Language and the Brain

Tan: the case that started a science

In April 1861, a fifty-one-year-old man named Louis Victor Leborgne was transferred to the surgical ward of the Bicêtre hospital outside Paris, under the care of the surgeon and anthropologist Paul Broca. Leborgne had been in the institution for twenty-one years. He had gradually lost the ability to speak. By the time Broca saw him, his speech had reduced to a single syllable, which he repeated, usually twice, in answer to any question: tan. He became known on the wards as Tan, and he is known by that name still.

The crucial observation, and the one that made the case matter, is what Leborgne could still do. He was not demented. He was not deaf. He could gesture, he could indicate numbers with his fingers, and he understood a great deal of what was said to him. He could vary the intonation of his one syllable to convey emotion. What had gone was the machinery of speech output, and it had gone while leaving comprehension, and evidently intelligence, standing.

Leborgne died within days of Broca's examination. Broca performed a post-mortem and found a lesion in the posterior part of the third frontal convolution of the left hemisphere: the region now called the left inferior frontal gyrus, and, by convention, Broca's area. In 1865, after further cases, Broca made the claim that made his name: we speak with the left hemisphere.

Thirteen years later, in 1874, a twenty-six-year-old German physician named Carl Wernicke described the mirror image. His patients had damage not to the frontal lobe but to the posterior part of the superior temporal region of the left hemisphere, and their deficit was the opposite of Leborgne's. They spoke fluently. The rhythm and melody of their speech were normal, the grammar was superficially well formed, the words came easily. And the speech was largely empty of meaning: a stream of substituted, invented, or wrongly chosen words. Worse, they could not understand what was said to them.

Why the classic model was a triumph, not a mistake

It is easy, and lazy, to treat the classic model as a quaint nineteenth-century error. It was not. It was one of the best pieces of reasoning in the history of the brain sciences, and understanding why is the only way to understand what its successor had to beat.

Wernicke, and after him the physician Ludwig Lichtheim, did not simply record two syndromes. They built a mechanism. The proposal, in what became the Wernicke-Lichtheim model, ran like this. A store of the sound-images of words sits in the posterior temporal region: this is what you need in order to recognise a word you hear, and it is Wernicke's area. A store of the motor programmes for producing words sits in the inferior frontal region: this is what you need in order to say a word, and it is Broca's area. To repeat a word you have just heard, you must get from the first to the second, so the two must be connected. The connection was identified with a great arching bundle of white matter running from the temporal lobe forward to the frontal lobe: the arcuate fasciculus.

Now comes the part that makes this science rather than description. If the model is right, a lesion that cuts the connection while sparing both stores should produce a patient who is fluent, because the production store is intact; who comprehends, because the comprehension store is intact; and who nonetheless cannot repeat what he has just heard, because the route between the two has been severed. No such patient had been described. The model predicted one anyway.

That patient exists. The syndrome is called conduction aphasia, and its cardinal feature is exactly what the model said it would be: fluent, comprehending speech with a selective and often dramatic failure of repetition, together with frequent self-corrected errors in word sounds, as the patient hears that the output is wrong and tries again. A model that generates a novel, risky prediction and has it confirmed is a good model. It is what a scientific theory is supposed to do.

So the classic model is not to be dismissed. It is to be superseded, which is a different and more respectful operation. The question this page has to answer is: superseded on what evidence, and by what?

The aphasias: the evidence the model was built on

Everything in this field begins with aphasia: an acquired loss of language caused by damage to the brain, most often from a stroke in the territory of the left middle cerebral artery, and sometimes from tumour, trauma, or degenerative disease. The word is worth defining sharply, because what it excludes is as important as what it includes.

Four classical syndromes carry most of the historical weight. They are described here as the classic model described them, because that is the evidence base the model was answering to. All four are real clinical pictures. What is not real is the neat one-region-one-syndrome mapping that was drawn onto them.

The awareness asymmetry across the first two of these is one of the most quietly devastating findings in clinical neurology, and it is not a footnote. It is a clue. Monitoring your own speech for sense requires that you can comprehend it. In Broca's aphasia, comprehension is largely intact, so the patient hears their own broken output, recognises it as wrong, and is tormented by the gap between the sentence they intend and the fragment they produce. In Wernicke's aphasia, the comprehension system is itself the damaged one, so the internal monitor is gone. The patient talks fluently, hears nothing wrong, and cannot understand why listeners look confused. Sometimes they conclude that the problem is with the listener.

That asymmetry tells you something the two-blob diagram cannot: comprehension is not just an input function sitting in a box. It is also the feedback loop by which production is checked. The moment you notice that, the tidy separation of the two stores starts to look less tidy.

Four findings that broke the classic model

The classic model was not overturned by one dramatic refutation. It was eroded, over decades, as four different problems accumulated, each individually survivable and collectively fatal. This is the core of the page, so each is set out plainly.

1. Damage confined to Broca's area does not reliably produce Broca's aphasia

   This is the single most damaging finding, and it is the one most often left out of introductory accounts. Patients with lesions restricted to the left inferior frontal gyrus frequently show a transient speech disturbance that resolves, sometimes within weeks. They do not typically end up with the chronic, agrammatic, effortful syndrome that bears Broca's name. To produce a persistent non-fluent aphasia, the damage generally has to be far larger: extending into the underlying white matter, into the insula, into adjacent frontal and parietal cortex, and often into the deep tracts that carry information in and out of the region. In other words, the syndrome is not the signature of the area. It is the signature of a large lesion that happens to include the area, and the white matter beneath it may matter more than the cortex above it.

2. Nobody agrees where Wernicke's area is

   This sounds like a technicality. It is not. If you compare the maps drawn in different textbooks and papers, "Wernicke's area" is placed variously in the posterior superior temporal gyrus, extended back into the supramarginal and angular gyri of the parietal lobe, extended down into the middle temporal gyrus, or drawn as some combination of these. There is no consensus boundary. A discrete organ with a specific function ought to have a location that specialists can agree on. Persistent disagreement about where a region is, after 150 years of looking, is a strong hint that the region is a label applied to a stretch of cortex rather than a natural functional unit.

3. The regions are not language-specific

   The classic model treats Broca's area as a store of speech-motor programmes. But functional imaging shows Broca's area, and its right-hemisphere counterpart, activating in tasks that involve no language at all: learning and executing structured motor sequences, tracking hierarchical structure in music or in artificial grammars, and holding and manipulating information in working memory. Its neighbourhood in the prefrontal cortex is exactly where you would expect to find machinery for sequencing and hierarchy, and that is a plausible description of what it contributes to language rather than a description of language itself. A region that also does non-linguistic sequencing is not a speech organ. It is a general-purpose piece of circuitry that language recruits.

4. The historical evidence was thinner than the textbook implies

   Broca's original case series was small, and the anatomical claim rests on far less than a century and a half of confident repetition suggests. Leborgne's brain was preserved rather than sectioned, which was fortunate, because it meant it could be examined again with modern methods. When Dronkers and colleagues put the preserved brains of Leborgne and of Broca's second patient, Lelong, into a high-resolution MRI scanner (Brain, 2007), they found that the damage in Leborgne's brain extended well beyond the classic Broca's area: into the insula, into deeper structures, and, importantly, into the underlying white matter, including the fibre pathways running beneath the region. The surface lesion Broca saw was the visible part of something considerably larger. The founding case of localisation was, on re-examination, a case of a big lesion.

What replaced it: two streams, not two spots

The account that now dominates the field was set out by Gregory Hickok and David Poeppel in Nature Reviews Neuroscience in 2007, and it is known as the dual-stream model. It borrows its logic, deliberately, from vision, where a comparable division into a ventral "what" pathway and a dorsal "where and how" pathway had already proved its worth.

The starting point is a question the classic model never really asked. What does the brain actually have to do with speech? Two quite different things, as it turns out, and they pull in opposite directions.

The first job is to get from a pattern of sound to a meaning. You hear the acoustic smear of the word "harbour" and you must arrive at the concept. This is a mapping from a highly variable signal onto a stable, abstract representation, and it must be robust: it has to work across accents, across speakers, in noise, at speed.

The second job is to get from a pattern of sound to an articulation. You hear a word you have never heard before, a foreign name, a nonsense syllable, and you have to say it back. Meaning cannot help you here, because there isn't any. What you need is a translation between the auditory code and the motor code, a sensorimotor interface that converts "what it sounds like" into "what my mouth must do".

These are different computations, and the model's claim is that the brain runs them in different places.

Now watch what this explains, because this is where the model earns its keep. It resolves a fact the classic model could never account for, and which sits in plain sight in every stroke unit.

The dual-stream framework also re-describes the old syndromes in a way that fits the lesion data better. Conduction aphasia, on this account, is not the severing of a cable between two stores; it is damage to the dorsal sensorimotor interface itself, which is exactly why repetition, the task that requires a sound-to-articulation mapping and cannot be rescued by meaning, is the function that fails most selectively. Classical Wernicke's aphasia, with its poor comprehension, tends to follow lesions large enough to disrupt both posterior temporal cortex and a good deal besides, rather than a small strike on a single named area.

None of this means language has no anatomy. It has a great deal of anatomy. It means the anatomy is a pair of distributed processing streams, dependent as much on the white matter joining regions as on the regions themselves, and not a pair of dots joined by a line.

Lateralisation: the numbers, and what they do not mean

Language is the clearest example of hemispheric lateralisation: the fact that the two halves of the cerebrum, which look nearly identical, are not functionally identical. It is worth having the actual figures, because this is a topic on which vague impressions do enormous damage.

The most useful dataset comes from Knecht and colleagues (Brain, 2000), who used functional transcranial Doppler ultrasound, a non-invasive measure of blood-flow changes in the two hemispheres during a word-generation task, in healthy volunteers spanning the range of handedness. Their finding, in round numbers:

Read that carefully, because the popular version gets it backwards. Being left-handed does not mean your language is on the right. Roughly three left-handers in four are left-dominant for language, just like right-handers. Handedness shifts the probability of atypical dominance upwards, and does so in a graded fashion as left-handedness becomes more pronounced, but it does not flip the picture. A left-handed patient who has a left-hemisphere stroke is, on the odds, still at serious risk of aphasia.

Two further points keep the numbers honest. First, "dominance" is a statement about which hemisphere carries the heavier load, not a claim that the other one is idle. Even in a strongly left-dominant brain, the right hemisphere is doing real linguistic work: prosody, the melody and stress that carry emotional tone; aspects of discourse, inference, and the interpretation of metaphor, irony, and jokes. Patients with right-hemisphere damage can produce and understand grammatical sentences and still be strikingly bad at getting the point of a story or hearing that a remark was sarcastic. Second, lateralisation is a fact about language, and, as the myths section below insists, it licenses no conclusion whatever about personality.

The split-brain evidence, told accurately

The two hemispheres communicate through the corpus callosum, a thick sheet of roughly two hundred million axons crossing the midline. In the mid-twentieth century, a small number of patients with severe, drug-resistant epilepsy underwent surgical section of the corpus callosum, to stop seizures spreading from one hemisphere to the other. The operation helped them. It also created, inadvertently, the most informative experiment in the history of lateralisation research, exploited above all by Roger Sperry and Michael Gazzaniga.

The method exploits an anatomical fact about vision. What falls in the left half of the visual field is projected to the right hemisphere, and vice versa. In an intact brain this hardly matters, because the callosum shares the information across within milliseconds. In a split brain, it matters enormously: information flashed to one visual field stays in one hemisphere.

Now be careful about what this shows, because this is the point at which a genuinely important experiment gets turned into a genuinely stupid slogan.

What it does show. That the two hemispheres can be functionally dissociated: cut the connection, and each can hold information the other lacks. That speech output is, in these patients, a left-hemisphere function: the hemisphere that talks is the one that did not see the object, so it reports nothing, and the hemisphere that saw the object cannot talk. That the right hemisphere is not empty: it recognised the object, retrieved its identity, and directed a hand to find it. Gazzaniga's later work also documented the left hemisphere's readiness to confabulate, generating confident verbal explanations for actions that its right-hemisphere partner had in fact initiated for reasons it had no access to, which says something uncomfortable about how readily any of us explain our own behaviour.

What it does not show. It does not show that the hemispheres are two personalities, or two styles of thought, or that one is analytical and the other artistic. These patients had their hemispheres surgically disconnected, and even so, outside the laboratory, they behaved as unified people. In an intact brain, with 200 million fibres carrying traffic across the midline continuously, the hemispheres are not two minds having a difficult conversation. They are one system. The split-brain work tells you what is lateralised. It tells you nothing about what you are like.

Recovery: why a language area is not a life sentence

Because the classic model presented language as living in particular boxes, it carries an implicit and cruel corollary: destroy the box and the function is gone for good. Clinically, this is false, and the reason it is false matters for the science as much as for the patient.

The single most important variable is age. In young children, the language system has a remarkable capacity to relocate. Following extensive damage to the left hemisphere early in life, including, in extreme cases, surgical removal of most of the left hemisphere to control catastrophic epilepsy, children can go on to develop functional language, largely supported by the right hemisphere. The outcome is often not perfectly normal, and syntax tends to suffer most, but the gulf between this and the adult outcome from comparable damage is enormous. This is neuroplasticity operating at its most impressive, and it is why the concept of a sensitive period for language acquisition has real neural teeth.

In adults, recovery is more modest but it is real. Most people with post-stroke aphasia improve, and improvement continues for months, in some cases years, well past the window of spontaneous recovery in which oedema resolves and undamaged tissue resumes function. Imaging studies of recovering patients show reorganisation of activity in perilesional left-hemisphere tissue and, particularly after large left lesions, increased recruitment of right-hemisphere homologues, that is, the mirror-image regions on the other side. Speech and language therapy is effective, and intensity matters. The brain that recovers language is not restoring the old wiring. It is finding another route.

What none of this licenses is optimism without honesty. Severe global aphasia after a large left middle cerebral artery stroke frequently leaves lasting, life-altering impairment, and no amount of plasticity talk should be allowed to obscure that. The claim is narrower and truer: damage to a language region is not automatically the permanent loss of language, and the outcome depends on the size and location of the lesion, the age of the patient, and the therapy that follows.

Myths this subject attracts

Related topics