Broca and Wernicke: The Two Patients Whose Brain Damage Explained Language

Introduction

In April 1861, a man died in a hospital ward outside Paris. He was 51 years old. For the last 21 years of his life he had been unable to speak, apart from one syllable he repeated in every situation, with different intonations, as though the meaning lived in the melody rather than the sound. The syllable was "tan". The staff called him Tan. His name was Louis Victor Leborgne.

Six days before he died, a surgeon named Pierre Paul Broca examined him. After the death, Broca performed the autopsy, found damage in the left frontal lobe, and reported the case to the Société Anatomique de Paris. That report is the reason a region of your brain is named after a man who never saw a living brain work.

Here is the part almost nobody mentions. Leborgne's brain was not dissected and discarded. It was preserved, and it still exists. In 2007, Nina Dronkers and colleagues put it into a high resolution MRI scanner, along with the brain of Lelong, Broca's second patient. They looked inside the tissue that Broca could only look at from the outside [1].

The damage was bigger than Broca thought. It reached deeper than the surface lesion he described, extending into pathways running under the cortex, and into regions he never named. The observation that founded the localisation of language in the brain was made from the outside of a brain whose interior told a different story.

That is not a gotcha. Broca had no scanner and no way to see beneath the surface, and the discovery he made was real and enormous. It is, though, the beginning of a pattern that repeats through this entire subject for the next 160 years. Somebody makes a careful observation. The observation gets compressed into a simple map. The map gets taught. And then, slowly, evidence accumulates that the map is a rough sketch of something considerably more complicated, and the teaching does not change.

If you search for Broca and Wernicke today, you will find a first page of results that mostly teaches the 1874 version with better graphics. Two of the pages currently ranking put Broca's area in the wrong lobe. One of them inverts the logic of handedness and language. Almost none of them mention that a review published in 2016, in a serious journal, carried the title "Broca and Wernicke are dead" [2].

So this article does two things at once. It tells the story properly, with the patients named and the dates right, because the story is genuinely good. And it tells you which parts of what you learned have survived contact with modern imaging, lesion mapping and awake brain surgery, and which parts have not.

Some of what follows is settled. A lot of it is not. Where the field is still arguing, this article says so rather than picking a winner, because a topic this heavily taught deserves better than confident summary.

The Man Who Could Say One Word

Start with what Broca actually saw, because the popular version compresses it badly.

Leborgne had been admitted to the Bicêtre hospital 21 years earlier, having lost the ability to produce speech. Over those two decades his condition worsened. He developed weakness on the right side of his body, then paralysis, and by the end he had an infected and gangrenous right leg, which is what brought him to a surgeon rather than a physician. Broca was that surgeon.

What made the case famous was not the loss of speech by itself. It was the dissociation. Leborgne could understand what was said to him. He could respond appropriately with gesture and with the intonation of his one syllable. His intelligence, as far as Broca could assess it, was intact. What was gone was production, and only production.

Broca called the condition aphemia, a term that did not survive. He examined the brain and found a lesion centred on the posterior part of the third frontal convolution of the left hemisphere. Within months he had a second case, Lazare Lelong, an 84 year old man with a similar pattern of speech loss and a lesion in a similar location. Two patients, two lesions, one conclusion: the faculty of articulate language has a seat, and the seat is in the left frontal lobe.

Over the following years Broca accumulated more cases and made a further claim that mattered as much as the first. The lesions were on the left. Not sometimes on the left. Consistently on the left. The two hemispheres of the brain, which look like mirror images, are not doing mirror image work.

That claim reorganised neurology. It is also the claim that somebody else had made first, and lost.

Ronald Lazar and J.P. Mohr revisited Broca's contribution in 2011 and made a point worth holding onto [3]. Broca's lasting achievement was not really the discovery of a speech centre. It was the demonstration that a specific mental function could be tied to a specific piece of tissue at all, and that the method for establishing this was to compare behaviour in life with damage found after death. That method, the lesion method, is what he actually invented, and it ran the field for the next century.

The Man Who Was First and Lost

In 1836, 25 years before Broca stood up in Paris, a country doctor named Marc Dax presented a paper at a medical meeting in Montpellier. He had collected cases of patients who had lost speech, and he had noticed that the damage was on the left.

Nothing happened. The paper was not published. Dax died the following year.

His son, Gustave Dax, spent much of his career trying to establish his father's priority, and the resulting dispute is a small, bitter footnote to one of the largest discoveries in the history of neurology. Manning and Thomas-Antérion went back over the record in 2011 and laid out what can and cannot be established about what Marc Dax knew and when [4]. Stanley Finger and Daniel Roe made the case for the son in a 1999 paper whose title asks the question directly, whether Gustave Dax deserves to be forgotten [5].

Two things are worth taking from this. The first is small and factual, and it gets muddled constantly: Marc Dax and Gustave Dax are two different people, father and son, and articles that merge them into a single "Dax" are wrong. The second is larger. Left hemisphere dominance for language was noticed by more than one person, working independently, before anyone had a mechanism to explain it. That is usually a sign the underlying phenomenon is real and strong. It is one of the parts of this story that has survived everything since.

Wernicke's Turn, at Twenty-Six

Thirteen years after Broca's report, a young German physician published a monograph that completed the picture, or appeared to.

Carl Wernicke was 26 years old when Der aphasische Symptomencomplex appeared in 1874 [6]. He had noticed a different kind of language failure. His patients could speak fluently, with normal rhythm and normal grammatical scaffolding, but what came out did not mean much. They also could not understand what was said to them. The lesions in these cases were not in the frontal lobe. They were further back, in the temporal lobe.

Wernicke did something with this that was cleverer than simply adding a second box to the map. He proposed a system. If one region stores the sound patterns of words, and another region controls the motor programmes that produce them, and the two are connected, then you can predict the existence of aphasias that nobody had described yet. Damage the connection while leaving both regions intact, and you should get a patient who understands and speaks fluently but cannot repeat what they hear. That prediction is conduction aphasia, and it turned out to exist.

Ludwig Lichtheim formalised the scheme into a diagram with a small number of centres and the connections between them, and the whole apparatus became the connectionist or diagram-maker model of language. It is genuinely impressive work. It predicted syndromes before they were observed, which is the thing scientific models are supposed to do and mostly fail at.

It also had a weakness that would take a century to become obvious. The model treated brain regions as boxes holding functions, and connections as wires carrying content. That is a very natural way to think, and it is how almost everyone still explains the brain informally. Whether it describes what cortex actually does is a separate question.

The Model Goes Global

The diagram-maker approach fell out of fashion in the early twentieth century and then came back, hard, because of one person.

Norman Geschwind revived it. In 1965 he published a long two-part paper on disconnexion syndromes, arguing that a large class of neurological deficits arise not from damage to centres but from damage to the white matter connecting them [7]. In 1970 he set out the case for the whole framework in Science, in an article that became one of the most widely read statements of how language is organised in the brain [8].

This is the version that reached textbooks, medical schools, psychology courses and, eventually, the internet. It is why the model is properly called the Wernicke-Lichtheim-Geschwind model, and it is why a framework built between 1861 and 1885 is still the thing most people are taught in the 2020s.

The story, as it settled, goes roughly like this. Wernicke's area, in the posterior superior temporal gyrus, handles comprehension. Broca's area, in the inferior frontal gyrus, handles production. A bundle of fibres called the arcuate fasciculus connects the second to the first. Damage the front and you get effortful, non-fluent speech with preserved understanding. Damage the back and you get fluent, meaningless speech with impaired understanding. Damage the cable and you get a repetition problem.

It is clean, it is memorable, and it is teachable in four minutes. Those are exactly the properties that make a model hard to dislodge once the evidence turns.

The First Crack Appeared in 1978

The classical model was in trouble within a decade of Geschwind reviving it, and the paper that did the damage is barely known outside clinical neurology.

In 1978, J.P. Mohr and colleagues published a study in Neurology with the flat title "Broca aphasia" [9]. They looked at what actually happens when the damage is confined to the region Broca described. The answer was not Broca's aphasia. Patients with lesions restricted to that area had a speech disturbance that largely recovered. The persistent, severe, non-fluent syndrome that carries Broca's name requires damage well beyond Broca's area, extending into surrounding frontal cortex, the insula, and underlying white matter.

Read that again, because it is the single most important clinical fact in this article and it contradicts the thing everyone is taught. Broca's area damage does not reliably cause Broca's aphasia. The syndrome and the region do not line up.

And notice the date. This was 1978. Not 2016. The mismatch has been in the literature for nearly fifty years.

It fits, in retrospect, with what the 2007 scan of Leborgne's brain showed [1]. Leborgne had 21 years of progressive deterioration and extensive damage, not a tidy lesion in one gyrus. The founding case of Broca's aphasia was a case of large, deep, spreading damage. The tidy version was always a simplification of the tissue Broca held in his hands.

Broca's Area Is Not One Thing

The second crack is more recent, and it goes deeper, because it questions whether the region is a meaningful unit at all.

Start with anatomy. Katrin Amunts, Karl Zilles and colleagues mapped Broca's region at the level of cell architecture in 1999, examining the cytoarchitecture of Brodmann areas 44 and 45 across a set of postmortem brains [10]. Two findings mattered. The region is not architecturally uniform. And the borders differ substantially from one brain to another, in position and in extent, well beyond what the tidy textbook blob suggests.

Anwander and colleagues arrived at a similar conclusion from a completely different direction in 2006, parcellating Broca's area by its connection patterns rather than its cell types and again finding internal subdivisions [11]. Two methods, two sets of subdivisions, one message: the blob has structure inside it.

Then came the functional evidence. In 2012, Evelina Fedorenko, John Duncan and Nancy Kanwisher used single-subject fMRI, which means they identified each individual's language regions in that individual's own brain rather than averaging everyone together onto a template [12]. Averaging is what had been hiding the answer, because if the borders vary between people, the average smears them.

What they found settled a long argument by declaring both sides right. Inside Broca's area there are subregions that respond specifically during language processing and almost nothing else. Sitting immediately next to them, sometimes millimetres away, are subregions that respond during arithmetic, working memory tasks and cognitive control, the network usually called the Multiple Demand system. They are interleaved. A follow-up study in 2013 mapped the domain-general side in more detail [13].

This is why group-averaged imaging produced twenty years of contradictory results about what Broca's area does. Half the studies were reading the language subregions and half were reading their neighbours, and the average of the two is a region that appears to do everything vaguely and nothing precisely.

Fedorenko and Idan Blank pushed the conclusion to its logical end in 2020, in a paper titled "Broca's Area Is Not a Natural Kind" [14]. Their argument is that the term picks out a piece of anatomy with a nineteenth century name, and that piece of anatomy contains functionally distinct systems. Asking what Broca's area does is like asking what the postcode does. There are several things in there and they are not the same thing.

Two other findings make the same point from outside. Luciano Fadiga and colleagues reviewed evidence that Broca's area responds during action observation and during music processing, not only during speech [15]. And Mairéad MacSweeney and colleagues showed with fMRI that native users of British Sign Language recruit the classical left perisylvian language regions when processing sign [16]. Whatever Broca's area is doing, it is not driving the muscles of the mouth. Deaf signers using their hands activate it too. That single result kills the most common intuitive reading of the region, and it is worth remembering when you next see it described as the speech muscle area.

The Wernicke Conundrum

If the front half of the model has problems, the back half has worse ones.

Marsel Mesulam and colleagues published a study in Brain in 2015 that gave the problem a name [17]. They studied 72 patients with primary progressive aphasia, a condition in which language declines gradually as specific regions atrophy. Because the atrophy varies from patient to patient, the group gave them a wide spread of damage sites to correlate against a wide spread of language impairments, which is difficult to obtain from stroke patients.

The result was not what the classical model predicts. Neuronal loss in the temporoparietal areas traditionally counted as Wernicke's area left single word comprehension intact. Patients with substantial damage there could still understand words. Sentence comprehension was affected inconsistently. And the most severe comprehension impairments were associated with a heterogeneous mix of atrophy sites that variably included temporoparietal cortex, Broca's area, and dorsal premotor cortex.

That last detail deserves emphasis. Severe comprehension failure was associated with damage to Broca's area, the production region. The neat division between a front that talks and a back that understands does not hold up.

Nina Dronkers had reached a compatible conclusion from stroke data a decade earlier [18]. She and her colleagues assessed 64 chronic left hemisphere stroke patients, along with 8 right hemisphere stroke patients and 15 neurologically normal older adults, on 11 subtests of a receptive language battery. Comprehension deficits mapped onto several regions, including areas well outside the classical Wernicke territory. Comprehension is not a thing that lives in one place.

Jeffrey Binder and colleagues made the distributed picture explicit in a 2009 meta-analysis of 120 functional neuroimaging studies of semantic processing [19]. The semantic system is a large, multi-region network spanning temporal, parietal and frontal cortex. A single dot on a diagram is not a reasonable summary of it.

There is also a third hub the classical model has no room for. Mohamed Seghier reviewed the angular gyrus and its subdivisions, a region heavily involved in comprehension, particularly in reading [20]. Of the pages currently ranking for this topic, exactly one mentions it. If you want to understand how the comprehension side connects to reading, our article on the neuroscience of reading and the visual word form area picks up that thread, and the storage side is covered in semantic memory.

Nobody Agrees Where Wernicke's Area Is

Here is a problem that sits underneath all the others and rarely gets said out loud.

In 1976, Bogen and Bogen published a paper in the Annals of the New York Academy of Sciences whose title is a question: where is Wernicke's region? [21] They went through the published definitions and found that authors did not agree. Some restricted it to the posterior third of the superior temporal gyrus. Some included the supramarginal and angular gyri. Some extended it into the middle temporal gyrus. The boundaries moved depending on who was drawing them.

Nearly fifty years later this has not been resolved. Textbooks state confidently that Wernicke's area is Brodmann area 22 in the posterior superior temporal gyrus, and that is the most common convention, but conventions are not measurements.

Combine that with the finding that individual anatomy varies substantially [10], and a specific problem appears. If different researchers mean different things by the term, and if the underlying anatomy differs between people anyway, then two studies of "Wernicke's area" may not be studying the same tissue. Some of the contradictions in this literature are not disagreements about the brain. They are disagreements about a word.

This is what Pascale Tremblay and Anthony Steven Dick were driving at in the 2016 review with the deliberately blunt title [2]. Their argument is not that the regions do not exist or that damage to them does not matter. It is that the terminology has become an obstacle. The names carry a nineteenth century theory inside them, and using the names keeps the theory alive after the evidence has moved on.

The Cable Does Not Run Where You Think

The arcuate fasciculus is the third element of the classical model, and it has had a similar reassessment.

The textbook version is a single bundle running from Wernicke's area forward to Broca's area, carrying the sound patterns of words to the machinery that produces them. Cut it, and you get conduction aphasia, in which comprehension and fluent speech survive but repetition fails.

Marco Catani and Marsel Mesulam traced the history of this idea and its current standing in 2008 [22]. The short version is that the anatomy is more complicated than one cable between two boxes. Catani, working with Derek Jones and Dominic ffytche in 2004, used diffusion tractography to map the perisylvian language pathways in living brains and found a segmented structure, with an indirect pathway running through the inferior parietal lobe alongside the direct one [23]. That is a network, not a wire.

Byron Bernal and Alfredo Ardila reviewed the specific claim that arcuate damage causes conduction aphasia in 2009 and found the relationship is not as tidy as the model requires [24]. Conduction aphasia is real and it does occur. Whether it is caused by a clean disconnection of one bundle is a different question, and the answer appears to be no in many cases.

Some of the strongest evidence about these pathways comes from an unusual source. Hugues Duffau operates on brain tumours while the patient is awake, stimulating the cortex and the white matter underneath with a small electrode and watching what happens to speech in real time. His 2005 study in Brain used cortico-subcortical electrostimulation to map the semantic system in living, talking people [25]. It is about as direct a test of function as neuroscience gets, and the picture it produces is one of distributed networks with specific pathways whose disruption produces specific errors, not one of boxes and wires.

Method matters here in a way that is easy to skip past. Elizabeth Bates, Stephen Wilson and colleagues introduced voxel-based lesion-symptom mapping in 2003, which allowed lesion studies to move from grouping patients by syndrome to testing, voxel by voxel, which damaged tissue predicts which deficit [26]. A great deal of the modern picture exists because the statistics improved, not because the brains changed.

What Replaced the Classical Model

The classical model has been criticised for decades. The harder question is what to use instead, and this is where the field genuinely disagrees. What follows are three serious contenders. None of them has won.

The dual stream model, set out by Gregory Hickok and David Poeppel in 2007, borrows an idea from vision [27]. Sound arrives in auditory cortex and splits. A ventral stream, running forward through temporal cortex, maps sound onto meaning. A dorsal stream, running up and back through parietal cortex toward frontal regions, maps sound onto articulation. The ventral stream is largely bilateral, which explains why comprehension often survives damage that devastates speech production. The dorsal stream is strongly left lateralised.

Dorothee Saur and colleagues tested the anatomy directly in 2008, combining functional imaging with diffusion tractography to show that the two functional streams correspond to two identifiable fibre pathways [28]. That is a rare and satisfying result: a model proposed on functional grounds, confirmed by anatomy.

The MUC model, developed by Peter Hagoort, divides the problem differently, into Memory, Unification and Control [29]. Memory is the stored inventory of words and their properties, held mainly in temporal cortex. Unification is the process of combining them into structures that mean something, and this is what Hagoort assigns to the frontal regions including Broca's area. Control governs which language you are speaking and whose turn it is. Hagoort and Peter Indefrey laid out the broader case for looking past single words in 2014 [30]. The appeal of MUC is that it explains why a production region would be involved in comprehension, which is exactly what Mesulam's patients showed.

Angela Friederici's account, set out at length in 2011, starts from the structural connections and builds the functional story on top of them, with a strong emphasis on the time course of processing and on the specific pathways that support syntax [31].

These are not minor variations on one idea. They make different claims about what the frontal regions compute and about how much of the work is bilateral. Anyone who tells you the field has settled on a replacement is describing a preference, not a consensus.

What all of them share is the rejection of the box-and-wire picture. Language runs on distributed networks with heavy internal traffic, considerable individual variation, and no single point where meaning is stored or speech is produced.

What Aphasia Actually Looks Like

Now the clinical reality, which is where the popular account diverges from the data most sharply and where the numbers are most useful.

Almost everybody who has read anything about this can name two aphasias. Broca's, the effortful one. Wernicke's, the fluent but empty one. Ask how common they are and most people would guess they are the main two.

They are not. The Copenhagen Aphasia Study enrolled 270 consecutive acute stroke patients with aphasia across three hospitals in Copenhagen, of whom 203 had suffered a first-ever stroke, assessed everyone with the Western Aphasia Battery, and reassessed them a year later [32]. Here is the distribution in first-ever stroke.

The two famous ones together account for 28 percent of cases. Global aphasia alone is 32 percent, and anomic aphasia, which most people outside a clinic have never heard of, is 25 percent. The two types the internet teaches are not the two types people mostly get.

The pattern in that table is also more informative than the two headline names suggest. Notice that repetition and comprehension vary independently of fluency. Transcortical sensory aphasia has impaired comprehension with preserved repetition, meaning the patient can repeat a sentence they do not understand. That combination is impossible in a model where comprehension has to happen before repetition can occur, and it was one of the observations the diagram-makers used to build their scheme in the first place. Antonio Damasio's 1992 overview remains a clear statement of how these syndromes are distinguished [33].

The Copenhagen study also reported something hopeful and rarely quoted. Over the first year, aphasia type always shifted toward a less severe form. Non-fluent aphasia could become fluent, with global aphasia evolving into Wernicke's or Broca's and Broca's evolving into anomic. Fluent aphasia never evolved into non-fluent aphasia. The direction of travel is one way.

Recovery keeps going for longer than the first year, and the brain keeps changing while it does. Thomas Hope, Alex Leff and colleagues followed patients years after left hemisphere stroke and found continuing structural change in the right hemisphere tracking changing language skills [34]. Speech and language therapy for aphasia after stroke has been assessed in systematic reviews of the randomised evidence, which is the appropriate place to look for what actually helps [35].

One caution, stated plainly. This section describes group patterns from research studies. It is not a guide to assessing anyone, including yourself, and nothing here should be used to interpret a real person's symptoms. Aphasia after a suspected stroke is a medical emergency and belongs with clinicians.

Things You Will Read That Are Not True

Some of these appear on pages currently ranking on the first page of Google for this exact topic.

"Broca's area and Wernicke's area are both in the temporal lobe." Broca's area is in the inferior frontal gyrus, in the frontal lobe. Wernicke's area is in the posterior superior temporal gyrus, in the temporal lobe. Two currently ranking pages state the frontal region is temporal. It is a straightforward factual error and it propagates because pages copy each other.

"Language damage is more common in left-handers." The logic runs backwards. Szaflarski, Binder and colleagues used whole-brain fMRI to measure language lateralisation in 50 healthy non-right-handed people [36]. Activation was predominantly left hemispheric in 78 percent of them, 39 out of 50. It was symmetric in 14 percent, 7 out of 50, and predominantly right hemispheric in only 8 percent, 4 out of 50. Most left-handers are still left dominant for language. Atypical organisation is more common in non-right-handers than in right-handers, which is the real finding, but it remains the minority pattern in both groups.

"Damage to Broca's area causes Broca's aphasia." Damage confined to the region produces a deficit that largely recovers. The persistent syndrome requires much more extensive damage [9].

"Damage to Wernicke's area causes Wernicke's aphasia." In 72 primary progressive aphasia patients, atrophy in the traditional Wernicke territory left single word comprehension intact [17].

"Broca's area produces speech, Wernicke's area understands it." Severe comprehension impairment was associated with damage that included Broca's area [17], and Broca's area is engaged during comprehension of structurally demanding sentences [29].

"The arcuate fasciculus is the cable from Wernicke's area to Broca's area." The perisylvian pathways are segmented, with an indirect route through inferior parietal cortex in addition to the direct one [23], and the disconnection account of conduction aphasia is contested [24].

"Broca's area is the speech muscle region." Native signers activate it while using their hands [16].

"Broca's patient was called Tan." His name was Louis Victor Leborgne. "Tan" was the only syllable he could produce. Using it as his name is a small thing, and it is also the habit of turning a person into a symptom.

"Marc Dax and Gustave Dax are the same person." Father and son [5].

What This Means If You Are Learning a Language

This is the part the clinical literature does not cover and the exam-prep pages do not attempt, and it is where the science becomes relevant to something most readers actually do.

Start with a result from 1997 that is still the cleanest demonstration of its kind. Karl Kim, Norman Relkin, Kyoung-Min Lee and Joy Hirsch scanned bilingual people and compared where their two languages were represented [37]. In Broca's area, a second language acquired in adulthood occupied territory that was spatially separable from the native language. In people who had acquired both languages in early childhood, the two overlapped. In Wernicke's area, the representations overlapped regardless of when the second language was learned.

Two different regions, two different developmental stories. The comprehension side appears to accommodate a new language within existing tissue whenever you learn it. The production side, for a language learned late, appears to set up somewhat separately.

Age of acquisition turns out to matter for some things and not others. Isabell Wartenburger, Hauke Heekeren, Jubin Abutalebi, Stefano Cappa and colleagues separated the effects of age of acquisition from the effects of proficiency and found they load differently onto grammatical and semantic processing, with grammatical processing more sensitive to when you started [38]. Perani and colleagues had shown a few years earlier that high proficiency can substantially reduce the differences between a first and second language in the brain, even when the second was learned late [39]. If you want the fuller picture on that, we cover it in how bilingual brains store languages.

Now the mechanism that connects all of this to the ordinary experience of learning vocabulary.

Alan Baddeley, Susan Gathercole and Costanza Papagno published a paper in 1998 arguing that the phonological loop, the part of working memory that holds and rehearses sound-based information, is not a general purpose scratchpad that happens to be useful for language [40]. Its function, they argued, is language learning. The evidence came from several directions at once: patients with impaired phonological short-term memory struggle specifically to learn new words while retaining the ability to learn other kinds of material, children's phonological memory capacity predicts their vocabulary growth, and adults learning a foreign language show the same relationship.

The title of the paper is the claim. The phonological loop is a language learning device.

That reframes what happens when you try to hold a new word in your head long enough for it to stick. You are not using a general memory buffer for a language task. You are using a system whose apparent purpose is exactly this, and whose capacity predicts how quickly your vocabulary grows. The mechanism is set out in more detail in our article on the phonological loop, and the wider story of how language gets built in the first place is in how the brain learns a language.

Two cautions, because this is where articles about brain science usually overreach.

Nothing in the imaging literature tells you how to study. Knowing that a second language learned in adulthood occupies partly separate territory in the inferior frontal gyrus does not imply any particular practice schedule, any technique, or any product. The distance between a brain map and a study plan is enormous, and the honest position is that the map does not yet contain the plan.

And the regions involved are not doing only language work. Fedorenko's finding that domain-general subregions sit interleaved with language-selective ones inside Broca's area [12] means that some of what looks like language activation during a hard learning task is general effort. Uri Hasson and colleagues have argued that explanations of language processing need to be grounded in mechanisms that are not language-specific at all [41]. Susan Bookheimer's overview of what functional imaging can and cannot establish about language remains a good corrective for anyone inclined to read too much into a coloured blob [42]. The related machinery in the frontal lobe is covered in our piece on the prefrontal cortex, and the cellular basics are in how neurons communicate.

What Is Actually Settled

After all that revision, it would be easy to conclude that nothing is known. That would be the wrong conclusion, and it is worth being precise about what survives.

Language depends on the left hemisphere in the large majority of people, including most left-handers [36]. This has held since 1836 and nothing has dented it.

Damage to left perisylvian cortex impairs language. The argument in this field is about which damage produces which deficit, never about whether the region matters.

Broca's area is in the inferior frontal gyrus of the frontal lobe. Wernicke's area is in the posterior superior temporal gyrus of the temporal lobe. The boundaries are debated. The lobes are not.

The broad clinical distinction between fluent and non-fluent aphasia is real, useful and used daily by clinicians who know perfectly well that the underlying model is dated.

Aphasia after left hemisphere stroke is common, and its type reliably shifts toward less severe forms over the first year [32].

Broca and Wernicke were right about the thing that mattered most. Specific mental functions can be tied to specific brain tissue, and you can find out how by comparing behaviour in life with damage after death. Every piece of evidence in this article that contradicts them was produced using the method they invented.

Conclusion

Go back to the jar.

Leborgne's brain sat in preservation fluid for 146 years before anyone could see inside it. When the scanner finally looked, it found damage that went deeper and further than the man who examined it in 1861 could have known. Not because Broca was careless. Because he was looking at a surface and drawing a conclusion about a volume.

That is a decent summary of the whole field's problem for the next century and a half. Every generation looked at the best evidence available, drew the tidiest map that evidence would support, and taught the map with more confidence than the evidence deserved. Then the tools improved.

The current position is more interesting than the version in the textbook, and it is not really a story of two men being wrong. Broca's area turned out to contain at least two functionally different systems sitting millimetres apart, invisible to any method that averages across brains [12]. Wernicke's area turned out to be a term different researchers use for different tissue [21], and damage to it does not do what its name predicts [17]. The cable between them turned out to be a segmented network [23]. The syndrome named after Broca requires damage well outside Broca's area, and has done since 1978 [9].

What replaced the model is genuinely unsettled. Dual stream, MUC and the structural accounts each explain some of the data well and none of them explains all of it, and choosing between them is a live research question rather than a matter of preference.

The thing worth holding onto is smaller than any of the models and older than all of them. In 1861 a surgeon looked at a man who could say one syllable, and then looked inside his head, and concluded that the two facts were connected. Everything since, including all the evidence that he oversimplified, comes from taking that idea seriously.

The names on the map are wrong in most of the ways names can be wrong. The map is still where everyone starts.