Finding Language in the Brain
Brought to you by Curio
What exactly is language? At first thought, it’s a continuous flow of sounds we hear, sounds we make, scribbles on paper or on a screen, movements of our hands, and expressions on our faces. But if we pause for a moment, we find that behind this rich experiential display is something different: the smaller and larger building blocks of a Lego-like game of construction, with parts of words, words, phrases, sentences, and larger structures still.
We can choose the pieces and put them together with some freedom, but not anything goes. There are rules, constraints. And no half measures. Either a sound is used in a word, or it’s not; either a word is used in a sentence, or it’s not. But unlike Lego, language is abstract: Eventually, one runs out of Lego bricks, whereas there could be no shortage of the sound b, and no cap on reusing the word “beautiful” in as many utterances as there are beautiful things to talk about.
Language is a calculus
It’s tempting to see languages as mathematical systems of some kind. Indeed, languages are calculi, in a very real sense, as real as the senses in which they are changing historical objects, means of communication, inner voices, vehicles of identity, instruments of persuasion, and mediums of great art. But while all these aspects of language strike us almost immediately, as they have philosophers for centuries, the connection between language and computation is not immediately apparent — nor do all scholars agree that it is even right to make it.
It took all the ingenuity of linguists, like Noam Chomsky, and logicians, like Richard Montague, starting in the 1950s, to build mathematical systems that could capture language. Chomsky-style calculi tell us what words can go where in a sentence’s structure (syntax); Montague-style calculi tell us how language expresses relations between sets (semantics). They also remind us that no language could function without operations that put together words and ideas in the right ways: The sentence “I want that beautiful tree in our garden” is not a random configuration of words; its meaning is not completely open to interpretation — it is the tree, not the garden, that is beautiful; it is the garden, not the tree, that is ours.
Language in the brain
At this point, most linguists would probably be content with saying that calculi are handy constructs, tools we need in order to make rational sense of the jumble that is language. But if pressed, they would admit that the brain has to be doing some of that stuff, too. When you hear, read, or see “I want that beautiful tree in our garden,” something inside your head has to put together those words in the right way — not, say, in the way that yields the message that I want that tree in our beautiful garden.
Linguists, logicians, and philosophers, for at least the first half of the 20th century, resisted the idea that language is in the brain. If it is anywhere at all, they estimated, it is out there, in the community of speakers. For neurologists such as Paul Broca and Carl Wernicke, active in the second half of the 19th century, the answer was different. They had shown that lesions to certain parts of the cerebral cortex could lead to specific disorders of spoken language, known as “aphasias.” It took an entire century — from around 1860 to 1960 — for the ideas that language is in the brain and that language is a calculus to meet, for neurology and linguistics to blend into neurolinguistics.
If we look at what the brain does while people perform a language task, we find some of the signatures of a computational system at work. If we record electric or magnetic fields produced by the brain, for example, we find signals that are only sensitive to the identity of the sound one is hearing — say, that it is a b, instead of a d — and not to the pitch, volume, or any other concrete and contingent features of the speech sound. At some level, the brain treats each sound as an abstract variable in a calculus: a b like any other, not this particular b. The brain also reacts differently to grammar errors, as in “I want that beautiful trees in our garden,” and incongruities of meaning, as in “I want that beautiful democracy in our garden”: Rules and constraints matter. We are slowly figuring out how the brain operates with the abstract system that is language, how it arranges morphemes — the smallest grammatical units of meaning — into words, words into phrases, and so on, on the fly. We know that it often looks ahead in time, trying to anticipate what new information might arrive, and that words and ideas are combined by a few different operations, not just one, kicking in at slightly different times and originating in different parts of the brain.
Software and hardware?
The language-as-calculus idea may well be the best model of language in the brain we currently have — or perhaps the worst, except for all the others. Like all ideas in science, it has limitations. Like all crisp, powerful ideas, it can easily misguide. For example, it may seem to suggest that the language calculus is a program run by the brain. And in some sense it is, just not in the familiar sense of personal computers, of software and hardware.
Brains are peculiar computational environments, unlike anything humans have engineered. Neurolinguists like to say that words and their meanings are “stored” in memory and “retrieved” from memory. And in some sense they are, just not in the sense of personal computing. When I open a text file on my computer, I expect it to look exactly the same regardless of other files and browser tabs open at that time. This is not how information is “retrieved” in the brain. Some details of the word’s meaning that is activated will depend on context: The meaning of “tree” that one uses to derive an interpretation of “I want that tree in our garden,” its nuances and implications, may differ depending on whether we are looking at a potted artificial Christmas tree or at a rare olive tree.
The case of “storage” is subtler, more intriguing. One idea in neurolinguistics is that the human brain is “language ready”: Any infant can acquire any language on Earth, while no other animal can. But the infant brain is not ready for language in the way a brand new laptop is ready for Spotify and Zoom, right out of the box. If one could upload a language to a newborn’s brain, one would certainly not magically create a competent, fluent speaker of that language: The “hardware,” a developing — but still immature — brain, would not be ready to run the “software,” a full-fledged language. In language learning, the “software” would have to be programmed in gradually, as a function of the stage of growth and maturation of the “hardware.” That this is not how computers and programming work just shows that the software-vs-hardware metaphor is not quite right for language and the brain.
The future of an idea
Neurolinguists sometimes grumble that we know oh-so-little about language in the brain, and that there is still much to figure out. True, but what has been achieved is remarkable, especially in recent times. Accelerated histories are not uncommon in new fields of science — think of genetics, or informatics — but they raise questions about the accelerating factors: It is not always easy to tell real progress from hype. The language-as-calculus idea has been scrutinized and perfected by philosophers, logicians, linguists, and computer scientists. How long would a bad idea survive, caught in such crossfire? It has guided our best theory-building and experimental efforts. It has been a hard wall to bounce new ideas off against.
Still, there is a risk we take it too far, that we fail to see the difference between genuine aspects of the language calculus in the brain and what we can just model mathematically or simulate easily and elegantly “in silico.” No climate scientist would think a heat wave is a computational process just because they can simulate it in a computer. For those who study the mind and brain, making such distinctions is much harder. Where does the language calculus begin and end? Is it just about syntax, or do we also compute sound and meaning? And what about the rich experiential dimension of language? It is perhaps ironic, but ultimately a blessing, that no calculus, no algorithm, will give us these answers — only clear thinking, open criticism, and tireless, imaginative research.