.

Dennett points out that the lesson about the personal/sub-personal distinction has occasionally been misconstrued ‘as the lesson that the personal level of explanation is the only level of explanation when the subject matter is human minds and actions’.Footnote ⁹ Now a strict and literal reading of Wittgenstein's remark about ‘only of a human being and what resembles a living human being…’ would presumably entail this claim that the personal level is the only level of explanation that matters when the subject matter is human minds and actions. But this is not Dennett's view. In his commentary on Bennett and Hacker's idea of a mereological fallacy, he writes,

Dennett's view is clearly that the distinction between the personal and the sub-personal levels of explanation is clearly compatible with two ideas: (i) that we might extend to the sub-personal level words whose paradigmatic application is at the personal level; (ii) that we might use this extension to help us explain how intentionality at the personal level is possible. One could reject either or both of these ideas, but this rejection is not implied by the very distinction between the personal and the sub-personal.

So this brings us back to our question, what does it mean to say that the brain thinks, or represents the world? I take Frith's comments quoted above as representative: many cognitive neuroscientists and others are perfectly happy to talk in terms of the brain representing the world. Given this, Hacker's approach leaves us with a mystery: how is it that so many people, apparently highly competent in their use of language and highly knowledgable about the empirical facts, fall so easily into such a simple fallacy? What is more, why is it that they are so resistant to recognising that they have made this apparently simple mistake?

I will argue in the rest of this paper that in fact it is not a mistake, and I will attempt to explain what it means to say that the brain represents the world. Just as it makes perfect sense to say that people fly, and it makes perfect sense to say that a Jackson Pollock painting is a picture, so similarly it makes sense to say that the brain represents the world, or even that it thinks. If we allow for words to extend their meaning through a historical process, then we can see how this is not obviously meaningless. In fact, on broadly Wittgensteinian grounds, we should accept that such transfers and extensions of meaning are part of the essence of our language, and there is nothing in principle stopping the talk of neuroscientists being such a case. But how should we spell out this idea that the brain represents?

2. Contents as Models

One answer to this question, going back to the 1970s, is that representation in the brain involves there literally being symbols written in your brain: there are symbols which represent the contents of your beliefs and other personal mental states, or your sub-personal states, or both. These symbols are part of what's known as a ‘language of thought’; this view was defended by Jerry Fodor in 1975.Footnote ¹¹ Fodor argued that there can't really be beliefs unless there are representations in the brain.

The main problem with the Language of Thought hypothesis, to be blunt, is that there is no reason to believe there is such a thing as the Language of Thought. Some people hypothesise that there might be such a thing for the sake of argument, but few of those who say this never defend the idea that it really exists (and even Fodor himself is sceptical these daysFootnote ¹² ). As Frances Egan remarks: ‘There isn't much empirical support for this view. It's a very elegant picture but it's not likely to be the way that minds developed naturally, as a product of evolution, in fits and starts’.Footnote ¹³

Instead of speculating about the inner structure of the brain, or simply denying that representation in the brain makes any sense, we should ask instead of what it is that people are doing when they attribute representations to the brain – when they say things like ‘the brain knows that light comes from above’. What is it that those who attribute intentional states are actually doing in this kind of case?

The interpreter is the theorist – the neuroscientist or psychologist – who is trying to provide the best explanatory structure to account for how the system moves from one state to the next. The system, in the cases we are considering, is the brain, and what the theorist is trying to do is to explain why the system moves from one state to another. There are states of the system which the theorist can then map on to what I'm going to call its contents. The content might be something like the rule that if the object is light at the top and dark at the bottom then this is a convex object, for example. Contents are related to the intrinsic or non-intrinsic states of the system by a mapping provided by the theorist; that is to say, a correlation between the states of the system and the content. I suggest that we think of the content as part of a model of the system, in the sense in which this word is used in the philosophy of science.

It is a familiar claim in recent philosophy of science that scientific theories are ‘collections of models’; but given the variety of things the word ‘model’ has come to mean, some clarifications are needed. The original proposal that theories should be thought of in terms of models is normally traced back to Patrick Suppes's work in the 1960s, subsequently developed by Bas van Fraassen.Footnote ¹⁴ Suppes and van Fraassen used the word ‘model’ in the sense of model theoretic semantics; a model of a theory is a collection of objects which renders true the claims of the theory. Hence the association of the idea of a theory as a collection of models with the label ‘the semantic view of theories’, contrasted with the ‘syntactic’ view of theories defended by the logical empiricists. ‘Semantic’ is appropriate because scientific models were being conceived of in terms of model theory, the standard semantic framework for formal languages.

Later work on models by Ronald Giere and others emphasised something quite different: the use of simplified mathematical structures (e.g. equations describing ideal populations), imagined comparisons (the atom is like a solar system) or even concrete objects (the actual wire and wood construction which represents the double-helix structure of DNA).Footnote ¹⁵ All these things are classified as models by scientists, and philosophers of science attempted to make sense of them. But models in this second sense look very different from the models of model theory, as a number of writers have emphasised.Footnote ¹⁶

In model theory, a model is collection of objects and operations on these objects – a set-theoretic structure – which makes the sentences of a theory true. (Nothing metaphysically weighty is meant by ‘making true’ here; this is just a standard definition of a model.) Even if model theory can be used to illuminate scientific theorising, as Suppes and van Fraassen argue, it is plain that a model-theoretic model does not look much like Rutherford's solar system model of the atom. For the solar system does not make true any claims about atoms; and nor does the comparison between the solar system and the atom make this true. Rather, the solar system is used to represent an aspect of how things are with the atom, just as equations describing an idealised population in biology is used to represent an aspect of how things are with some real population. Neither the equations nor the idealised population can be described as making true the claims about the real population.

The model-theoretic conception of models makes more sense if we think of the model as the comparison between the solar system and the atom. Here we might consider the model to be the mapping itself (sometimes described as an isomorphism, as it is by Suppes) between aspects of the real-world system and those of the object used to represent it. Whether this is a good way to understand models in science is much discussed in the philosophy of science; but what is clear is that it is a different idea from that of those who talk of one system being a model of another. Peter Godfrey-Smith puts this point well:

In what follows, I will talk of models as being what Godfrey-Smith calls the ‘model system’, and not as the ‘model description’ (i.e. the description mapping the system to the real-world system under investigation). To the extent, then, that the semantic conception of theories is tied up with the model-theoretic conception of models, we should avoid talking about ‘the semantic conception’ of theories here.

Models, in the sense I intend, are used to understand the behaviour of real-world systems by being used to represent how things are with those systems at a time, or how they evolve across time. Typically, the models are simpler than the real-world systems under investigation. They may involve idealisations (frictionless planes), or even empirical falsehoods (rational actor models in economics), and they may be unspecific in certain respects (a model of a cell may leave out information relating to what kind of cell it is).Footnote ¹⁸ The point of the model is to facilitate understanding of the real-world system by examining the behaviour of the model system: it is what Michael Weisberg calls an ‘indirect theoretical investigation of a real-world phenomenon’.Footnote ¹⁹

If we apply this to the case of the brain, the following picture emerges. The brain is the system under investigation (the ‘real-world system’) and the theorist attributes to it certain states, in order to predict or explain certain outputs. The theorist models these states by relating them to the abstract objects which are what I call the contents of the system. They expect to give a better understanding of the transitions between states of the brain by relating these states to contents than they would by merely citing neurochemical interactions, say, or gross external behavioural changes. This is what is going on when Frith claims that ‘the brain has this simple rule built into its wiring; it uses this rule to decide whether an object is concave or convex’. The rule is that because light generally comes from above, concave objects will be dark at the top and light at the bottom; convex objects will be light at the top and dark at the bottom. That doesn't mean that those words are written in the brain, that those words are written in English or any other language, or even in the Language of Thought. The attribution of content is, rather, an abstraction away from the activity of the brain in a way that can help you with predicting and explaining what's going on. The claim, then, is that the process in the brain is understood by modelling it with a rule relating shading to convexity and concavity.

As Giere has emphasised, modelling works by exploiting relations of similarity.Footnote ²⁰ What is similar to what, in the case of the brain? It's not that the brain state, whatever it is, resembles an abstract object, any more than a population resembles the equations used to model its growth. It's rather that the movement between states is similar to the stages of an explicit inference from the rule about concrete objects, plus the input from the image, to the conclusion about how the object looks. Inferences relate propositions; so the claim is that what is going on in the brain resembles a relationship between propositions. The appropriate comparison here is between modelling the brain with an inference, and modelling some target system by using a mathematical model.

Notice that this idea contains an echo of Fodor's famous argument for the Language of Thought based on the nature of mental processes.Footnote ²¹ I agree with Fodor that there is this similarity; but I resist the jump to the conclusion that this gives us a reason to say that there are symbols in the brain. Once equipped with a proper understanding of models, there is no need to move to this more outlandish hypothesis.

This does however raise an important question. What is the difference between a theory which models the transitions among states of the brain by relating them to representational contents (i.e. treats the states as representations) and one that merely treats the transitions as law-governed processes? When looking at the Kanizsa triangle, for example, normal perceivers see a white triangle as occluding segments of three black circles at its corners. Some neuroscientists and psychologists talk about the representation being ‘completed’ in the visual system by means of it making certain ‘assumptions’ about objects and how they normally relate to one another. On this understanding, the brain is conceived of as making an inference: in moving from state to state, its states are modelled by contents. But on another understanding, the brain is simply a law-governed system which produces certain visual outputs by being governed by a law about how objects normally look. On this view, there is no reason to call this an inference, not least because an inference must be rationally sensitive to further information, and this one isn't: no matter what you know, what you see will not change.

To determine whether it is correct to model a brain process by relating it to an inference will depend on the details of the case. My aim here is not to defend any particular hypothesis that treats the brain as making an inference; my aim is only to argue that it makes sense, and to say something about what kind of sense it makes.

3. The Propositional Attitudes

My discussion so far has been about things going on the sub-personal level; but does the modelling picture apply to the states of the whole person? In particular, does it apply to beliefs, desires, hopes and the other propositional attitudes? Some philosophers may agree that the modelling picture makes sense applied to sub-personal states but not to personal-level states; perhaps because these states have ‘original’ rather than ‘derived’ intentionality.Footnote ²²

Here I want to resist this objection. In the remainder of this paper, I will argue that ascriptions of propositional attitudes employ propositions as models in a similar way that ascriptions of sub-personal states employ contents as models. But I don't say this because I reject the distinction between original and derived intentionality. Rather, I would draw the significant distinction in this area not between sub-personal states and personal-level propositional attitudes, but between intrinsically intentional states of consciousness and all the other mental states. However, I cannot defend this further thesis in this paper; so I will confine myself to the claim about models.Footnote ²³

Contemporary philosophy of mind contains two influential views about the propositional attitudes. The first is that the concept of intentionality, the mind's representation of the world or direction upon its objects, should be understood entirely in terms of the propositional attitudes. This is a thesis that's been held by many people – Donald Davidson and many others. I call it ‘propositionalism’.Footnote ²⁴

The second view – which I will call the ‘relational thesis’ – is that the propositional attitudes should be thought of as (literally) relations to propositions: a propositional attitude like believing that the sun is shining is a relation between the person and the thing believed, namely, the proposition that the sun is shining. When you believe that the sun is shining you have one kind of relation to it, and when you hope that the sun is shining you stand in another kind of relation to it. The conjunction of propositionalism with the relational thesis implies that intentionality should be understood in terms of relations to propositions.

For a long time the relational thesis has been something of a dogma in analytical philosophy of mind. Fodor made his argument for this thesis a cornerstone of his intentional realism: ‘Believes looks like a two place relation so it relates two things the thinker and the proposition and it would be nice if our theory of belief permitted us to save the appearances’.Footnote ²⁵ And in a famous paper, Hartry Field said a similar thing: ‘propositional attitude attributions appear to relate people to non-linguistic entities called propositions’ and he then claims that this fact is a problem for materialists.Footnote ²⁶ The picture we are given is of a real relation to a proposition – the proposition is the thing you believe, and your belief state is a relation to it.

Before deciding whether this claim is true, there is a prior question: what does it mean? What does it mean to say that you're ‘related to a proposition by the relation of believing’, if that is not just another way of saying that you believe it? What's the point of saying that you stand in a ‘real relation’ to a proposition?

The picture I want to reject is that it is a basic psychological or metaphysical fact, something that needs to be explained, that people are related to propositions. But this does not mean that it is not true. Rather than being something that needs to be explained, it is rather part of the theoretical explanation or description of your state of mind. In other words, we should think of the ascription of propositional content to a person – placing someone in a relation to a proposition – as a way of modelling their mental state.

When we say that someone believes that p, or thinks that p, what we are doing is picking out a feature of their state of mind – their whole psychological outlook, or world picture – by relating it to this abstract object, the proposition p. The proposition serves to pick out part of the way the subject represents the world. We use a sentence to express the proposition, but we shouldn't think of the belief simply as a relation to a sentence, since the same feature of the state of mind can be expressed by different sentences (in the same or different languages). So for example, if we say that someone thinks Scotland should be an independent country, we are picking out an aspect of their world view, by relating them to the proposition expressed by the sentence ‘Scotland should be an independent country’ and all sentences that mean the same.

It is not plausible that someone could believe this without believing, for example, that some countries should be independent; this is an instance of the well-known ‘holism of the intentional’, which should be accepted by everyone. This means that if we attribute the first belief we should also attribute the second, unless we have some countervailing reason not to. Why this is so, is a question which divides theories of mind and intentionality: ‘inferentialists’ take these facts to be the basis of intentionality, ‘representationalists’ take them to be explained by facts about intentionality. Here my aim is not so much to contribute to this debate but to point to the role of ‘the proposition’ in describing beliefs and the relationships between them.

In identifying a belief by using a proposition, we are exploiting the logical and conceptual relationships between propositions to reflect the holism of the intentional. The fact that if you are ascribed the belief that Scotland should be an independent nation means that you can also be ascribed the belief that some countries should be independent mirrors the fact that the proposition that Scotland should be an independent nation entails that some nations should be independent. I presuppose here that propositions stand in logical and conceptual relationships – they are inconsistent, they entail one another, they support one another and so on. On the assumption that we have a fairly good understanding (or at least an agreement) about which logical and other relationships hold between propositions, we can then use this understanding or agreement to underpin our conjectures about propositional attitudes. It is because we understand that A entails B that we can use this understanding to say that ‘if someone believes that A then they will/should believe that B’.

This is why I say that the relation to a proposition is a way of modelling a belief. Just as Rutherford used our antecedent understanding of the structure of the solar system to model or picture something we are trying to understand – the atom – so we can use our antecedent knowledge of the relations between propositions to model the relationships between beliefs, and thus various aspects of the beliefs themselves. This is one feature of models: they use something which is in some way already understood to enlighten us about something that is less understood. Models do this by idealisation. If you say that someone believes that Scotland should be an independent nation, you do not thereby say everything about their conception of Scotland, nations and independence. They may have a very complex conception which is not easily summarised by a single sentence; yet it may still be perfectly true that they believe this, and in that sense the proposition models what they believe. As Robert Cummins puts it, ‘attributing a belief is going to be a bit like attributing a point of view to an editorial’.Footnote ²⁷ You can attribute a point of view to an editorial even if in doing so, you don't actually pick out a sentence that is contained in that editorial.

I mentioned above that when we attribute a belief, we typically commit ourselves to attributing certain related beliefs; or rather, we commit the believer to having certain related beliefs. This is the holism of the intentional. But of course, people do not always believe the logical consequences of what they believe, and our belief ascriptions should reflect this fact. Thinking of propositions as modelling beliefs captures this situation: the model is an idealisation. We say that if someone believes A then they should believe B but that may be because proposition A entails proposition B, or because it gives strong inductive support to B, or because it provides some other kind of reason for believing B. We might expect that a thinker will have the second belief if they have the first; this is a reasonable idealisation. But there might be all sorts of reasons why they don't have the second, and if so we would have to revise our model in the light of other things we find out about them. (I do not mean to imply here that thinkers are obliged to believe all logical consequences of what they believe, only that there are some cases where they do have an obligation to believe some of the consequences.)

There are two central questions about the propositional attitudes which this modelling approach answers. The first concerns what kind of objects propositions themselves are; the second concerns which specific propositions (however conceived) actually are the contents of any particular belief.

The first question is about the metaphysics of propositions. Some philosophers (Russellians) think that the constituents of propositions are objects and their properties; some (Fregeans) say that the constituents of propositions are senses or ‘modes of presentation’; others (Lewisians/Stalnakerians) say that a proposition is a set of possible worlds.Footnote ²⁸ Which view is right? Which objects are thinkers related to in their propositional attitudes?

There are things to be said in favour of each view. For example, sets of possible worlds are formally tractable – relations of entailment (etc.) can be understood set-theoretically. But there are also aspects of the possible worlds view of propositions which are very bad for representing distinctions between mental states. As has been observed many times, if a proposition is a set of possible worlds, then a necessarily true proposition is the set of all possible worlds. So all necessarily true sentences express the same proposition; and all necessary false sentences express the same proposition. We could take this to be a reductio ad absurdum of the view, but I don't think we should do this. Rather, I think what it shows is that propositions as sets of possible worlds constitutes a partial model of certain states of mind. It is a model which is good for some things and not for others.

A similar thing can be said about the view that propositions are Russellian. This too is a partial model. It might express the fact that the object of your thought is that very particular object you are thinking about, and no other object will do. Similarly the Fregean view that distinguishes propositions (Gedanken) more finely than objects and properties may be used to express a finer-grained perspective on a subject's mental life. But we should resist the idea that only one of these views is correct in giving the content of someone's belief, that this and not that is the proposition they are related to. The modelling picture allows that people can be related to propositions of all these kinds; in other words, objects of all these kinds can be used to pick out their beliefs.

In a recent insightful discussion of this subject, Ian Rumfitt says:

Unlike Rumfitt, I am happy to say that ordinary belief and speech reports do place thinkers in relation to propositions, since I think that ascribing a belief is modelling, and this places the thinker in a relation to a proposition, in a metaphysically innocuous sense. But this is a minor disagreement; if we read Rumfitt as saying that our ordinary maxims for reported speech and belief do not determine that one kind of proposition rather than another are the objects of belief or speech, then his remarks here get the matter exactly right.

The second question which the modelling picture answers is about which specific proposition actually is the content of a given belief, once we have settled on a general type of proposition. How specific or determinate should the content of a belief be, for example? And how should we identify the concepts which a thinker employs when they have a given belief? These questions arise particularly starkly in connection with animal belief. Consider Norman Malcolm's example of whether a dog believed that a cat went up a certain tree.Footnote ³⁰ Some say that if the dog believes that the cat ran up the tree, this requires that the dog has the concept of a tree, and this requires that it can think of a tree as something living, and the tree has leaves and the tree is made of wood … Most or all of these things we believe about trees. But how can a dog believe any or all of those things? The same applies to the beliefs of children. When a child believes – to use Dennett's example – that her father is a doctor, what exactly does she believe? If beliefs are determinate states of subjects, then surely there must be an answer to this question.

The modelling view gives an answer. The child and the dog have a conscious perspective on the world, a point of view. What we do when we ascribe them a belief is to attempt to make sense of their behaviour (running towards the tree, saying ‘Daddy is a doctor’) by identifying some feature of their point of view. This feature we pick out using a proposition. The proposition can truly describe part of that point of view – after all, it is a tree that the dog is thinking about, and it is being a doctor that the child ascribes to her father. But that does not require that the dog or the child means tree or doctor in exactly the same sense that we do. The relation to the proposition is a partial model.

And the same applies to more sophisticated thinkers. There is normally not just one sentence that expresses what you believe – in other words, there are many propositions that characterise your belief in different ways. As you struggle to express your beliefs, what you're trying to do is fix on what is the most appropriate way of expressing them. This does not mean that there is no fact of the matter about what you believe, that it is ‘simply’ a matter of interpretation. I say there is such a fact; but it is often a very complex fact, because of the holism of the intentional. Even where the simplest beliefs about your perceived environment are concerned, what you believe about here depends on so many beliefs about other things, that we should not expect that one sentence can adequately capture it all. It might be replied that a very long sentence could do it; maybe, but the suggestion is practically worthless when it comes to understanding real belief ascriptions.

It should be obvious that the view I am defending here owes a lot to Daniel Dennett's views about intentionality and belief-ascription.Footnote ³¹ But this does not mean that the view is some sort of anti-realism or instrumentalism about the mental (regardless of where Dennett himself stands on this vexed issue). On the contrary; I claim that my view involves a robust psychological realism. There is a psychological reality out there: this is the reality that you're attempting to model by using propositions in these various different ways. And just as the modelling conception of the atom is compatible with a hard-headed realism about atoms, so the modelling conception of the propositional attitudes is entirely compatible with realism about the psychological. Psychological reality is precisely that which we are trying to model in our personal and sub-personal psychological ascriptions.Footnote ³²