Meanings as brain states
A natural thought about meaning is to identify it with brain states: understanding or intending a certain meaning, on this identification, would just be having the neurons of one’s brain in a particular configuration. From one point of view, this identification seems very plausible. After all, isn’t language ultimately a product of our brain? If we understood how the brain works, wouldn’t we understand all the details of language, semantic ones included? Isn’t it only the rudimentary state of our current understanding of brain processes that prevents us from giving the details of this identification? According to this line of thinking, semantics, along with the rest of linguistics, will one day be reduced to, or unified with, brain science, in the same way that the classical theory of genetics can be reduced to or unified with that of molecular biology. In other words, once brain science has progressed, we will no longer need the technical vocabulary of semantics, but will be able to talk wholly in terms of synaptic connections, neurotransmitters, proteins and so on. In just the same way, the modern language of molecular biology, involving chromosomes, nucleotides and DNA sequences, has at least partly replaced the older one of genes as the best way of describing the details of inheritance.
This is an attractive position in many ways: brain states must ultimately cause all behaviour, including language. But it is going too fast to conclude from this that we will eventually be able to identify meanings with brain states and reduce semantics to brain science. The first obstacle is that it’s hard to see how brain states could have the properties that meanings have. Meanings, the way we normally think of them, have mutual connections to each other – of synonymy, antonymy, class inclusion and so on. For example, the meaning ‘cat’ has the following relations with other meanings:
• it is an instance of a broader class of meanings, ‘mammals’, ‘domestic animals’, ‘four-legged animals’ and so on;
• it is, in some sense, the ‘opposite’ of the meaning ‘dog’;
• it can be synonymous with the meaning ‘feline’.
These are facts about the meaning of cat that we will presumably want a theory of semantics to refl ect. It is not clear, though, how this could hap pen in a theory which identified meanings and brain states. How can one brain state be the opposite of, or synonymous with, another? The brain is just physical matter; it makes as little sense to say that a state of brain matter is the opposite of, or synonymous with, another as it does to say the same of the state of the electrons in my computer at any one time. It therefore seems that meanings have a property which prevents them from being completely identified with brain states.
This problem is just one instance of the broader problem of intentionality. Intentionality is the term philosophers use to describe the essential aboutness or contentful nature of language. A word like cat has a certain psychological content: it refers to (is about) a certain class of creatures, cats. The same is true of a verb like sit: this refers to, or is about, a certain class of actions, the actions of sitting. Many philosophers think there is something deeply special about intentionality, in that it is a property that is distinctively mental: purely physical things like my brain or my computer, which consist of configurations of electrons, just aren’t the types of thing which can possess intentionality. Electrons, whether in my brain or in my computer, aren’t about anything; they’re just there. As a result, any attempt to simply identify something intentional like language with something non-intentional like a brain state cannot be successful.
How do we square this with the obvious truth that it is the brain that is ultimately responsible for linguistic production and understanding? If meaning is one of the factors to be taken into account in the production of utterances, and if brain processes will ultimately explain the whole production of utterances, then surely they must explain meaning too! It would just be illogical to say that everything that happens in language is determined by brain processes, and in the same breath to exclude meaning. Here we need to invoke the concept of levels of explanation or levels of description, an important notion in cognitive science discussed by Marr (1982). Attending to the notion of levels of explanation/description will show us that there is room for both intentional meanings and non-intentional brain states in our explanations of language.
Consider a computer chess program. There seem to be several levels on which we can describe and explain what the program is doing. The first is the informal level, on which we describe the computer as simply following the rules of chess with the aim of beating its opponent. A particular move might be described as ‘castling to protect the king’, for example, or ‘taking a pawn’, or ‘sacrificing the bishop’. This mode of description uses ordinary, everyday vocabulary of the sort we could also use for explaining people’s behaviour. It makes reference to beliefs, intentions, and desires: the computer’s belief that its king could be threatened, its desire to protect it, and its intention to castle in order to achieve this. In one way, the computer doesn’t really have beliefs, desires or intentions, of course, but we talk as though it does, since this is a useful way of describing and understanding what is happening. Let’s call this the intentional level of explanation.
A second, lower level of explanation is more detailed: this level consists in specifying the different steps of the program that the computer is running. This explanation wouldn’t use the ordinary terms of everyday intentional, psychological explanation, but would lay out the sequence of individual input and outputs that the computer processes. The computer has some way of representing the input positions (the position of the pieces after each move), and a set of algorithms for turning inputs into outputs in a way that makes it likely to win. Let’s call this the algorithmic level of explanation. Understanding this level will give us a detailed way of predicting what the computer is going to do next: if we have access to the specific program it is running, we can work out its next move in advance. Notice that there are several different ways in which the intentionally described actions the computer performs could be realized algo rithmically. There’s more than one possible chess program that a computer could run, yet all of them produce behaviour which is open to a single type of intentional explanation: the difference between different chess programs disappears at the intentional level of explanation where, whatever program the computer is actually running, we can still always describe it as ‘castling to protect the king’, ‘taking a pawn’, ‘sacrificing the bishop’ and so on. The details of the program become invisible as we move to the higher level.
Finally, there’s the lowest level, the level of implementation: this is the level of description/explanation which concerns the specific way in which the algorithm is instantiated physically in the particular machine involved. Just as a single fact on the topmost intentional level can correspond to several different states on the lower algorithmic level, so a single algorithm can be implemented in multiple ways in an actual physical machine. This is most obvious if we think about the difference between the most up-to-date type of computer, which runs on a solid-state drive, a conventional one using a spinning hard disk, and an old-fashioned one using magnetic tape or punch cards. All these machines can run the same algorithms, but the physical details of how they do so are completely different.
Clearly, all three levels of explanation are necessary to understand what is going on when the computer plays chess. Since there is a variety of possible physical realizations of the program on the implementational level, the next highest level, the algorithmic one, gives us a powerful way of abstracting from the details of the actual physical system that is performing the operations and describing the inputs and outputs of the program. But it is the intentional level that is the most relevant when we ask why the computer is behaving as it is. The intentional level, which consists of explanations like ‘protecting the king’, ‘taking a pawn’, ‘sacrificing the bishop’, makes sense of the computer’s actions as a chess-player, not just as a machine. The algorithms and their physical instantiations are just a meaningless sequence of actions if we can’t place them in the context that allows them to make sense, and it is only the intentional level that does this. Marr (1982) draws an analogy with trying to understand bird flight: we can’t understand bird flight by limiting ourselves to feathers, the implementational level. We have to go beyond this to look at the wider place of feathers within a complex of notions like lift, air pressure, energy, gravity, weight and so on. Studying the physical constitution of the feathers in the absence of these other considerations will be fruitless.
Language is arguably the same way. Studying brain states will only tell us how language is implemented. It will tell us nothing about the higher-level relations that tie this implementation in with the rest of our psychology. As a result, meanings are unavoidable as part of the explanation of utterances. If I tell you that my head is killing me, then part of the explanation for my utterance involves my belief that my head hurts, my desire to communicate this fact to you, and the fact that those words convey that idea as their meaning. I could have expressed the same belief in a number of different ways, for example by saying I’ve got a migraine, or my headache’s come back, or by clutching my head and saying the usual problem again in a long-suffering tone of voice. Since each of these utterances is expressed differently, they would correspond to different brain states. But we can concisely capture what they have in common by appealing to the level of their meaning: even though the brain states that produce them are different, they are united by the similarities of the meanings they convey. Just talking about brain states makes this elementary generalization impossible.
Especially at the current rudimentary stage of our knowledge of the brain, then, we have no choice but to continue to appeal to meanings in our explanations of language. Brain states are too complicated and too variable (both within and between individuals) to allow us to capture the straightforward generalizations we can capture using the intentional vocabulary of meaning. Meanings are the thread that guides us through the variety and confusion of brain states and input–output sequences; only by invoking meanings can we relate language to human behaviour and psychology in general. Understanding brain states will be important for understanding language, but not at the expense of meaning. Studying brain states will tell us how the brain does what it does. Studying meaning as part of an intentional level study of human psychology and behaviour will tell us what it is doing and why it is doing it. It is thus a confusion of explanatory levels to claim that meaning can be reduced to brain state.
