The Science of Learning & Speaking Languages | Dr. Eddie Chang

概览

This episode features Andrew Huberman in conversation with Dr. Eddie Chang, chair of neurosurgery at UCSF, about how the brain hears, produces, and understands speech and language. The discussion connects early auditory development, neuroplasticity, awake brain surgery, epilepsy, bilingualism, dyslexia, stuttering, and brain-machine interfaces.

A central theme is that speech is not a simple function of one textbook “speech area.” Chang explains that speech depends on distributed, highly precise circuits for hearing, vocal tract control, motor planning, auditory feedback, and language meaning. He also argues that some classical teaching about Broca’s area is oversimplified or wrong.

The second half turns toward clinical translation: Chang describes work decoding speech-related brain activity in locked-in patients, especially the Bravo trial participant Pancho. The conversation closes with the future of speech neuroprosthetics, avatars, neural augmentation, and the mental practices Chang uses to stay focused as a neurosurgeon.

分段落总结

[00:00] Guest Introduction and Episode Frame

[事实] Huberman introduces Dr. Eddie Chang as chair of neurosurgery at UCSF and an expert in speech, language, epilepsy, movement disorders, and bioengineering. [事实] The episode is framed around speech and language circuits, critical periods, bilingualism, stuttering, epilepsy, ketogenic diet, Neuralink, and brain-machine interfaces. [事实] Huberman notes that he and Chang have known each other since childhood and later reconnected at UCSF.

[05:00] Auditory Plasticity and Critical Periods

[事实] Chang describes studying baby rodents with Mike Merzenich to understand how sound patterns shape auditory cortex organization. [事实] Raising rat pups in continuous white noise masked natural environmental sounds and kept the auditory critical period open much longer than usual. [事实] Chang says this delayed maturation of the auditory cortex, showing that sensory experience helps regulate when plasticity opens and closes. [事实] He states that early sound exposure, including sounds heard in utero, influences how neural networks organize and how people later hear sounds.

[10:12] Infant White Noise and Structured Sound

[事实] Chang says there is not yet a definitive answer about whether infant white-noise machines during sleep are harmful. [事实] He emphasizes that humans are not naturally exposed to continuous white noise and that structured, salient sounds are valuable for normal auditory development. [事实] Chang says he did not use a white-noise machine for his own three children and prefers more natural soothing sounds. [推测] The discussion suggests caution with continuous unstructured sound for infants, while acknowledging that nighttime-only exposure has not been directly resolved in humans.

[16:36] Awake Brain Surgery and Brain Mapping

[事实] Chang explains that awake brain surgery allows clinicians to talk with patients while mapping areas important for speech, language, movement, seizures, or tumor removal. [事实] A sterile drape separates the surgical field from the patient’s face, while neuropsychologists interact with the patient during testing. [事实] Brain mapping uses small electrical stimulation to probe specific sites near pathological tissue. [事实] Stimulation can move a hand, stop speech, slur words, or interfere with naming objects, helping surgeons identify areas to protect.

[23:21] Emotion, Stimulation, and Anxiety-Like Seizures

[事实] Chang says stimulation of orbital frontal cortex can reduce stress in anxious patients, while stimulation of the amygdala or insula can evoke anxiety or disgust. [事实] He describes a young woman diagnosed with anxiety disorder whose spontaneous panic-like episodes were ultimately shown to be seizures. [事实] Her MRI did not reveal the issue; implanted electrodes localized the events to the amygdala. [推测] The case illustrates why some sudden, untriggered emotional episodes can require neurological evaluation, even though Chang stresses most anxiety is not caused by seizures.

[27:20] Epilepsy Treatment, Surgery, and Seizure Types

[事实] Chang defines epilepsy in the conversation as having three or more seizures and says many patients’ seizures are controlled by medication. [事实] About one third of people with epilepsy do not achieve control with modern medications, and after two or three medications, later medication attempts are less likely to help. [事实] Some patients benefit from surgery that removes a seizure focus, while others may receive stimulators that modulate brain activity. [事实] Absence seizures can involve temporary loss of awareness without full-body convulsions. [事实] Temporal lobe seizures can involve metallic taste, burning-toast smells, deja vu, or anxiety; nocturnal seizures can relate to sleep stage and circadian timing.

[30:23] Ketogenic Diet and Epilepsy

[事实] Chang says he does not know whether the ketogenic diet was originally designed for seizures, but it can be life-changing for some people with epilepsy. [事实] He says the mechanism for why ketosis can reduce seizures remains to be determined. [事实] The discussion notes that ketogenic approaches have been especially associated with pediatric epilepsy, while adults may also try them. [推测] The episode treats ketosis as a potential clinical tool for some epilepsy cases, not as a universal replacement for medical treatment.

[37:08] Broca, Wernicke, and Textbook Revision

[事实] Chang reviews the classical account: Broca’s patient “Tan” supported the idea of a left frontal “seat of articulation,” while Wernicke’s observations linked posterior temporal lobe damage with impaired comprehension and fluent jargon. [事实] Chang says his clinical experience often did not match what he learned in medical school. [事实] He argues that Broca’s area as the basis of speaking is fundamentally wrong. [事实] He identifies the precentral gyrus, closely associated with motor cortex and vocal tract control, as crucial for speech expression. [事实] He says Wernicke’s area remains highly important for comprehension, word retrieval, and aspects of speech production.

[49:14] Language Lateralization and Handedness

[事实] Chang says that in right-handed people, language is on the left side of the brain about 99% of the time. [事实] In left-handed people, language is still more often left-sided, around 70%, but is more likely to be bilateral or right-sided. [事实] Handedness is described as strongly, though not entirely, genetic. [事实] Chang says the machinery for language may exist on both sides, but the brain usually specializes one side for functional use. [事实] After left-hemisphere strokes, recovery can involve plasticity in nearby left-sided areas or recruitment of the right hemisphere.

[54:51] Bilingualism and Language Learning

[事实] Chang says bilingual people use shared brain circuitry for different languages, but the activity patterns are not identical. [事实] His lab studies how bilingual brains respond when hearing one language versus another. [事实] He says the brain may process unfamiliar language through the listener’s native-language system. [事实] Early, intense, immersive, and sustained exposure helps language learning, and real social interaction appears to add something beyond recordings. [事实] Chang agrees that learning new languages is possible across the lifespan.

[59:21] Speech, Language, and Auditory Coding

[事实] Chang defines speech as the communication signal generated by vocal tract movement and heard as sound. [事实] He defines language more broadly, including pragmatics, semantics, syntax, sign language, and reading. [事实] The ear decomposes sound into low, middle, and high frequencies with very high temporal precision. [事实] The cortex transforms auditory signals into words and language, and Chang’s lab records millimeter-scale, millisecond-resolution brain activity from patients with implanted electrodes.

[67:09] Articulation and the Vocal Tract

[事实] Some recording sites are tuned to consonants, vowels, or features of consonants such as plosives and fricatives. [事实] Plosives require temporary closure and release in the mouth, while fricatives involve narrow airflow and turbulence. [事实] Speech begins with exhalation; the larynx brings the vocal folds together to create voicing. [事实] Chang says male voices average around 100 Hz and female voices around 200 Hz because of differences in larynx size and shape. [事实] The pharynx, oral cavity, tongue, jaw, and lips shape voiced sound into consonants and vowels.

[71:57] Vocalizations Beyond Speech

[事实] Chang distinguishes speech from vocalizations such as crying, moaning, and laughter. [事实] Vocalizations also use exhalation and phonation, but rely on different brain areas than speech and language. [事实] People with injuries to speech and language areas may still be able to moan or vocalize. [推测] This distinction supports the broader theme that communication is built from multiple partially separable neural systems.

[74:52] Speech Maps, Phonemes, and Language Complexity

[事实] Primary auditory cortex contains organized maps of sound frequency, moving from low to medium to high frequencies. [事实] Chang says speech-related cortex has a map, but it appears more “salt and pepper” than universally ordered across people. [事实] English has about 40 phonemes, while Chang says these can be reduced to roughly 12 articulatory features. [事实] Languages differ in their sound inventories and consonant clusters; Hawaiian has about 12 to 14 phonemes, while English has about 40. [事实] Russian, English, and German are discussed as languages with substantial complexity.

[81:50] Reading, Writing, and Dyslexia

[事实] Chang says reading and writing are human inventions that reuse existing brain architecture rather than relying on newly evolved brain areas. [事实] The visual word form area is sensitive to written or typed words. [事实] Reading maps visual word forms onto auditory speech and phonological systems, especially during learning. [事实] Skilled readers may later access meaning more directly without consciously sounding out every word. [事实] Chang says dyslexia can involve difficulty mapping visual words to speech sounds and may reflect problems in phonological awareness, not only vision.

[90:31] Language Change and Foreign Accent Syndrome

[事实] Chang says speech and language change over time, and dialect shifts are normal rather than deviations from one “proper” form. [事实] He has not seen or read a convincing case where a stroke causes someone to suddenly speak a language they never knew. [事实] He describes foreign accent syndrome, where injury to speech-control areas can make someone sound as if they have another accent without acquiring that language’s meaning or grammar. [推测] The discussion leaves open whether modern reading, texting, or caption habits directly change speech patterns.

[96:19] Distributed Auditory and Motor Memory

[事实] Chang says memory is distributed rather than stored in one specific speech or auditory area. [事实] People with injuries to speech areas may still sing “Happy Birthday” or count because some sequences are embedded in motor memory. [事实] Chang says he rarely sees surgeries erase long-term memories or motor skills unless injury is very severe and widespread. [推测] The brain’s distributed memory organization appears protective against losing a whole category of personal memories from a small focal injury.

[101:39] Locked-In Syndrome and the Bravo Trial

[事实] Chang describes locked-in syndrome as intact cognition and awareness with no voluntary movement or ability to speak. [事实] Causes discussed include brainstem stroke and ALS. [事实] Chang’s lab began the Bravo clinical trial to test whether speech-related brain signals could be decoded in paralyzed people. [事实] The first Bravo participant, Pancho, had been paralyzed for about 15 years after a brainstem stroke following a car accident. [事实] Pancho communicated by moving his neck to control a stick attached to a baseball cap and typing letters one by one.

[111:09] Speech Neuroprosthetic Decoding

[事实] Chang’s team implanted electrodes over Pancho’s speech-related cortex, including areas for the larynx, lips, tongue, and jaw. [事实] The electrodes connected to a port attached to the skull, and a wired system sent brain signals to a computer. [事实] Chang says the port carries infection risk and that future systems will need to become wireless. [事实] Machine-learning algorithms translated subtle brain activity patterns into text, but training took weeks. [事实] Pancho’s giggling interfered with decoding, and Chang says that issue had not been fixed.

[115:02] First Words and Language Modeling

[事实] One early sentence Pancho produced through the system involved asking to bring his family in. [事实] The initial system used a vocabulary of 50 words. [事实] The team generated possible sentences from those words and used a computational language model similar to autocorrect. [事实] Chang says the vocabulary was expanding and expects much larger vocabularies over time. [推测] The result is presented as proof of principle rather than an approved therapy.

[118:10] Neuralink, BCI, and Augmentation

[事实] Huberman raises Neuralink and the broader possibility of using brain-machine interfaces for clinical treatment and future augmentation. [事实] Chang says brain-machine interface research has existed for decades, including work restoring movement or controlling cursors. [事实] He says industry involvement is moving the field toward commercial medical products. [事实] Chang describes augmentation as enhancement beyond normal function, such as faster communication or stronger memory. [事实] He says society has not fully addressed ethical questions about invasive neurotechnology, access, and whether such enhancement is desirable.

[125:01] Limits of Neurotechnology

[事实] Chang says natural communication systems are supported by neural structures with bandwidth involving millions of neurons. [事实] He says current commercial and research technologies do not come close to that evolved biological bandwidth. [事实] He argues that cognitive augmentation is not entirely new because humans already use tools such as phones, caffeine, nicotine, medications, and other technologies. [推测] Chang expects enhancement to arrive incrementally and subtly rather than immediately as dramatic superhuman memory or communication.

[129:43] Avatars and Richer Communication

[事实] Chang says communication includes facial expressions, mouth movements, and nonverbal signals, not just text. [事实] Seeing a speaker’s mouth and jaw movements can improve intelligibility and make speech feel more natural. [事实] His team is interested in avatars that reflect speech movements and facial expressions for paralyzed patients. [事实] Chang says avatars may help users learn speech neuroprosthetic control by making the system feel embodied. [推测] The clinical avatar work may also become relevant as more social interaction moves into digital and virtual spaces.

[133:20] Stuttering, Anxiety, and Auditory Feedback

[事实] Chang defines stuttering as a speech problem in which ideas, meaning, and grammar are intact but words cannot come out fluently. [事实] He says stuttering reflects breakdowns in coordinated vocal tract control, not simply a language deficit. [事实] Anxiety can provoke or worsen stuttering, but Chang says anxiety is not the root cause. [事实] Therapy often focuses on creating conditions that help words initiate and come out. [事实] Auditory feedback can change stuttering for better or worse, suggesting interaction between speech motor commands and hearing systems.

[139:20] Exercise, Focus, and Neurosurgical State

[事实] Chang says he runs and swims primarily for mental rather than physical reasons. [事实] He notices changes in how he feels in the operating room and in interpersonal interactions if he goes a day or two without exercise. [事实] He treats the operating room as a focused, disconnected space and does not bring his phone in. [事实] He prefers no music or background distraction during surgery so speech in the room is clear. [事实] Chang describes surgery as a sacred moment because a patient’s life is directly under his hands.

播客点评/总结

[推测] The strongest value of this episode is that it connects basic neuroscience with clinical realities. Chang repeatedly grounds abstract topics like cortical maps, plasticity, and phonology in awake surgery, epilepsy cases, and speech prosthetic trials.

[推测] The highlight is the Pancho/Bravo trial discussion, because it shows how decoding speech features from the brain can move from laboratory recordings to meaningful communication for a locked-in person. The episode also does well at correcting oversimplified textbook ideas without dismissing the historical foundations.

[推测] The main limitation is that some practical questions, especially infant white-noise use, ketogenic diet mechanisms, dyslexia intervention specifics, and neurotechnology ethics, are left unresolved because the transcript itself presents them as open or still developing.

[推测] This episode is best suited for listeners interested in neuroscience, neurosurgery, speech and language, epilepsy, clinical brain mapping, neuroprosthetics, and the future of human-computer communication.