127 | Erich Jarvis on Language, Birds, and People
What this covers
Sean Carroll hosts a conversation with neurogeneticist Erich Jarvis on the origins and mechanisms of spoken language. Jarvis argues that vocal learning—the capacity to imitate and produce novel sounds—is not unique to humans but appears in a small number of distantly related mammal and bird species (humans, cetaceans, bats, elephants, and seals among mammals; parrots, songbirds, and hummingbirds among birds). Rather than invoking a uniquely human language module, Jarvis proposes that vocal production learning and speech rely on the same forebrain circuit responsible for motor learning more broadly, likely arising from brain circuit duplication of ancient pathways that control movement. This architecture appears to have evolved independently yet similarly across distant lineages—a signature of convergent evolution.
The discussion moves across several layers of evidence and implication. Jarvis explains how the specialized vocal circuit shares cellular machinery with motor-learning pathways, how differential gene regulation rather than entirely novel genes distinguishes vocal learners, and how social communication over distance—enabled by sound traveling farther than sight or smell—selects for greater intelligence and thus indirectly for language ability. He addresses the hand gesture theory of language origins by noting that one human vocal circuit sits adjacent to hand-control areas in the cortex. The conversation also challenges longstanding assumptions: the descended larynx hypothesis for human speech is questioned using work by Tecumseh Fitch, and Chomsky's notion of a separate language module is critiqued as unsupported. A secondary thread covers Jarvis's work leading large-scale genome sequencing projects to understand the genetic basis of these traits across vertebrate diversity.
Jarvis argues that vocal learning—the rare ability to imitate and produce novel sounds—evolved independently in a handful of mammal and bird lineages via duplication of an ancient motor-learning brain circuit, and that this circuit, not a separate Chomskyan language module, underlies spoken language and drives the social communication that selects for greater intelligence.
- Vocal learners possess a specialized forebrain motor circuit absent in non-learners, likely arising from duplication of pre-existing motor pathways.
- Because vocal learning evolved independently yet similarly in birds and mammals, studying birds reveals analogous and deeply homologous mechanisms applicable to human language.
- Vocal learning enables long-distance social communication, which Jarvis proposes selects for intelligence rather than language ability per se.
Vocal learning circuits arise from duplication of existing motor learning pathways, not novel genes or anatomical structures.
- Species capable of producing learned sounds have a specialized forebrain circuit absent in non-learners, and this circuit shares many similarities with the brain pathways that control learning how to move—suggesting vocal learning is built from motor learning circuitry.
“we find that the species that can produce learned sounds have a specialized forebrain circuit that you don't find in the species that can't. Alright, and that circuit actually has a lot of similarities with brain pathways that control learning how to move.”
- Jarvis proposes that a pre-existing motor learning pathway duplicated to give rise to a vocal learning pathway, and that in humans and parrots that new vocal pathway duplicated again to produce a second one—an analogy to gene duplication, where brain circuits may duplicate during development as a natural process, making it easier than building a new circuit from scratch.
“What we think is that initially a motor learning pathway duplicated and gave rise to a vocal learning pathway, and then in humans and parrots, that new vocal learning pathway duplicated again and gave rise to a second one.”
- The vocal learning circuit likely duplicated from one of two specific motor circuits—either the oral-facial musculature circuit (since one human vocal circuit is intermingled with it) or the hand control circuit (since another vocal circuit is directly adjacent to the cortical hand area), the latter consistent with the hand-gesture theory of language origins given that humans naturally use and learn hand gestures while speaking.
“It's either the circuit that controls the oral-facial musculature, or the circuit that controls the hands.”
- Humans, songbirds, and parrots do not have extra novel genes associated with language; rather, existing genes found in mice, birds, reptiles, and fish have been modified (regulated differently or with sequence changes) in a similar way in the independent vocal-learning lineages—with the open exception of possible extra gene copies that might drive brain pathway duplication.
“we're not finding that humans and songbirds and parrots have extra genes that we think of as associated with language. We think it's modification of existing genes.”
- The vocal learning circuit is a motor learning circuit—the brain learning to control body parts to move muscles in specific ways (like learning piano or dancing)—but applied to the larynx; many species have learning and motor control capacities that are not linked, whereas vocal learners have linked their learning ability to their vocalization muscles.
“I'm proposing this is a motor learning circuit, and what do I mean by motor learning? I mean learning, the brain controlling body part to learn how to move muscles in certain ways.”
- The vocal learning circuit in a parakeet and a human is analogous (independently evolved in a similar way), not homologous (the same); but the surrounding motor circuit for learning to move body parts, found in all vertebrates, is homologous—specifically a 'deep homology' because it gave rise to the analogous speech circuits in each lineage.
“the word analogy or analogous means that it's independently evolved but in a similar way, and homologous means it's the same... the surrounding motor circuit found in all vertebrates for learning how to move other parts of the body, that's homology. And we would call that a deep homology, because it gave rise to the analogous circuit for speech.”
- Although birds do have a cortex, its cellular organization differs from the mammalian cortex: rather than being arranged in layers as in humans, bird cortical cells are clustered, even though it functions as an ancient motor-learning pathway.
“the bird brain, the organization of the cortex, part of the bird brain, they do have a cortex, but it's not layered as it is in humans. The cells are actually clustered in their organization.”
- Although vocal learning evolved independently in songbirds and humans, because traits that evolve independently can evolve in similar ways, the similarity means studying these animals' brain mechanisms reveals relevant insights into human spoken language mechanisms—turning birds into a tractable model system.
“we're finding because things that evolve independently can evolve in a similar way, it does mean that since it's similar, you can learn about the human mechanisms from studying these animals, and this is what we've been discovering.”
Vocal production learning, distinct from auditory comprehension, is the rare ability enabling true spoken language across species.
- A cat's meow is a genetically pre-determined sound; the cat can learn to use it in different contexts (vocal usage learning), but cannot produce novel sounds it was not born able to produce (vocal production learning), which is the rare ability of interest.
“The cat can learn how to use a meow in different contexts, like for food and water, and we call that vocal usage learning, but not vocal production learning, which is the ability to produce novel sounds that you're not born to be able to produce.”
- Koko the gorilla understood roughly 2000 to 7000 words and could sign about 1000 to 2000 with her hands, which is a lot for a non-human animal, but she could not utter even a single word vocally—illustrating the huge gap between comprehension/gesture and vocal production.
“I will say that just putting it in relative perspective, Koko can't even utter a single word.”
- Spoken language has multiple components: auditory learning (forming auditory memories and understanding sounds, which is common across species like dogs) is necessary but not sufficient; vocal production learning (imitating and modulating novel sounds) is the rare component required for true speech.
“This is necessary for spoken language because we need to form an auditory memory of what we're going to imitate in our speech, but it's not sufficient.”
- Species without vocal learning can still possess symbolic understanding: dogs can learn the meaning of objects and whole sentences, and vervet monkeys naturally associate distinct innate alarm calls (eagle vs snake) with different meanings that are culturally passed between generations, even though the ability to modify those sounds is innate.
“in a vervet monkey, you can get them to... they do this naturally out in the wild where they all associate even some of their innate sounds, whether an eagle in the sky or a snake on the ground, different innate sounds will mean different things that they culturally pass on from one generation to the next. But the ability to modify those sounds is innate.”
Vocal learning evolved independently in eight non-closely-related vertebrate orders, demonstrating convergent evolution like wings.
- Crows are vocal learners: although they lack a loud advertising song, they produce low-volume song to each other and their loud crowing vocalizations are learned, and there are documented cases of crows picking up human speech sounds.
“the crowing vocalizations that they produce that you can hear and that's pretty loud, is also learned. And there have been cases where crows have picked up human speech sounds.”
- Of roughly 70,000 vertebrate species grouped into about 260 orders, only 8 orders are vocal learners: among mammals it is humans, cetaceans (whales and dolphins), bats, pinnipeds (seals and sea lions), and elephants; among birds it is parrots, songbirds, and hummingbirds—and these groups are not close relatives, so the ability evolved independently.
“of those 260 orders, 8 of them are the vocal learners.”
- Because something as complex as spoken language can naturally evolve multiple independent times from the same biological substrate, Jarvis infers that complex life—and life itself—could likewise evolve more than once, making it more than an N of 1 in the universe.
“if you're starting from the same substrate and just can naturally evolve something as complex as spoken language, or at least something close to it, multiple times, why can't you evolve something as complex as life multiple times?”
- Vocal learning evolving in wildly different species is convergent evolution analogous to wings, which independently evolved in bats, birds, and ancient flying dinosaurs always at the sides of the body due to the physical selection pressure for aerodynamic center of gravity—nature repeatedly selects for the same trait given similar constraints, and spoken language is the same kind of repeated solution.
“it's kind of like the evolution of wings. They evolve in bats and birds, some ancient flying dinosaurs, and each time it's at the sides of the body... because the center of gravity, there's this selection pressure... So nature is selecting for this trait, and I think it's the same thing when it comes to spoken language.”
Brain circuit duplication during development and differential gene regulation of conserved genes, not novel genes, drive vocal learning evolution.
- Parrots and songbirds lack the extra SRGAP2 copy but have roughly twice the density of neurons packed into the same brain volume as non-vocal-learning species, and this doubling—without increasing brain size—is argued to provide extra brain circuits available for vocal learning and other behaviors.
“in parrots and songbirds, the vocal learning birds, they don't have an extra copy of this gene, but they do have a greater, twice the density of neurons packed in the same amount of space as a non-vocal learning species, and I think that this doubling of neurons in the brain, without increasing brain size mathematically, allows them to have more extra brain circuits for vocal learning and other behaviors.”
- Although species including humans have huge genomes, they contain only about 20,000 protein-coding genes; nature could fit far more but instead uses this limited set as a reusable toolkit, deploying the same genes in different ways via intergenic regulatory regions to control different brain connections and cell types.
“only 20,000 protein-coding genes. If nature wanted to, it can fit 100 times that many genes inside our genome... So nature has a way of taking a limited set of 20,000 genes as a tool kit. Think about 20,000 tools and then use them in different ways”
- Despite birds looking very different from humans, over 80% of their genes are the same genes; they differ in how those genes are regulated and in some protein-coding sequence divergence, and it is this differential regulation that makes a bird a bird and a mammal a mammal.
“Even though a bird looks very different from a person, over 80% of our genes are the same genes. They're just regulated in different ways and do have different protein-coding sequences to them as well”
- Humans have an extra copy of the gene SRGAP2 that keeps neurons in a more immature state throughout life, which is argued to allow us to learn more readily across our lifespan—though Jarvis notes we may also forget equally more readily as a consequence.
“in humans, we have an extra copy of a gene called SRGAP2, and this extra copy causes the neurons to stay in a more immature state throughout our lives, and that more immature state is argued to allow us to learn more readily throughout our lives.”
- FOXP2 is a transcription factor that regulates other genes and has two nucleotide mutations producing two amino acid differences in humans not seen in chimps or most other vertebrates, which is thought to somehow enhance the spoken language circuitry's function—one of the genes consistently different between humans and chimpanzees.
“FOXP2. It's a transcription factor that regulates other genes, that it has two nucleotide mutations, just two amino acid differences that you don't see in chimps or most other vertebrate species that we think somehow enhances spoken language”
Human language ability differs from other vocal learners in degree, not kind, despite unprecedented capacity for abstraction and recombination.
- Jarvis proposes humans have greater capacity for abstract thought because thinking occurs in the auditory, visual, and speech pathways: when we think through the speech pathway we are talking to ourselves silently, sending that internal speech to the auditory pathway to 'hear' it—so parrots may talk in their own heads while a dog can only hear in its head, not talk in it.
“when we have thought through our speech pathways, we are talking to ourselves, we're not actually producing the sounds, but we're talking in our head, and that speech that's being produced in our head is being sent to the hearing pathway... So I think parrots will talk in their own heads, but a dog will only hear in its own head, it won't talk in its own head.”
- Humans overrate their own abilities and thereby underrate other animals; among vocal learners there is anecdotal evidence of greater intelligence—particularly social intelligence—brought on by the greater capacity for learned communication, rather than a tight necessary link between vocal learning and general intelligence.
“Overall, I think we humans overrate ourselves and thereby underrate other animals, including the non-vocal learner, but among the vocal learners there is some anecdotal evidence that the more learned kind of communication we do... there is a greater intelligence, but I'm going to say more of a social intelligence than anything”
- Humans are special for spoken language, but the specialty is one of degree, not a have/have-not distinction: we are far more advanced than even other vocal learners at recombining sounds and using them for things like writing, reading, and forming new concepts, even though some parrots can produce sounds humans cannot.
“we are special for spoken language, but the specialty is not have and have not, okay? The specialty is that we're much more far advanced for this ability than even the other vocal learners.”
- Jarvis accepts a genetic predisposition (language instinct) to learn to imitate sounds but disagrees with Chomsky's separate language module: there is no good evidence for it, and the brain pathway controlling grammar and syntax is the spoken-language motor learning pathway itself, fed by the pre-existing auditory pathway (including Wernicke's area) common to many vertebrates—which is why dogs can understand rudimentary human speech.
“there's a language module in the brain that then controls the vocal pathway and the auditory pathway... I disagree with that. I don't think there's good evidence for a separate language module. I think the brain pathway that's actually controlling the grammar, the syntax... is the spoken language pathway itself, is the motor learning pathway”
Vocal learning enables greater social communication networks and selects for increased intelligence through mate attraction and social signaling.
- Jarvis proposes that it is vocal learning—not language ability generally—that is associated with greater intelligence, via a causal sequence: vocal learning leads to greater social communication, and greater social communication then selects for greater intelligence; syntactic communication via gesture, vision, or olfaction already exists in many species but does not select for intelligence the way vocal-learned communication does.
“I'm going to call it vocal learning ability associated with greater intelligence, and then why that could have occurred, is that vocal learning leads to greater social communication. And greater social communication, I think then selects for greater intelligence, in that sequence of order.”
- Vocal learning is used in social contexts for mate attraction and territorial defense, and because more varied vocalizations signal greater intelligence and health it improves mate attraction; crucially, since sound propagates over long distances unlike vision or smell, vocal communication enables a far greater social network with individuals you cannot see (e.g. dolphins using signature whistles as names).
“sound propagates a long distance so you can communicate with somebody a long distance without actually having to see them. You have a greater social network with this than you get from visual or smell.”
- Stephen Nowicki showed that nutritionally depriving a songbird at a young age negatively impacts the brain so that it grows up to sing an impoverished song; females dislike the impoverished song, so the bird fails to pass on its genes—demonstrating that song quality is an honest signal of developmental health used in mate selection.
“Stephen Nowicki at Duke University has shown that if you nutritionally deprive a songbird at a young age, and then it grows up to be an adult, it sings what sounds like maybe an impoverished song... Females don't like that impoverished song, and that bird doesn't get to pass on its genes.”
Parrots demonstrate vocal learning capacity by recombining learned sounds into novel contextually appropriate utterances not taught to them.
- Parrots can take learned human speech sounds and recombine them into new words and sentences that were not taught to them, conveying meaning in social context; Irene Pepperberg's experimental work on African gray parrots showed this occurs in a statistically significant, non-random, contextually appropriate manner rather than being human pattern-projection.
“Irene Pepperberg, who's done some work on African gray parrots, doing experimental analysis of this, demonstrates that this happens in a non-random manner. It's statistically significant that her parrots were producing words and combination words in appropriate social context.”
- Parrots can produce more complex learned sounds than other vocal-learning birds for two reasons: they use the tongue (like humans) in addition to the larynx to modulate sound, and they—like humans—have an extra brain pathway, a 'song system within a song system,' that controls this behavior.
“One is because they use the tongue more than the other species, like we use our tongues, not just our larynx to modulate the sound. Second is they, parrots and humans, have an extra brain pathway. It's like a song system within a song system that controls this behavior.”
Vocal learning requires high energy and predator exposure costs, so species must gain substantial social or reproductive benefit to evolve it.
- Vocal learning is costly in two ways: it exposes the animal to predators (you'll just be dead), and it requires large amounts of energy (glucose, ATP) because the vocal learning brain circuit has very high firing rates and the larynx muscles are among the most rapidly firing muscles in the body—so a species must benefit enough to justify the resource cost.
“the other is, it takes a lot of energy, glucose, ATP, and so forth, in this brain circuit, which has very high firing rates... the larynx muscles are the most rapidly firing muscles in the body. So I think it takes a lot of actual real energy for the body to actually produce spoken language.”
- Vocal learning may be associated with being near the top of the food chain, because producing varied novel sounds makes a species more conspicuous to predators that have a hard time habituating to it; evidence includes that songbirds and parrots descended from apex predator 'giant parrot birds' of the dinosaur era, while bats avoid the cost by vocalizing in the ultrasonic range.
“if you evolve this trait, you're producing varied sounds that predators have a hard time habituating to and you're more likely to be eaten.”
- The long-standing Lieberman hypothesis that the descended human larynx enables greater sound variety is weak or refuted: Tecumseh Fitch's work shows other animals can lower their larynx as far as humans when raising their heads, some animals (e.g. certain lions) independently evolved a descended larynx without speech, and blowing air through a post-mortem macaque larynx produces human-like phonemes regardless of laryngeal descent.
“the work of Tecumseh Fitch has shown that this is not true, that other animals, when they raise their heads up, can get their larynx descended as low as humans, and there are some animals that independently evolve a descended larynx, in some lions, but yet they don't imitate sounds and produce speech like we do.”
High-quality genome sequencing of all vertebrate species and trait databases will reveal genes controlling vocal learning and disease resistance.
- Once all species are sequenced, sequencing becomes unnecessary and the focus shifts to building a database of species traits alongside genetic codes; combining trait and genetic databases will make it far easier to identify genes needed to create drugs against diseases, since many species are disease-resistant and others are not, revealing the responsible genes quickly.
“with a database of traits and a database of their genetic codes, that'll make it a lot easier to figure out what kind of genes you need to create certain drugs against certain diseases, because there would be lots of species that are resistant to a disease and others that are not.”
- Jarvis chairs an international consortium aiming to sequence the genomes of all roughly 70,000 vertebrate species at high quality—because high-quality, gap-free, error-free assemblies enable far better biological investigation—currently in phase one covering 260 species representing all vertebrate orders, as part of an even larger effort to sequence all eukaryotic species (the Earth BioGenome Project).
“I've now become chair of a large international consortium whose mission it is to sequence the genomes of all vertebrate species on the planet, so all animals. 70,000 of them. And to do it at high quality”
- Jarvis's lab is testing whether language arises from modification of existing genes rather than new genes by inserting human versions of vocal-learning-related genes into the mouse brain to see if a non-imitating species can be made to imitate sounds; so far only one gene has been modified, and he believes dozens must be modified to produce the effect.
“what we're trying to do is take the human version of those genes and put them in the mouse brain, a species that can't imitate sounds the way we do. To test the hypothesis, if we take the human version of these genes that causes disconnectivity, will we get the mouse to actually learn how to imitate sounds?”
- The Pan-Human Genome Project predicts that sequencing roughly 300 people representing about 350 populations or ethnic groups will determine the conserved human genome everyone shares versus what differs between groups, and identify the percent that is unequivocally human—the sequence beyond which one is no longer human—a value not yet known but believed to exist.
“if you sequence the genomes of about 300 people, representing roughly 350 different populations or ethnic groups on the planet, that you'll basically determine what is the conserved human genome that everybody shares and then what is different from one group to another.”
- Contrary to common assumption, the Tree of Life is far from resolved: the bird family tree kept changing year to year (owls and hummingbirds shifting positions), which prompted Jarvis's group to use whole genomes to produce a more stable bird tree, and the ultimate tree will be partly network-like rather than a simple bifurcating tree, reflecting that all individuals are cousins rather than strict parent-child lineages.
“every year, every other year, the bird tree, the family tree of birds, is changing. Owls going from one position to another... I said, 'Wait a minute, these guys don't know what they're talking about.' The tree is not really set... the Tree of Life is not resolved. It's far from resolved.”
- Jarvis jokes that he wishes humans would evolve a little faster given current politics, to which Carroll replies that evolution is not directed, so evolving may not be what we want.
“I just wish we humans, we'd evolve a little faster considering our current politics.”
Specialized forebrain circuits controlling vocalization muscles distinguish vocal learners, resolvable through electrode recording and molecular mapping.
- Scientists identify which brain areas are active during behaviors like singing using two main approaches: implanting electrodes to record neuronal firing during the behavior, and molecular mapping of immediate-early genes whose messenger RNA and protein products go up or down with neural activity, measured via techniques like immunocytochemistry or in situ hybridization after dissecting the brain.
“There are certain genes in the brain that respond to that activity... Their messenger RNA and protein products go up or down, and we can use this like a molecular mapping tool to identify brain areas that are active during particular behaviors.”
- The key connection Jarvis's lab is trying to engineer is the direct control the forebrain cortex has over the neurons modulating vocalization muscles—a huge robust connection present in humans and songbirds but weak or absent in other species—by locally up- or down-regulating specific gene products that control these connections without changing expression elsewhere.
“the one connection we're trying to modify the most is one that the cortex in the forebrain has a direct control of the neurons that modulate the muscles for vocalizations. In other species we don't get that direct connection... In humans and songbirds it's a huge, robust connection from the forebrain.”
Critical periods for vocal learning exist in humans and songbirds, reflecting developmental constraints on acquiring learned acoustic production.
- Like humans, songbirds have a critical period in which they learn sounds best at an early age and find it harder to learn new sounds later in life; some songbirds simply cannot learn as adults, while others (like humans) can but never as well as in infancy.
“this critical period like we humans have, when you learn best at an early age in life, and then it's harder to learn new sounds later in life. And for some songbirds, they just can't.”
- Within birds there is no evidence that vocal learning evolved independently in different lineages, but evidence that it was lost in some, such as the Anna's hummingbird and in the females of many songbird species; near the Equator both adult males and females learn to sing, while further from the Equator females more often lost the ability amid greater division of labor between the sexes.
“It turns out, the Equator, where most songbirds are at, adult males and females learn how to sing. The further you get away from the Equator, to higher probability that females lost the ability, where there has been more of a division of labor between the sexes.”