Language: In April 1861, a 51-year-old man named Louis Victor Leborgne died in a Paris hospital after a 21-year illness. He had been admitted in 1840 as a young man able to understand everything said to him but capable of producing only a single syllable: 'tan.' Over the years his paralysis had spread, but his comprehension never failed.
The surgeon Paul Broca performed the autopsy and found a lesion in the left inferior frontal lobe.[1] He presented the case to the Societe d'Anthropologie de Paris, proposing that this small region was essential for the articulate production of speech.
The case of 'Tan' launched the scientific study of language in the brain, a field that has generated more than 160 years of increasingly sophisticated investigation.
What Broca could not know was that the region he had identified was one node in a network that would take a century and a half of increasingly sophisticated technology to map.
What his successors in the 19th and 20th centuries would add - Wernicke's area, the arcuate fasciculus, the Wernicke-Geschwind model - seemed to settle the question of where language lived.
Then functional MRI came along and showed that the answer was far more distributed, far more bilateral, and far more dynamic than any of them had imagined.
Understanding how the brain processes language matters beyond academic interest. Aphasia - the loss of language ability from brain damage - affects approximately one million Americans, most following stroke. Reading disorders affect 5-10% of the population.
Second language education, speech therapy, and neurological rehabilitation all depend on an accurate picture of language's neural architecture.
And the deep question of how meaning arises from the electrical activity of neurons - how 'socks' in the wrong context surprises a brain, how a child acquires grammar from babble, how a bilingual speaker toggles between languages without apparent effort - touches something close to the heart of what makes minds minds.
"Language is not a cultural artifact that we learn the way we learn to tell time or how the federal government works. It is a distinct piece of the biological makeup of our brains." - Steven Pinker, The Language Instinct (1994)[11]
Key Definitions
Broca's area: Region in the left inferior frontal gyrus (Brodmann's areas 44 and 45) associated with speech production, grammatical processing, and articulatory planning; named for Paul Broca, who linked the region to speech production in 1861.
Wernicke's area: Region in the left posterior superior temporal gyrus (Brodmann's area 22) associated with speech comprehension and lexical processing; named for Carl Wernicke, who identified it in 1874.
Aphasia: Partial or complete loss of the ability to produce or comprehend language, resulting from brain damage, most commonly stroke or traumatic brain injury.
fMRI (functional magnetic resonance imaging): A neuroimaging technique that measures brain activity by detecting changes associated with blood flow, enabling researchers to observe which brain regions are active during language tasks in living participants.
Dual-stream model: A framework proposing two processing pathways for language: a dorsal stream linking acoustic areas to motor regions for sound-to-articulation mapping, and a ventral stream linking acoustic areas to temporal regions for sound-to-meaning mapping.
The Aphasia Spectrum: Lesion Location and Clinical Presentation
| Aphasia Type | Lesion Location | Fluency | Comprehension | Repetition | Characteristic Feature |
|---|---|---|---|---|---|
| Broca's | Left inferior frontal gyrus | Non-fluent (effortful, halting) | Relatively preserved | Impaired | Telegraphic speech; omits grammar words |
| Wernicke's | Left posterior superior temporal gyrus | Fluent | Severely impaired | Impaired | "Word salad"; patient unaware of errors |
| Conduction | Arcuate fasciculus | Fluent | Relatively preserved | Severely impaired | Can understand and produce but cannot repeat |
| Global | Large left perisylvian lesion | Non-fluent | Severely impaired | Severely impaired | Most severe; all language modalities affected |
| Anomic | Variable (often angular gyrus / temporal) | Fluent | Preserved | Preserved | Word-finding difficulty; circumlocutions |
| Transcortical motor | Anterior frontal (outside Broca's) | Non-fluent | Preserved | Preserved | Can repeat but not initiate spontaneous speech |
| Transcortical sensory | Posterior (outside Wernicke's) | Fluent | Impaired | Preserved | Opposite of transcortical motor |
"The neurological record of aphasia has given us more insight into the organization of language in the brain than decades of normal psychology." - Andrew Kertesz, Aphasia and Associated Disorders (1979)
The Classic Neuroanatomy: Broca, Wernicke, and the Two-Area Model
Paul Broca and the Discovery of Speech Production
Paul Broca's 1861 case of Leborgne - 'Tan' - was followed quickly by additional cases with similar lesion locations, and by 1865 Broca had formalized the claim in two papers: speech production depends on the left inferior frontal region, and the left hemisphere is dominant for language.
The second claim - left hemisphere dominance - was one of the earliest documented examples of cerebral lateralization of function.
Broca's aphasia, as it came to be known, presents with non-fluent speech. Patients produce utterances slowly and effortfully, often omitting grammatical function words, verb inflections, and sentence structure while preserving content words.
A patient trying to describe a cookie-theft picture might say 'boy... cookie... fall... woman... water' rather than 'the boy is stealing a cookie while the woman washing dishes lets the sink overflow.' Comprehension of simple sentences is relatively preserved. Repetition is impaired.
Carl Wernicke and the Discovery of Comprehension Areas
Carl Wernicke was 26 years old when he published his monograph 'The Aphasic Symptom Complex' in 1874, describing a new aphasia type with the opposite profile from Broca's.
His patients produced fluent, grammatically intact, rhythmically normal speech that was semantically empty - filled with substituted words (paraphasias), invented words (neologisms), and what listeners described as 'word salad.' Comprehension was severely impaired; patients often seemed unaware their speech was meaningless.
Wernicke traced the lesions to the left posterior superior temporal region.[2]
Wernicke made a further contribution that went beyond describing the new syndrome: he proposed a model in which spoken words were received and recognized in what became known as Wernicke's area, then transmitted via a connecting pathway to Broca's area for speech production.
Damage to the connecting pathway would produce a third syndrome: fluent speech, intact comprehension, but severely impaired ability to repeat heard words. This was precisely the syndrome that clinicians subsequently documented and termed conduction aphasia.
The Wernicke-Geschwind Model
Norman Geschwind at Harvard Medical School formalized the two-area model in a landmark 1965 paper in Brain, connecting it to white matter pathways and extending it to explain a broader range of language disorders.[3]
In the Wernicke-Geschwind model, the arcuate fasciculus - a white matter tract curving from posterior temporal regions through the parietal lobe to frontal regions - connects Wernicke's area to Broca's area.
The model predicted and organized a classification of aphasias by lesion location that has remained the backbone of clinical aphasia diagnosis.
The model's predictive success is genuine. The three-way distinction among Broca's aphasia (frontal), Wernicke's aphasia (posterior temporal), and conduction aphasia (arcuate fasciculus) is well-supported. The localization of production to frontal and comprehension to posterior temporal is directionally correct.
However, the model significantly understates the complexity of the language network. It implies that language lives primarily in two regions connected by one pathway.
Decades of neuroimaging have shown this is wrong: language engages distributed bilateral networks, individual variation in organization is substantial, and the functions attributed to 'Broca's area' and 'Wernicke's area' are far more complex and varied than the simple production/comprehension division suggests.
The Modern Neuroscience: Distributed Networks and Dual Streams
What fMRI Revealed
The first fMRI language studies in the mid-1990s immediately demonstrated that language engaged far more brain territory than the Wernicke-Geschwind model suggested.
The inferior temporal lobe, angular gyrus, middle frontal gyrus, supplementary motor area, premotor cortex, and bilateral temporal regions all showed consistent activation across various language tasks.
The model had been built from lesion data - areas whose damage impairs language - but functional imaging shows areas involved in language processing, a broader and different set.
Angela Friederici at the Max Planck Institute for Human Cognitive and Brain Sciences, working across three decades, has produced the most systematic account of functional differentiation within the language network.[5]
Her research, synthesized in 'Language in Our Brain: The Origins of a Unique Human Capacity' (2017), establishes that different components of the language network contribute distinct operations.
Broca's area, particularly BA44, contributes to hierarchical syntactic structure building - the computation of who did-what-to-whom relationships in complex sentences.
BA45, immediately adjacent, is more involved in semantic composition - combining word meanings into phrase and sentence meanings. Left anterior temporal regions handle early combinatorial processing. Left posterior temporal regions, including Wernicke's area, handle fine-grained phonological and lexical processing.
Prosody - the rhythm, stress, and intonation of speech that conveys emotion and discourse structure - engages bilateral networks, with right hemisphere regions playing particularly important roles in emotional prosody processing.
A patient with a right hemisphere stroke may speak with flat, affectively empty intonation even with fully intact left-hemisphere language systems.
The Dual-Stream Model
Gregory Hickok at UC Irvine and David Poeppel at New York University proposed the dual-stream model of auditory language processing in a 2000 paper and expanded it in a highly cited 2007 review in Nature Reviews Neuroscience.[4]
The model draws on the parallel dorsal-ventral stream architecture that had been established for visual processing.
The dorsal stream processes sound-to-motor mappings. Running from posterior superior temporal cortex through inferior parietal cortex (Spt, the sylvian-parietal-temporal region) to premotor cortex, this stream converts acoustic input to articulatory motor programs.
It is critical for speech repetition, for learning to produce new sound sequences, and ultimately for speech production planning. Damage in this stream disrupts repetition, consistent with conduction aphasia and related disorders.
The ventral stream processes sound-to-meaning mappings. Running from posterior superior temporal cortex through middle and inferior temporal cortex toward anterior temporal regions, this stream maps acoustic input to lexical-semantic representations - word meanings, conceptual knowledge, and the combinatorial processes that build phrase and sentence meaning.
Damage in this stream disrupts comprehension.
The dual-stream model explains why lesions in different locations can produce superficially similar symptoms and why the same lesion can produce different deficits depending on which processing demands are placed on the patient.
It also provides a framework for understanding the complementarity of production and comprehension: both streams originate in shared auditory cortex regions, diverging to serve distinct computational needs.
Language Acquisition: Critical Periods and the Poverty of the Stimulus
Chomsky's Universal Grammar
The speed and accuracy of child language acquisition has been among the most discussed phenomena in cognitive science. By age two, most children produce multi-word utterances.
By age four or five, children command the recursive grammatical structures of their language - forming relative clauses, passives, complex questions - despite receiving input that is imperfect, context-dependent, and far from the complete explicit grammar that linguists require years to describe.
Noam Chomsky at MIT argued in work beginning in the late 1950s and most accessibly in 'Language and Mind' (1968) that this acquisition feat is inexplicable unless children begin with substantial innate knowledge of the abstract principles underlying all human languages - universal grammar.[10]
The 'poverty of the stimulus' argument holds that the grammatical rules children infer are not uniquely determined by the input they receive: children avoid certain errors they have never received explicit correction for, and they acquire grammatical distinctions that the input would not by itself reveal.
Chomsky's proposal is that human language capacity is a biological endowment, a specialized cognitive module shaped by evolution, rather than a general learning process applied to linguistic input.
The Critical Period
Eric Lenneberg's 'Biological Foundations of Language' (1967) formalized the critical period hypothesis: first language acquisition is possible only within a window bounded by early childhood and puberty, after which neuroplasticity decreases and full acquisition becomes difficult or impossible.[8]
Evidence for a critical period comes from multiple sources.
Deaf individuals who receive cochlear implants before age two acquire spoken language at near-normal rates; those implanted after age five show substantially impaired spoken language development even with years of subsequent training.
The developmental trajectory of phonological, syntactic, and lexical development all show similar critical period dynamics, though with different timing.
The case of Genie Wiley, discovered in 1970 in Los Angeles after being kept in isolation and severe neglect from 20 months of age until 13 years, became the most extensively studied case of extreme language deprivation.
Following her discovery and sustained language instruction, Genie acquired vocabulary but never acquired productive grammar, providing what many researchers interpreted as evidence that the critical period for grammatical acquisition had closed.
Her case is complicated by the severe psychological trauma she suffered, which confounds the linguistic evidence, but it has remained an influential point of reference.
Patricia Kuhl and Native Language Neural Commitment
Patricia Kuhl at the University of Washington documented a narrowing of phonological sensitivity in early infancy that she termed 'native language neural commitment.' Using high-amplitude sucking and other infant habituation procedures, Kuhl and colleagues showed that infants at 6 months of age can distinguish phonemic contrasts from any language, including contrasts absent in their home language.
By 10-12 months, this universal sensitivity has narrowed to language-specific sensitivity: English-learning infants can no longer reliably distinguish the Japanese /r/ and /l/ distinction that 6-month-old English-learning infants could still detect.
Kuhl's interpretation is that early statistical learning from the linguistic environment leads the brain to reorganize its phonological processing around the structures of the heard language.[9]
This reorganization is efficient - it sharpens processing of relevant distinctions - but involves loss of sensitivity to distinctions that are irrelevant for the native language.
The neural commitment hypothesis accounts for the difficulty of achieving native-like accent in a second language acquired after the phonological critical period.
Reading in the Brain: Cortical Recycling and Dyslexia
The Brain Was Not Built for Reading
Reading is a cultural invention approximately 5,000 years old - far too recent to have driven genetic evolution of dedicated neural circuits. Writing systems did not exist when the human brain's current architecture evolved.
Yet literate adults process written words with extraordinary speed and automaticity, recognizing common words in under 150 milliseconds.
Stanislas Dehaene at the College de France has developed the most comprehensive neuroscientific account of how the brain accomplishes this, summarized in 'Reading in the Brain' (2009) and subsequent papers.
His central framework is cortical recycling: cultural learning co-opts neural circuits evolved for other purposes and repurposes them through intensive training.
Reading recruits circuits in the ventral visual stream that evolved for object recognition, reprogramming them to recognize the specific visual shapes of written language.
The visual word form area (VWFA), located in the left fusiform gyrus in the left occipito-temporal region, is the most striking evidence for this recycling.
In literate adults across all tested writing systems - alphabetic, syllabic, logographic - this region responds selectively to words and letterstrings over non-letter stimuli, in a position and with a selectivity that is remarkably consistent across individuals.
Dehaene and colleagues showed in cross-cultural imaging studies that the same anatomical region, showing the same functional selectivity, is recruited by reading Arabic, Hebrew, Chinese, and Roman script.[7]
The literacy-dependent reorganization of this region has been confirmed in longitudinal studies of children learning to read.
The Dual-Route Model of Reading
Max Coltheart and colleagues developed the dual-route model of reading, distinguishing two pathways by which written words are converted to meaning and sound.[15]
The lexical route maps familiar written words directly to their stored phonological and semantic representations, enabling skilled readers to recognize high-frequency words as whole visual units without phonological assembly.
The phonological route assembles pronunciation from grapheme-phoneme correspondence rules, enabling readers to sound out novel words or pronouns they have never encountered in written form.
Both routes contribute to skilled adult reading. The phonological route is essential during the acquisition of reading: learning to read requires mastering the correspondence rules between letters and sounds before the lexicon is large enough for the lexical route to carry most of the load.
Developmental dyslexia, affecting approximately 5-10% of the population, is now understood as primarily a deficit in phonological processing - the ability to represent and manipulate the sound structure of language independently of meaning.
Children with dyslexia show difficulty with phoneme awareness tasks (identifying that 'cat' and 'bat' rhyme; counting the phonemes in 'split') that predict and mediate difficulty learning grapheme-phoneme correspondences.
fMRI studies by Guinevere Eden, Sally Shaywitz, and others have consistently found reduced activation in left temporoparietal and occipitotemporal reading networks in individuals with dyslexia. Effective interventions target phonological awareness directly and produce both behavioral and neural changes.
The Bilingual Brain
Two languages in one brain raise fundamental questions about neural organization. Do the two languages share the same networks? How does the brain avoid constant interference from the non-target language? Does managing two languages affect cognition more broadly?
Early versus late bilinguals organize their languages differently. Individuals who acquire a second language in early childhood tend to represent both languages in substantially overlapping cortical substrates, particularly in Broca's area.
Late bilinguals show greater spatial separation between first and second language representations in frontal regions, while posterior regions for phonological and semantic processing show more overlap.
These differences correspond to behavioral differences: early bilinguals typically achieve native-like accent in both languages, while late bilinguals almost universally retain a foreign accent in the later-acquired language, consistent with phonological critical period effects.
Jubin Abutalebi at the University of Milan-Bicocca has conducted extensive imaging research on bilingual language control.[12] His findings implicate the anterior cingulate cortex and left caudate nucleus in the inhibitory processes by which bilinguals suppress the non-target language during use of the other.
Code-switching - alternating between languages within or across utterances - activates these inhibitory control regions, consistent with the view that managing two languages requires continuous active competition management rather than simple switching.
The bilingual cognitive advantage hypothesis, associated with Ellen Bialystok at York University, proposed that lifelong experience managing two competing language systems produces broader benefits to executive function and attentional control.
Bialystok and colleagues also reported that bilingualism delayed the diagnosis of Alzheimer's disease symptoms by 4-5 years. These claims attracted substantial attention but have proven difficult to replicate.
A 2014 meta-analysis by Paap, Johnson, and Sawi found no reliable bilingual advantage on executive function tasks in studies with adequate statistical power.[14] The dementia delay claim has faced similar replication failures in large population samples.
The current scientific consensus is that strong versions of the bilingual advantage claim are not well-supported by the available evidence.
Predictive Processing: The Brain as Language Anticipation Machine
The predictive processing framework, developed most comprehensively by Karl Friston at University College London, proposes that the brain continuously generates predictions about forthcoming input and updates its model only upon prediction error - when the prediction fails.
Rather than passively processing sensory input, the brain actively anticipates what is coming and devotes computational resources primarily to unexpected events.
Applied to language, predictive processing provides an account of the remarkable speed of comprehension. Skilled readers and listeners do not wait for each word to arrive before processing it: they use syntactic structure, semantic context, world knowledge, and discourse expectations to predict what is coming next.
When predictions are accurate, comprehension is fast and effortless. When predictions are violated, a prediction error signal is generated, and the model is updated.
The N400: Direct Neural Evidence for Prediction
Marta Kutas and Steven Hillyard at UC San Diego reported in a landmark 1980 paper in Science the discovery of the N400 - a negative electrical potential peaking approximately 400 milliseconds after an unexpected word is encountered.[6]
Using event-related potentials recorded from electrodes on the scalp, they showed that sentence-final words that violated semantic expectations produced a large negative deflection, while semantically expected words produced a small one.
The original sentence pair was: 'He spread the warm bread with butter' (small N400) versus 'He spread the warm bread with socks' (large N400).
Five decades of subsequent N400 research have mapped the parameters that modulate prediction strength: semantic coherence, syntactic plausibility, discourse context, speaker-specific expectations, and world knowledge all contribute.
The N400 is now among the most extensively studied cognitive neuroscience components and has been documented across dozens of languages, modalities, and experimental paradigms.
Embodied Simulation in Language Comprehension
Rolf Zwaan at Erasmus University Rotterdam developed the embodied simulation hypothesis: language comprehension involves simulating the perceptual, motor, and affective experiences that the language describes, rather than computing abstract propositional representations.[13]
Comprehending 'the carpenter hammered the nail into the floor' activates circuits for downward arm movements; comprehending 'the carpenter hammered the nail into the wall' activates circuits for outward arm movements.
Reading 'the eagle in the sky' activates larger visual field representations than 'the eagle in the nest.'
These findings, published in a series of studies from the early 2000s through the present, suggest that semantic knowledge is not stored as amodal symbols but is grounded in sensorimotor experience. Language understanding, on this view, is fundamentally about mental simulation - running an internal model of the described state of affairs.
The implications for theories of meaning, concepts, and the relationship between language and thought continue to be actively investigated.
Sources & Further Reading
- Broca, P. (1861). Remarques sur le siege de la faculte du langage articule. Bulletins de la Societe d'Anthropologie, 2, 330-357.
- Wernicke, C. (1874). Der aphasische Symptomencomplex. Cohn and Weigert.
- Geschwind, N. (1965). Disconnexion syndromes in animals and man. Brain, 88, 237-294.
- Hickok, G. and Poeppel, D. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience, 8(5), 393-402.
- Friederici, A.D. (2017). Language in Our Brain: The Origins of a Unique Human Capacity. MIT Press.
- Kutas, M. and Hillyard, S.A. (1980). Reading senseless sentences: Brain potentials reflect semantic incongruity. Science, 207(4427), 203-205.
- Dehaene, S. (2009). Reading in the Brain: The New Science of How We Read. Viking.
- Lenneberg, E.H. (1967). Biological Foundations of Language. Wiley.
- Kuhl, P.K. (2004). Early language acquisition: Cracking the speech code. Nature Reviews Neuroscience, 5(11), 831-843.
- Chomsky, N. (1968). Language and Mind. Harcourt, Brace and World.
- Pinker, S. (1994). The Language Instinct. William Morrow.
- Abutalebi, J. and Green, D. (2007). Bilingual language production: The neurocognition of language representation and control. Journal of Neurolinguistics, 20(3), 242-275.
- Zwaan, R.A. (2004). The immersed experiencer: Toward an embodied theory of language comprehension. Psychology of Learning and Motivation, 44, 35-62.
- Paap, K.R., Johnson, H.A. and Sawi, O. (2014). Are bilingual advantages dependent upon specific tasks or specific bilingual experiences? Journal of Cognitive Psychology, 26(6), 615-639.
- Coltheart, M. et al. (2001). DRC: A dual route cascaded model of visual word recognition and reading aloud. Psychological Review, 108(1), 204-256.