Home Andrew Huberman Notes
Andrew Huberman · 2022-10-24 · 2h 34m

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

UCSF neurosurgeon Eddie Chang explains how the brain builds speech and language, and how brain implants let paralyzed patients talk again.

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

Dr. Eddie Chang — Chair of neurosurgery at UC San Francisco and a world expert on the neuroscience of speech, language, epilepsy, and movement disorders. His lab pioneered brain-machine interfaces that let fully locked-in patients communicate.

The gist

Andrew Huberman talks with his childhood friend Dr. Eddie Chang about how the brain produces and understands speech and language. They cover critical periods for learning languages, the difference between speech and language, why the old Broca's/Wernicke's textbook model is partly wrong, and how the brain maps consonants and vowels onto motor movements. Chang explains the mechanics of the larynx and vocal tract, bilingualism, dyslexia, stuttering, and epilepsy treatments including surgery and the ketogenic diet. The centerpiece is his BRAVO trial work decoding speech directly from a paralyzed patient's brain, and the future of avatars, augmentation, and companies like Neuralink.

Big reveals

  • Chang says the textbook idea that Broca's area is the basis of speaking is 'fundamentally wrong'; the precentral gyrus is the key area.
  • When asked what fraction of medical-school brain teaching is correct, Chang estimates about 50% is accurate and 50% is oversimplification.
  • Speech features are laid out in a 'salt and pepper' map in the cortex, not a clean universal order, and speech may bypass primary auditory cortex.
  • All words are built from roughly 12 articulatory movement features, which Chang compares to DNA's four base pairs generating all of life.
  • Chang recounts Pancho, paralyzed 15 years, having his attempted speech decoded into text for the first time via brain electrodes.
  • Pancho's giggling fits of joy when words appeared on screen actually corrupted the decoding of the next word, a bug they never fixed.
  • Chang debunks 'stroke gives you a new language' stories but confirms foreign accent syndrome is real after precentral gyrus strokes.
  • Chang predicts avatars that speak our typed text and says digital/virtual social interaction is where communication is heading.

Things worth remembering

  • Raising rat pups in continuous white noise kept the auditory critical period open for months, delaying brain maturation.
  • Despite his expertise, Chang chose NOT to use a white noise machine for his own three children, preferring structured natural sounds.
  • A young woman was misdiagnosed with anxiety disorder for years; her panic attacks were actually seizures originating in the amygdala.
  • Handedness is strongly genetic; in right-handers language is on the left 99% of the time, but only about 70% in left-handers.
  • Male vocal folds vibrate at about 100 hertz and female voices at about 200 hertz due to larynx size differences.
  • Hawaiian uses only about 12-14 phonemes while English has about 40 consonants and vowels.
  • Patients who cannot speak a sentence can often still sing 'Happy Birthday' or count, because that memory is distributed.
  • The BRAVO trial started with a 50-word vocabulary and used autocorrect-style language models to fix imperfect brain decoding.
  • Chang notes humans have always augmented cognition with coffee and nicotine, so augmentation itself is not new territory.
  • Chang avoids caffeine and treats the operating room as a 'sanctuary' where he disconnects, never bringing his phone in.