The Neuroscience of Speech, Language & Music | Dr. Erich Jarvis

Andrew Huberman opens by introducing Dr. Erich Jarvis, a professor at The Rockefeller University and a Howard Hughes Medical Institute investigator, whose laboratory studies the neurobiology of vocal learning, language, speech disorders, and the relationship between language, music and movement, in particular dance. Huberman describes the work as spanning genomics, including the genomes of vocal species such as songbirds and parrots, through neural circuits and up to cognition. He notes that HHMI investigators must re-compete every five years to keep their funding.

Huberman previews two ideas he says he learned from Jarvis. First, while reading, people generate very low levels of motor activity in the throat, speaking the words below the level of perceptible sound, which an amplifier on the vocal-cord muscles could detect. Second, there is a direct link between the species that have song and movement, which many people associate with dance, and the ability to learn and generate complex language. He frames the episode as relevant to people with speech disorders such as stutter, to people learning multiple languages, and to anyone interested in how we communicate through written or spoken words.

Huberman starts with the distinction between speech, which he thinks of as motor patterns producing meaningful sound, and language. Jarvis calls the question provocative. He says he struggled with it for years and concluded that the terms are behavioral, psychology-derived categories that "don't actually align exactly with brain function." Looking at human brain research and at comparative work with vocal imitators such as parrots and songbirds, he no longer sees a sharp distinction.

Jarvis lays out the view he disagrees with: a separate language module containing the algorithms and computations that then steer a speech-production pathway and an auditory pathway. His alternative is that the speech production pathway, which controls the larynx and jaw muscles, has the complex algorithms for spoken language built into it, and the auditory pathway has the algorithms for understanding speech built into it.

And it turns out, I don't think there is any good evidence for a separate language module.

He adds an asymmetry between the two pathways. The speech production pathway is specialized to humans, parrots and songbirds, while auditory perception is far more widespread in the animal kingdom. That explains why animals can understand speech they cannot produce.

Dogs can understand several hundred human speech words. Great apes, you can teach them for several thousand, but they can't say a word.

Huberman asks which animals have language or language-like communication, mentioning whale songs, dolphins and birds. Jarvis says a narrow definition would count only produced sound, then asks about hand gestures and the aerial displays some birds use to communicate through body language. Going back to the brain, he says the regions for gesturing with the hands sit next to those controlling spoken language, and that this parallel hand pathway also has complex algorithms. Species differ in how advanced each circuit is: humans and a few others are most advanced for spoken language, while a non-human primate can gesture in a more advanced form than it can produce sound.

Just to say that humans are the most advanced at spoken language, but not necessarily as big a difference at gestural language compared to some other species.

Huberman asks about cultures, such as Italian, that he associates with a lot of hand movement. Jarvis does not cite imaging studies in reply. He points out that he and Huberman are gesturing unconsciously as they talk, and that people gesture on the telephone even though nobody can see them. Some have argued, and he says he agrees based on what his group has seen, that there is an evolutionary relationship between speech and gesture pathways. He thinks the speech pathways evolved out of the pathways that control body movement. He adds that each language, whether Italian, French or English, comes with a learned set of gestures.

His example is Koko (spelled "Coco" in the transcript), a gorilla raised with humans for 39 years or more. Koko learned gesture communication and sign language and could understand both signing and spoken speech, but could not produce the sounds with her voice. Jarvis's interpretation is that a number of species, though not all, have motor pathways for learned gesturing, which he calls rudimentary language with the limbs, but lack the extra pathway that would let them "gesture with their voice in the way that they gesture with their hands."

Huberman offers a hypothesis from his background in vision science. Primitive emotional sounds tied to respiration might be the base of language, the way visual images are built up hierarchically from dark and light contrasts. His examples are inhaling and saying "Mm" at a delicious smell, and wincing, turning away and exhaling at a putrid one.

Jarvis says the idea is not crazy and that it touches a key distinction in vocal learning research. Most vertebrates vocalize, but mostly with innate sounds they are born producing, such as babies crying or dogs barking. Only a few species can imitate sounds. He argues the split between innate and learned behavior is a bigger dichotomy for vocalizations than for other behaviors in the animal kingdom.

When people think of what's special about language, it's the learned vocalizations. That is what's rare.

The same split appears in the brain. Breathing, grunting and similar sounds are largely handled by brainstem circuits, around the level of the neck and below, in a reflex-like way, and some emotional aspects involve the hypothalamus. Learned behaviors, such as learning to speak, to play the piano, or a dog learning tricks, use forebrain circuits. Many species have forebrain control over learned body movement but not over vocalization. In humans, parrots and some other species, Jarvis says, the forebrain has taken over the brainstem and now uses it to produce learned vocal behaviors as well as innate ones.

Jarvis prefaces his answer with a caveat: "I think we humans overrate ourselves compared to other species." He says this leads even scientists astray when forming hypotheses, especially because fossil evidence of language is hard to find. Among primates, humans are the only ones with advanced vocal learning. It was assumed this ability belonged only to Homo sapiens.

His argument uses two lines of evidence. First, genomic data from living humans and from Homo sapiens, Neanderthal and Denisovan fossils indicate that human ancestors supposedly hybridized with these other hominids. Jarvis says he knows of no living vocal-learning species that can have offspring with a non-vocal-learning species, while adding that this "Doesn't mean it didn't exist." Second, in these ancestral hominids, genes that function in speech circuits have the same sequence as in humans.

He concludes that Neanderthals had spoken language but declines to say it was as advanced as in humans, adding "I don't know." His estimate is that the ability has existed for between 500,000 and a million years, and that humans have become more advanced with it culturally and possibly genetically.