About the guest: Ev Fedorenko is a neuroscientist. Her lab at MIT studies the neural basis of language, speech, and thought in the human brain.
Episode summary: Nick and Dr. Fedorenko discuss: the relationship between language and thought; the extent to which language is for thinking vs. communication; Noam Chomsky’s Universal Grammar theory; Sapir-Whorf hypothesis; language acquisition & language learning; language networks in the brain; neuroanatomy & brain lateralization; large language models (LLMs) & machine intelligence; and more.
Related episodes:
M&M #141: Evolution, Language, Domestication, Symbolic Cognition, AI & Large Language Models
M&M #20: Language, Symbolic Cognition, Evolution, Origins of the Human Mind | Terrence Deacon
*This content is never meant to serve as medical advice.
Full audio version: [Apple Podcasts] [Spotify] [Elsewhere]
Support M&M if you find value in this content.
Episode transcript below.
Full AI-generated transcript below. Beware of typos & mistranslations!
Ev Fedorenko 1:31
and I am a faculty member in the brain and cognitive sciences department at MIT, and I'm also a member of the McGovern Institute for Brain research also at MIT, and I work on the language system in humans, and trying to understand how it fits with all of the other stuff that we have in our minds and brains. So trying to kind of understand language in this broader landscape of the human mind.
Nick Jikomes 1:55
And, you know, briefly, how did you get into this? Why is this an area of particular interest for you?
Ev Fedorenko 2:00
Yeah, it's, you know, somewhat accidentally, really, like, I've always liked languages, but more from like, I was like math as well. And I liked kind of seeing, I was learning a bunch of languages as a kid, and I liked seeing, kind of the patterns and similarities and differences different solutions to the same kinds of problems that different languages solve in different ways, and I didn't really know what to do with that. And I came to to Harvard, Harvard College as an undergrad, and then I saw that there's a class called psychology of language was taught by Professor Alfonso caramazza. And I took this class, and it was just immediately obvious to me that this is what I'm gonna keep doing kind of forever, which is basically trying to use methods from psychology and cognitive science and neuroscience to try to understand how it is that humans understand and produce language and are able to express the kind of, you know, kind of complexity of our thoughts and all that stuff. And so I got hooked then, and then stuck with it before
Nick Jikomes 3:04
you started studying language as an academic discipline. So where are you from and What languages do you actually speak? Well,
Ev Fedorenko 3:10
I grew up in Russia, so I was speaking Russian until, well, until I moved, I guess, to the US. That was my primary language. And I was learning English in school. I was like in an English intensive school, and then I was additionally studying French and German and Spanish and polish and a little bit of this and that, a few other things. But I think it was basically my mom's way to build a way to escape for me, because she knew that the country was flawed in various ways, and she saw that, she saw some potential in me to, you know, pursue academic questions like or an academic career. And I think she just wanted to make sure that I have the tools, which, of course, first and foremost is being able to speak a language wherever you are, whatever it is that you're trying to escape to. And we were kind of covering broad ground with studying a lot of languages. But through that, I actually got really interested in the process of learning and processing language that kind of worked out in more than one way.
Nick Jikomes 4:17
So one of the things that's very interesting about language is that it's simultaneously extremely familiar to pretty much everyone and also very mysterious at the same time we all use it, we all learn it effortlessly, seemingly as children, and yet it's very mysterious, where it comes from, how it rises and really how it works, which is a lot of what we'll talk about today. So I'll just ask you a vague question. But what is language? How do you define it?
Ev Fedorenko 4:43
A very good question and a hard question. So the way that I define language, and then the way that I think about it is basically a set of conventions that we as a society kind of agree on, not necessarily explicitly. There's a lot of implicit le. Happening, but basically it's a set of conventions to refer to objects and actions and events in the world that we adopt as a group, and then we can use that code that we agree on to communicate that about our internal mental states, right, what we feel, what we think, how the world works, right? We can teach each other things using this code, but also just express our emotions and frustrations and beliefs, and basically, you know, communicate anything about our inner worlds and the outer world.
Nick Jikomes 5:33
And then, you know, another thing that's that's obvious but but often defies a systematic explanation for people, is so on the one hand, languages are very diverse, right? If you look at Russian or Chinese or Hindi English, you know, all of the languages you can think of, they use a different combination of sounds. They have a different structure to them. The syntactical rules are different. When you think about that diversity, and then you think about the fact that everyone speaks a language, you know, the a common, a common thought has been, there's, you know, there's maybe some something universal here that all languages share in common. Can you talk a little bit about the structure of language, maybe in terms of how a linguist would talk about it, and whether or not there truly are universals, and what those might be? Yeah,
Ev Fedorenko 6:19
another great question. I mean, so there are indeed a very large number of languages signed and spoken around the world. By current estimate, it's around 7000 but a lot of languages are in the process of dying, which is sad. It's losing part of our cultural heritage and languages, indeed, it's one of the kind of communication system that manifests in very different ways across different groups. And for a long time, the notion that languages share something, or languages share something, was couched within this framework of universal grammar, this idea that's due to Noam Chomsky, that we have some innate machinery that predisposes us to be able to acquire a system like language. There's not much support for this idea, in part, brought to the forefront by the success of large language models where nothing is built in, like GPT two for example, right? Nothing is built in, and yet you can learn language just fine. Nevertheless, there are indeed certain properties that have been discovered in the experimental tradition and in Corpus linguistics, where it seems like there are certain regularities that we find systematically across languages. Most of those regularities have to do it appears with communicative efficiency. So for example, the ways that languages choose to use certain classes of sounds and certain combinations of sounds to convey meaning differences. Typically ensures that the different words in the language are distinguishable enough, but the sound complexity is not too high, so that it's also possible to learn for a child learner, similarly, how the kind of the conceptual space, our meaning space, how that gets carved up into lexical items, right? Like, what do we choose to label with a word or construction and a language that seems to also follow some very clearly predictable, even mathematically predictable, regularities between kind of the meaning complexity and how well meanings are differentiated from one another, so how much information they can convey in particular context, and there's regularities in the grammar as well. So for example, all languages tend to put words that are connected to each other in terms of meaning, like a noun and a verb close by, and different languages do it in different ways. So the order in which those words come may differ across languages, but that tendency to try to kind of put things together that go together in terms of meaning that seems to be present very much cross linguistically, presumably driven by something like our memory constraints, right? It's hard for us to keep track of information that spans long context. And so we try to package things into these locally, kind of coherent sequences
Nick Jikomes 9:06
I see. So so let me try and parse that. So obviously, there's things that are sort of like trivially true across all languages. Obviously, we're all using sounds that we can classify as consonants and vowels in different ways. We're combining them in certain ways. There's some kind of syntax structure. It's obviously used for communication oftentimes, and we'll get into other things that might be used for elsewhere. But it sounds like what you're saying is there are so despite the fact that you have different languages that use different sets of sounds, right? There's phonemes or sounds that are present in English that are not present in Japanese and vice versa. All languages seem to share certain regularities whereby so there's trade offs between how the sounds are combined and the diversity of stuff that can be communicated. Created, and so languages all seem to be sort of optimized for combining sounds in a way that allows the language to be simple enough to be used and communicated to Even small children, and convey lots of information without getting too big and broad in what it's trying to communicate and requiring too many different sounds to be put together all at once. Right?
Ev Fedorenko 10:20
Yeah. And it's not just at the level of how sounds may are combined to make up words. It's also in how words get combined into phrases and sentences. So those kind of regularities tend to span multiple levels of structure in a language. But yes, that's exactly right.
Nick Jikomes 10:32
So, so even though different languages have different rules, syntax rules, so the subject and the object and the verb will be at different parts of the sentence. There are universal constraints, it seems that have to do with how far apart those things are within sentences.
Ev Fedorenko 10:46
Exactly, exactly. Yeah, yep, interesting.
Nick Jikomes 10:52
So, so I want to talk a little bit about communication in humans as compared to other animals, before we get into that exactly, is there a distinction that we should talk about between language and speech within humans? Yeah,
Ev Fedorenko 11:07
that's a that's a useful distinction, and I think helpful to highlight, because a lot of people conflate those, including in the scientific literature. But basically, there are perceptual aspects to language, right? So what you're hearing me right now, you're perceiving in your with your auditory regions, the sounds that I'm producing. I could switch to a language that you don't speak, right? That's totally unfamiliar to you, and those auditory regions would still be doing the same work. It's still perceiving the speech sounds. The only difference will be that you can't get any meaning out of the signals, right? But you'll still do the acoustic processing. And so this separation that we can make conceptually between kind of the lower level acoustic processing of speech sounds and the deeper interpretation of the signals that can be conveyed through speech as well as other modalities, right? Like sign languages rely on visual manual modality to communicate. This distinction seems to be instantiated in the neural mechanisms of how we process these signals. So like I was saying, you have these regions in the auditory cortex which process speech sounds. They're very selective for speech sounds relative to other kinds of sounds. So those regions that will respond to English speech sounds or Tagalog speech sounds or Hindi speech sounds will not respond or respond much, much lower when you're listening to music or, you know, hearing your dog bark or whatever. But those regions, in turn, are distinct from the higher level, what we call kind of, you know, language higher level language processing regions, which are the regions that recognize words, recognize structures in the signals when we know the language and put them together to create complex meanings. And so that distinction between the lower level speech perception and the higher level language interpretation is an important one. It's both conceptually and turns out empirically, that's how our system is set up, with those things being very separable, I
Nick Jikomes 13:04
see. So there's an important distinction to be made between the production and the the taking in of the speech sounds, the the acoustic or auditory stimulus itself, versus the capacity to map those sounds to meanings to meanings, exactly. That's
Ev Fedorenko 13:22
exactly, right? Yeah. So we didn't actually talk, I didn't say anything about production, but just like you have these auditory regions for speech perception, you have regions in your motor and premotor cortex that do the planning of speech patterns, right? And again, you can produce things that don't mean anything, right? I can say galapa doesn't mean anything, right? But I can say it, and those mechanisms that implement speech movements will be engaged just as much as if I'm saying something meaningful. But the high level system, of course, cares deeply whether you're saying something that actually conveys a meaning versus just has similar form formal property.
Nick Jikomes 13:56
And I suppose it's occurring to me that there are, I think there are probably neurological and psychiatric disruptions of the brain that can really sort of make this distinction jump out. So for example, I'm thinking of people with certain brain lesions where they appear to be talking fluently, but it's all just sort of garbled gibberish, or something like schizophrenia. For example, you might, you might observe someone who's speaking, they're producing they're using speech, they're producing words, but the logical structure doesn't carry a lot of meaning to it. Yeah,
Ev Fedorenko 14:28
schizophrenia is actually a little bit more complex. Schizophrenia, I think, affects an even higher level of thought that's somewhat independent of language. So even though it will, of course, meant those thought disruptions will manifest in language, because that's what we use to talk about, you know, to talk about our internal mental states. But I think schizophrenia is better thought of as actually a disorder of thinking, whereas aphasia, different kinds of aphasia may affect different pieces of the speech, perception of production machinery or this kind of higher level. The actual, you know, knowledge representations that we use to do the mapping between forms, linguistic forms. Yeah, and
Nick Jikomes 15:06
I think we'll get into later some more detail around, you know, distinctions and similarities between linguistic processing and thinking or other cognitive operations. But I want to come back to the communication piece of this, and the sort of Comparative Zoology that's, I think it's interesting here. So how is human language different from other forms of sophisticated animal communication? I'm thinking here of things like the songs of songbirds or or whales and dolphins and things like that.
Ev Fedorenko 15:35
Yeah, there's still a lot we don't know about animal communication, and we that's a very active area of research. As I'm glad that you said that the communication systems of other animals are sophisticated, and there's a lot of information that gets transmitted for a long time. For example, vocalizations of chimps, chimpanzees were thought to just be grunts, until some researchers went and studied this more rigorously using kind of controlled experiments. Turns out they convey a lot of information with those grunts. They just sound like grunts to us, you know, much like probably our speech sounds sound not very intelligible to them, obviously. So that's just like one big point is, I think there's a lot that we don't know yet that said there's obviously something different about human language. How exactly it's different, and what it is that we have that other animals don't have, as a very big question, a lot of people have tried to construe language as kind of the golden ticket that gave humans all of the sophistication that we have as a species, I think that's not, I think that's a little misguided. I think there's many things that differ between us and other animals, including the sophistication of mental processes that have nothing to do with communicative abilities, that have just to do with how we represent the world and reason about the world. But you know, obviously one thing that, given the richness of our knowledge structures and this communication system that we developed to communicate about those knowledge structures, one thing that language very unambiguously does is allows us to preserve knowledge and pass it along generations, right? So you don't need to start figuring out physics from scratch. You can just read a textbook and be like, Okay, here's what we know about the physics of the world right now, or here's what we know about how humans work as social agents, or here's what we know about math, right and you can build on this and this. Building knowledge upon knowledge is a really powerful thing that obviously has led to the civilizations we've been able to build. But in terms of how exactly language differs from other communication systems. I think the bigger, bigger differences lie in the sophistication of thought, which is merely reflected in the sophistication of the communication system. Because, of course, the given the social species, at least like if we look at the primate species, right, we're all very social animals. And so it has to be the case that our communication system kind of reflects whatever it is that's on our minds, such that we want to connect with each other and build social structures. And so, of course, if we have more sophisticated thoughts, this will be reflected in the sophistication of the communicative signals that we use. But I think it's not. I think the differences between humans and other species is not reducible to something like language in some isolated way.
Nick Jikomes 18:26
And so, you know, what's implied here is that the language faculty is one thing, but there are other cognitive faculties at play that don't necessarily depend on language, or that don't arise from language, and that's certainly been an area of debate, I think, in this realm for a while, to what extent is our thought coming from our linguistic abilities, and to what extent are linguistic abilities independent of, say, our ability to engage in advanced planning or complex social reasoning or sort of abstract mathematical abilities? Where do you stand on that? And sort of, what's, what's the thinking in the field right now? Is that still controversial? Or do you think we have a relatively clear
Ev Fedorenko 19:08
I mean, it's not controversial, as far as I'm concerned, but certainly will keep debating. I mean, it's a very interesting, it's a deep question that has been on the minds of many thinking people for many years. But, and, you know, of course, there's some people who are not so interested in data, like in some philosophical traditions, it seems there, they seem to be disconnected enough from data. So it kind of almost doesn't matter what the data tell you. They have their theories that they build based on, kind of just, you know, some, you know, conceptual primitives alone or something. But empirically, there is, I think, now, pretty compelling evidence that the language system is quite distinct from systems, different systems that support thinking abilities. So it's not like we have one brain region that does thinking. There's a whole bunch of distributed, large, large scale, what's called, what's called large scale, distributed. The networks in the brain that do different forms of reasoning, and the kinds of evidence that exist for the separability of the language faculty from these other systems are basically evidence from brain imaging studies, where you can show that different kinds of tasks selectively engage one or the other, and I can elaborate more on that and evidence from individuals with brain damage. So you already alluded to kind of the information we can glean from studying cases of brain damage, where basically you find clear dissociations between damage to the ability to use language as opposed to damage to something like the ability to reason or plan, and both of these kinds of evidence have painted what I think is a pretty clear and consistent picture, where we have the language system, we have a bunch of other systems that support reasoning, and they work largely independently. There's some open questions about how they co develop, which is harder to study because we can't experiment with what kind of inputs kids get, right? So, and a lot of things happen in the first few years. It's very hard to tackle those questions, but at least in the mature brain, those are very sharply distinct systems together. But yeah,
Nick Jikomes 21:17
yeah. And so just just to reiterate, there's long standing questions here around you know, is the language faculty a largely or mostly standalone part of our cognitive toolkit that's largely or at least partially independent of other forms of thinking, versus To what extent do other forms of thinking, all other forms of thinking, maybe just some forms of thinking, depend on linguistic ability. What you're seeing, what you seem to be saying, is that the data is pointing to our linguistic abilities being largely more of a standalone thing. And this would mean that if we look at patients with lesions or disruptions to language in some way, that their other forms of their other cognitive abilities are preserved. Can you? Can you give some examples of
Ev Fedorenko 22:01
that? Yeah, so, effectively, so, so the most kind of informative and striking cases in this realm are cases of like, really severe aphasia, where individuals basically get their whole language system wiped out, typically by by a large stroke. And to the best of our knowledge so far, like everything, every kinds of reasoning tasks that have been attempted with these patients, they can they can solve them just fine. They can do math. If they were able to play chess before, they'll continue to play chess. They can write music again, if they were a composer, they'll continue composing music. They can solve logic puzzles. They can do Sudoku, like any kind of they can social reason about others minds, like, why, why it might somebody behave in a certain way based on the beliefs that they have? Of course, all of these tasks have to be cast in a way that's you can convey in the instructions and without words, challenges too, but it's doable and but, but pretty much everything, all of these tasks they do, as well as control participants who have their language in fact, they're often really excited to do these tasks because they've been mandated with all sorts of things that they can't do. Right? Because, you know, physiologists would test them on various aspects of language, and they fail in all of those when
Nick Jikomes 23:17
you say these people have language deficits, is that in the ability to produce speech, or is also a comprehension? Can they no longer
Ev Fedorenko 23:23
knowledge? Yeah, no, no. They cannot map any forms to meanings anymore. They lost that set of mappings, which is what I think the kind of human language system stores and uses to interpret others utterances and encode our thoughts and to utterances, they lose that whole system. So whatever you've been learning across your lifetime, right about, okay, table means this, and run means this, and that's how you put them together. You lose all of that. And so imagine just kind of like being, you know, in LA, in a country where you don't speak the language, suddenly just don't have access to that code, right? But it turns out that their thinking proceeds okay, like it looks
Nick Jikomes 24:03
as if they can still do their taxes, they can still play board games, exactly. Yeah, interesting. So, I mean, a nice follow up to that is simply when you have patients with those kinds of lesions where the deficits seem to be limited to the language faculty, mostly or completely, what parts of the brain are involved there? Yeah,
Ev Fedorenko 24:24
so the language system is instantiated in a set of brain areas. They kind of span parts of the lateral frontal cortex and parts of the lateral temporal cortex. In most individuals, the language system lives in the left hemisphere, not always true. In rare cases, it lives in the right hemisphere, and that seems to work okay as well. But
Unknown Speaker 24:47
it's, it's this
Ev Fedorenko 24:49
set of brain areas which surrounds what some it's sometimes called perisylvian cortex, because the Sylvian fissure is this big kind of separate, separate separation between the front. Frontal cortex and the temporal cortex. Like you can if you take a brain in your hands, you can separate the temporal lobe quite, you know, clearly. And line that separates that from the frontal lobe is called the Sylvian fissure. And so para Sylvian just means around the Sylvian fissure. And there's a certain kind of stroke, certain kind of artery, middle cerebral artery, that supplies blood part of the brain, part of the brain, exactly. And so if you have a big stroke in that affecting that artery, you will often lose this kind of whole frontal temporal areas,
Nick Jikomes 25:31
okay, so you can disrupt certain parts of the brain. And the phenotype you see, the problem that manifests is largely, largely just a language deficit, other things are preserved. So these people, they can reason in other ways. As you said, they don't have other motor deficits or basically anything. Sometimes
Ev Fedorenko 25:47
they'll have motor problems, I mean, but again and of course, you know one thing that's important to remember when we talk about deficits is you will always have cases where multiple things are affected. Of course, there are individuals who have both language and cognition affected, and you'll have cases where a language and motor control is affected. But most striking are cases where you see these dissociations. So there's an old saying in the Neuropsychology world, which is dissociation, Trump's Association. Trump's known as bad connotations for other reasons, but dissociations are much more important. So when you see these cases where x is there and y is not, or y is there and x is not, that's really informative. It
Nick Jikomes 26:27
shows you that one isn't strictly necessary for the other, whereas when two things co occur, it's always tricky, because it might be coincidence,
Ev Fedorenko 26:34
yep, yep, yep, yep, and especially and the reason that those coincidences happen quite a lot is because language is right next door to many other brain areas. And so, of course, like if you have a big lesion, you'll often have damage to a bunch of nearby functionally distinct bits
Nick Jikomes 26:50
and so, SO, SO to SO, when we talk about a language deficit here, you were basically referring to the ability to map meaning, whatever exactly that means to sensory stimuli. In this case, we're talking about speech sound. So it's an auditory stimulus. Obviously, there's no strict limitation here for it to be auditory, because people can do sign language, for example, yeah, but there's some process by which more abstract meanings are directly linked to these, these sensory features. Can you unpack a little bit more about, like, what, what is going on there with that mapping and how it ties into, say, symbolic cognition more generally?
Ev Fedorenko 27:32
Yeah, that's, that's a big question. So you're basically indirect asking, like, what is meaning and like, what does that mean for you know, that mapping to exist. So how meaning is represented in the minds and brains is one of the biggest puzzles that a lot of people are still working on. But what we do know based on the evidence that exists is that meanings are in some sense abstract, in a sense that you can get to those meanings through different routes, right? I can tell you something in language, but I can show you a snippet of, you know, a silent film, and create complex mental structures in your mind, right? So you can get to these meaning representations by many different routes, and there's a lot of debates as to how much these meaning representations are tied to our sensory and motor experiences in the world. So these are the ideas that are referred to broadly as embodiment, embodied cognition. And I think there's a lot of evidence that you cannot reduce our meaning representations to sensory and motor experiences, in part, based on data from individuals who have very drastically different experiences, like individuals who are born blind or individuals who are born without limbs and therefore have very different experiential kind of path through the world, and yet they seem to acquire meaning representations that are very similar to the ones that individuals who have access to perceptual and motor modalities get so but the questions of, you know, how exactly these meanings are represented, Is it, you know, all kind of distributed vector like representations, like those that live in large language models, is there some more structured models of the world that emerge from this somehow? I think that's likely. But again, a very active area of research, and now, like, how those one is, like, how symbolic are those representations of thought, our structures of the world, that we infer from our experiences in the world, including through language, and how those map onto linguistic forms, is, you know, remains an active area of research, but it's, I don't think it's necessarily so different from kind of, you know, associations, in some sense, there's some association that you learn between, say, a string cat and some representation of cat Ness, right? Of, you know, fuzziness. I mean, there's a bunch of sensory features. There's also something abstract, something abstract enough so you can make inferences such that if I. Tell you, Okay, a dog is similar to a cat, and dogs can get, you know, this kind of disease. You can reason that cats probably can get that kind of disease too, based on their like, biological similarity and some structure representation of the animal kingdom, or something like that. But I don't know, maybe you have some more specific ways to couch this question, of like, well, I
Nick Jikomes 30:21
think, I mean, I mean it's, it's yes, yes and no, it's, sort of, it's a difficult area to even even talk about. But, you know, I have questions here around, sort of the I know that there's long standing questions around the abstractness of language. Are there truly abstract representations that are independent of more concrete sensory motor mappings and things like that. In the world, it sounds like you're saying that at least to some extent, there do seem to be abstract mappings of meaning that are parts of these language networks. And part of the evidence for that is that functionally, it sounds like you're seeing similar encodings of meaning in people with very different sensory motor experiences, either because they are missing limbs and things like this.
Ev Fedorenko 31:05
That's right, that's right. That's exactly right, yeah. So I guess it's also a useful distinction to make. So, so there's the language, which is kind of the forms that map onto the meanings. And then there's distinction between kind of like concepts, right? Like, what does cat mean, or what does run mean? And then kind of even more abstract representations of structure in different domains, right? Like reasoning about the physical world, like how gravity affects things, how things, kind of, you know, objects collide and whatever. And as well as, like social domains, like how people behave with respect to each other, like, what do they what inferences they make based on, you know, each other's behaviors and things like that. And there's a few of these domains that seem to be in the brain drawing on partially dissociable structures. So in addition to kind of this broad arsenal of kind of concepts of individual entities and actions, there's also these structured representations we build of different domains based on our experiences in the world. And all of that is abstract, but somehow through the link between word forms and concepts, we're able to, you know, learn about these domains from each other and communicate about these kinds of knowledge representations with each other pretty effortlessly.
Nick Jikomes 32:16
It's occurring to me too that you know something, there's a relationship here too between the the cognitive diversity you see among people and how you think about sort of the abstractness of language. And this is, I think, going to tie in, actually, to discussions we'll have around, you know, Language for Communication versus language for thought, and how language might reflect thoughts that don't themselves depend on linguistic faculties. But you know, sort of what's in my mind right now is, you know, I was just asking you about, like, how abstract is language? Blah, blah, blah, and it occurs to me that this, the answer to that question, in some sense, depends on the individual that you're talking about. So for example, I might ask you, what did you eat for breakfast
Ev Fedorenko 33:01
yesterday? I don't know, probably skipped it. Yes. Okay,
Nick Jikomes 33:05
so let's say you ate scrambled eggs for breakfast. I remember that that's a simple question. Anyone can answer it. Then I could ask you, well, how would you have felt if you had not eaten anything for breakfast? So now, sort of like another layer of abstraction, some people respond to that effortlessly, and they imagine, oh, I probably would have been, like, irritable, because I always need my morning breakfast or whatever. And some people really struggle to answer questions like that. And so, you know, if you were to actually image the brain of both people, you'd probably, you'd probably see a different level of engagement or abstractness in terms of how they're mapping those meetings.
Ev Fedorenko 33:40
Perhaps, yeah. I mean, it's certainly the case that some people make much stronger and richer inferences based on the information that's present. That's certainly true. And oftentimes we use language to judge about abilities like that, because that's the only thing that we can observe. Right? I can't look at somebody and make a judgment about how smart they are and how they think, right? But I can ask them, and through that, I can kind of probe the sophistication of the mental structures, which is kind of the whole basis of the Turing test, right? The idea is that if you can use that interface to get access to the internal structures, you can kind of, do you know, a bunch of controlled experiments that is yours, right? Like, if you know, ask about this kind of bit of knowledge. Does it have it or does it not have it? Does whatever system have it or not? If I ask you to make a few kind of reasoning steps in a particular domain, can you do it or not? But so, So language is a useful interface to probe the mental sophistication. But like I said, it doesn't seem that the medium of thinking is language itself.
Nick Jikomes 34:36
I want to talk a little bit more about the language network here in the neuroanatomy. So if I'm hearing you correctly, so far, there's not one discrete neuroanatomical locus of language in the brain where, if you it's you know, you lesion that spot, and language goes away. And it's not any other spot. But it does sound like there are multiple spots in this larger network, which are anatomically. Localized to certain regions that are all core, you know, coordinating with each other to give rise to the language faculty you mentioned. You know, roughly where some of those areas are. Can you talk a little bit more about where these areas are and the extent to which they are functionally specialized for language versus the extent to which they're doing other things that get co opted into language faculty, right?
Ev Fedorenko 35:20
Yeah. So, indeed. So, so I want to clarify a little bit that although it is absolutely true that language seems to be supported by this distributed network of areas, it does seem that there is one part of that network that is especially important, perhaps because it's kind of the first processing station when information comes in through the perceptual systems, but it's the posterior temporal cortex that seems to when damaged, to lead to kind of longer, lasting and more severe kinds of aphasia. So but you can also get aphasia by damaging other parts of that system. It just does seem that there's some privileged role that this region plays. But right? So one of the reasons that I kind of started using neuroscience approaches as I was interested in exactly this question, like, you know, language emerged pretty late in our species history, and so it would be very reasonable to kind of hypothesize that maybe when language emerged, it kind of CO opted bits and pieces of things that were already existing and kind of were used for similar kinds of things, right? Maybe, you know, a little bit of auditory machinery and a little bit of kind of social reasoning, because language is such a social tool, and a little bit of kind of abstract meaning representations, or something like that. And so the modern brain imaging tools are very well suited for asking these kinds of questions, because if you're a good experimentalist, you can design conditions that target a particular mental computation, like language comprehension, or, you know, mathematical reasoning or social perception and things like that. And so over the last, you know, bunch of years, almost, you know, decade, no more than like 15 years we've been at it. We've evaluated a whole bunch of these kinds of hypotheses of language sharing machinery with other things. And I was fully expecting to find that at least some parts of the system would be shared with other domains. But it turns out, again, empirically, none of the regions that respond strongly when we do language, and by do language, I mean either produce or understand language, because the high level language system gets engaged by both by both directions of mapping between forms and meanings. None of those regions respond when we do pretty much anything else from perceptual things like, you know, processing faces or spatial layouts to more cognitive things like doing math or doing reasoning, like social reasoning or doing logical reasoning, or, you know, episodic memory or representing meanings that got to us through a different medium than language. So it seems that the language system is really just this very specialized kind of store of these four meaning mappings, and everything else is separate and draws on distinct machinery.
Nick Jikomes 38:18
And, you know, so one of the things, so one of the things that's interesting here so, so you've told us that you can lesion parts of the brain and you see deficits in language without seeing deficits and other cognitive abilities. People can still play chess. People can still socially reason, they can do other sophisticated things. So the language faculty does seem to be standalone to some significant degree, and where that sort of makes me think, now is okay, if that's the case, if all of these other forms of cognition don't depend on language, then why did language evolve? And so this gets us to issues of what it's actually for, and why it was selected. And presumably it was strongly selected, although I guess I'm just making that assumption,
Ev Fedorenko 39:00
I assume. So, yeah, so if it wasn't selected
Nick Jikomes 39:03
for these other types of thinking or something like that, yeah, what do we think it's actually being used for primarily?
Ev Fedorenko 39:09
Yeah, I mean, I think it's, it's likely social functions, like being able to pass knowledge to each other through communication, right? So which, you know, you can very clearly see how that can improve evolutionary fitness, right? Like, if I tell you how to make a certain tool to protect yourself from predators or to, you know, get better food or whatever, that's very helpful, as well as kind of just facilitating cooperation, right? A lot of things that humans did which were required, for survival. Could not be done alone. So it required kind of people to work together, to, you know, scavenge or hunt, to protect, again, from, you know, other groups and so on, or other species and language of. Is very useful for coordinating action. It's a deeply embedded kind of social, social medium. So I suspect that it's some form of communicative use. What exactly it was, you know, we may never know, because there's just no way to conclusively answer evolutionary questions. It's just that tricky, although I think with some of the recent AI advances, there's interesting modeling work that could potentially be used eventually to ask these kinds of questions, like, under what conditions does a communicative system resembling language emerge? Like, what kind of pressures need to exist in the environment to have this system come about? But yeah, yeah.
Nick Jikomes 40:40
And it's so, you know, on the one hand, it's sort of, it's sort of trivial to say that language is about communication and communicating with each other. Like, of course it is. That's clearly one strong use case for it, but one of the like, one of the things. But the question is, is it only used for communication and social purposes, or is it used for these other modes of thinking, you've told us that there's, there's some indication that other modes of cognition don't depend on language. Is there anything in language, in the structure of how information is passed in the medium of language, that that support that it's primarily for communicating between individuals, rather than about engaging in other forms of cognition.
Ev Fedorenko 41:23
I mean, yeah. I mean, so there is the evidence that I talked about briefly where, like, basically, different features of natural languages seem to make them really well suited for efficient transfer of information. And that's not like a knock down evolutionary argument, because people may say, Well, there's all these cases of acceptation. Maybe language evolved for thinking, but then it became used for communication, and so it evolved these different features over time, or something like that. You know, it's possible, but
Nick Jikomes 41:52
it's also possible to test that. I think, yeah, yeah,
Ev Fedorenko 41:55
it's exactly it becomes kind of back is, because it's hard to test that's exactly right. But then again, in in adults, I can measure your activity when you engage in various forms of reasoning, and your language regions are not doing anything when you're doing when you're solving these different kinds of puzzles, engage in different kinds of thinking. And so that tells you that, well, at least in modern humans, the system is not doing thinking for you, because if it did, we would see evidence of that, and we just don't. So again, you know, I try to typically stay away from evolutionary claims, but I can tell you for sure that in a modern brain, in a modern brain of a modern human, in our species, we don't seem to be using the language system for any forms of thought that we've tested so far, which is a pretty extensive set, and it seems like you can also damage this system, and reasoning can proceed just fine.
Nick Jikomes 42:57
And you know, one of the things that's really interesting about language is, is the language acquisition piece of this so very young children pick up language spontaneously. They don't need to sit down in class and be instructed explicitly like they do for other subjects. It's a very natural thing to pick up and a very young age. And then layered on top of that, is how difficult adults often find it to learn an additional language, even though they easily learn their first language as a child. What does that tell it like? So why is it that the child brain that isn't fully formed, and the child can't do a lot of things on their own, their cognition is not fully developed? What does that start to tell us about language here that it can be learned so easily in a young brain that is otherwise not as sophisticated as the adult brain.
Ev Fedorenko 43:43
It's a good question. I mean, again, you know, lots of debates around this kind of host of questions. I should say that one thing I always say to my students when I like teach about language and language in the brain is that it's a very bad experiment, in a sense, like it's a very unfair comparison between a child learning a language and an adult learning a language, the experiences are completely different. When you're a child, you have caregivers around you whenever you're awake, and that like you are their main focus of attention for the first like, two years, for sure, right? They're around you. They're trying to interact with you. They find you like the most exciting thing in the world. It's great. It's good times, childhood, early childhood, and then when you're trying to learn a language later, it's a very different experience, one you've already learned a ton about the world, so there's no kind of pressure to, like, find a way to, you know, figure out what the world is like and talk about it. You already know everything about the world, and you're just trying to learn this other set of mappings between forms and meanings that you already know. But of course, the learning mechanisms change. You know, there are certain kind of plasticity that's present early on that we know is not present later on, as kind of knowledge structures come into place, which. Very rapidly during the first few years of life, right? We learn a ton like a five year old is a very sophisticated, you know, mental system. So a lot of these kinds of plasticity windows do become, you know, not, not as accessible at later ages. But why? Exactly it is. I don't think anybody has an answer to that, but, but there is, you know, there's interesting work on people trying to reopen the windows of plasticity and learning using like like psilocybin and whatever. There's interesting work that I think may tell us more about what kinds of learning may be possible even in an adult brain, under different conditions. But I think it's it's a very hard comparison to make between an infant learning their first language and an adult learning a second or third or whatever language.
Nick Jikomes 45:52
I think it's an important point because, you know, it's really easy to say infants or young children learn language easily. I don't necessarily think we can quite say that. I you know, if you ever watch an infant, they often seem to be quite angry and struggling to communicate. It's very intensive. But basically what you're saying is, it's an apples and oranges comparison. And I wonder if so when we say obviously, it takes longer for most adults to learn a language if they try to. But the question is, are they really in an environment where they have to try as hard as a child, right? Is there any data out there where an adult is truly fully immersed and they can't use their native language, and how long does it take to learn?
Ev Fedorenko 46:32
I mean, again, I don't think the right experiments have been done. You want to, like, match everything else and just vary the way you learn. But of course, there's still a lot of things that will vary if you're immersed without any access to your knowledge, you know, so it's, it's very hard to do these control, yeah, yeah, kinds of studies. But you know, from what, like second language teachers will tell you, yeah, immersion is definitely helpful. Like, if there is more of a need for you to survive in a group of people who you have to somehow interact with, of course, you're going to be more motivated and whatever. But you know what exactly is happening that's changing in those circumstances? I don't, I don't. So
Nick Jikomes 47:10
some of the things you've told us so far are that, you know, one there does seem to be the standalone ish language faculty that you can find in the brain that is dedicated largely to language, not other forms of cognition. There is a language network in the brain. There are parts of the of the brain, different parts that are interconnected that seem to be largely dedicated to linguistic processing. And I want you to unpack that in the context of something else you said earlier, which is, you know, if you go back to the ideas of Chomsky around universal grammar, the stuff that you often learn. I remember learning this in psychology and neuroscience classes. This is one of the dominant ideas, or big ideas, that's been there historically, that humans have this innate language faculty, and there's this sort of template in our brains that we all have. How do you think about so on the one hand, you're saying there is a standalone language faculty. There are parts of the brain dedicated to this, but also that universal grammar, that general idea doesn't have a lot of support. Can you unpack that a little bit for me? Yes,
Ev Fedorenko 48:06
of course, yeah, yeah, no, those are great questions. So, so this question ties into a question of, how does a language system come to be, and the way that I've talked about it as kind of a set of four meaning mappings that we can use to encode thoughts into Word sequences and decode meanings from other word sequences already. Should tell you that it can't be, it can't be innate, right? Because we have to learn those mappings. When I am born into the world, I don't know anything about what words may mean. I don't know anything about the world, right? Well, that's debated, but, you know, let's say, for simplicity, let's say we know nothing. And if I'm placed in a Japanese family, I will learn Japanese. And if I'm placed in a you know, family in the Philippines, I learned Tagalog or whatever. And so it's a very deep and interesting question of how these regions come to be language specialized regions. And the way to, of course, ask these questions is through developmental work. So you can ask, you know, if I scan a five year old, do they have a language network? Turns out they do. If you scan a four year old, they do two and two three year old, yeah, it seems to be already there, which, of course, makes sense, because a three year old can pretty, in the pretty sophisticated ways, communicate already using language. Now, the most action in language learning happens between about six months of age, when kids start learning to map words to meanings. The first evidence is around age six months and about one and a half years of age, like that's where kind of the most intensive change happens. Now that's a window during which studying brains of children is really, really challenging. Yeah, we can actually scan infants like PRI you know, in the first couple months of life, they're just mostly sleeping. And you can study auditory processing of enduring sleep, which is what a lot of people do, and then you can scan them using tools like functional MRI from about two and a half, three years away. Two and a half is very challenging. I have a postdoc currently doing this, and it's a lot of it's a lot of very. Very hard work, but like at around, you know, nine months or one, what these regions are doing is a very, very important question, because I think this can tell us something like, what is language made out of? Like, what what are the prereq? What are the circuits that become language circuits? To the extent that we find these regions, say, in a three year old, what are they doing when that kid is one, are they responding to general kind of social stimuli? Are they responding to kind of abstract representations of meanings in the world? Are they responding to some kind of more abstract reasoning challenges, right? And, you know, and we're, you know, trying to see if we can make some headway in those questions by using some combination of longitudinal scanning of like scanning infants and then scanning them later when they have the language regions, as well as scanning during this toddlerhood period to see if language regions respond to other things earlier on and later become more specialized. But we don't have an answer yet. There is some work from Zainab sagan's group at Ohio, showing that at least like by age three or four, the language system seems pretty specialized. So, you know, there is that bit of evidence. But again, what's happening earlier? Nobody knows yet. But yeah, you know, and a lot of like, a question that often follows this line of questioning is like, Well, why does it always end up in the same place? And the answer to that is, right, like, if it's if it's an acquired system, if it develops through experience, why is it always there? And the question, the answer likely lies in connectivity patterns. Like it has to be connected to things, where the information comes in and where it has to go out. You know, speech production.
Nick Jikomes 51:44
I mean, I would imagine too. I mean, just experientially, everyone's, everyone's experience is unique, but actually they're mostly, for most people, quite similar, right? We're all using an auditory mapping, except for the small number of people that are, that are auditorily impaired. And I would imagine the network does shift somewhat in those individuals? Yeah, yeah. Well, so
Ev Fedorenko 52:03
sign language, the high level processing is actually appears to be preserved, which is kind of cool. It seems that that system of four meaning mappings is very abstract, like, it doesn't matter what modality use, of course, the perceptual mechanisms are going to be in a different place, right? It's going to be in the visual cortex, because that's part of the venue used to get the initial information out of the signal
Nick Jikomes 52:22
I see. So, so there could be something here to do with connectivity that is independent of sensory modality, which would speak to why some of these regions you would point to on a brain map here are not in like a primary sensory area. They're in these in between areas that can connect to any sensory modality. That's exactly right. That's exactly right. Yep, what um, you know, I can anticipate your your answer to this, I think, based on what you've told me so far. But you know, if we go back to this question of, to what extent are Language and Thought dissociable, or certain other cognitive functions dependent on language, or, you know, synergizing with language, to some extent, there's this famous idea in neuro linguistics and psychology called the safer Warf hypothesis, which basically says that the language you learn constrains your cognition in very important ways. Um, what do you think about that hypothesis and how it fits with what you've told us so far?
Ev Fedorenko 53:17
Um, yeah, it's a it's a very interesting idea. And I liked that idea because, as the bilingual speaker, I kind of know the intuitions that people build on. But unfortunately, there's just no evidence that supports these kinds of effects. So people have made empirical claims based on certain phenomena that, for example, if you have multiple words for particular part of the color spectrum, you're better at differentiating those colors. Or if you have a word a way to encode certain aspects of motion and particular kinds of events, you may be better at remembering those aspects of events. And most of these studies just don't seem to be robust to kind of scrutiny and replication. In fact, that for almost any empirical claim that has been made in that space, there is now a few other papers that will say, Okay, it's a failure to replicate. It doesn't seem like there's robust evidential support for this that said there is something to there is something very true to the idea that culture, the cultural needs shape the language that the word, the kind of, the kinds of things that we talk about, right? So this idea of, you know, 100 words for snow, there is something to that. Of course, if our environment is such that it's important for us to distinguish between these kinds of things, of course, we'll come up with more words to cover that part of the conceptual space. But it's not, it doesn't have to be cross linguistic and cross cultural. It could be also witnessed within, within a society, right? Like, if you are a podcaster and you need to know a lot about like the you know, auditory recording machinery, right? You'll have all these terms in your. Cavalier that most people may not know, but it's important to you and your work, and so you'll expand that part of your conceptual space, and of course, you'll learn words for how to refer to those things. So some of that, in some sense, has to be true, but this idea that you know a particular language that you speak fundamentally reshapes your perception and cognition. I just don't think there's evidential support for it. I
Nick Jikomes 55:22
see so it sounds like the common denominator here is that people's vocabularies and the types of words they use in their language will reflect the social and cultural needs of the culture they're born into. But that does not translate into opening up the possibility for some more domain general ability that isn't opened up by learning a different language. Exactly. That's
Unknown Speaker 55:42
absolutely right. I
Nick Jikomes 55:43
see the other piece that's interesting here that I want to touch on a little bit more. So we talked about this language network. We talked about some of the anatomy here, and then we also briefly mentioned the lateralization component of this that typically this language network is largely lateralized to one half of the brain rather than the other. Usually it's a left hemisphere, but it can be the right also in some people, why does do we Why would this lateralization happen? Why is it at both sides? Do we have an explanation for why things would be localized to one hemisphere?
Ev Fedorenko 56:13
So I don't think we have an answer to that, which is surprising, because it's such a fundamental fact about language. So there's certainly reasons, computational, metabolic reasons, for why you might want to have a system within a hemisphere, as opposed to straddling both hemispheres. That is, you have, you can have faster information transfer between, say, frontal and temporal lobes within a hemisphere using the within hemispheric tracts, compared to having to go through the interhemispheric connections to the other hemisphere. So for so that's a general kind of argument for why you might have lateralization of function, why it's in the left as opposed to the right. I don't think we have a clear answer to, there's a lot of people working on that. There's stories, right? I mean, say, Oh, well, most people are right hand dominant. And maybe somehow, from the, you know, early use of the right hand for kind of manipulation, tool, manipulation, things like that, that somehow shifts the language. But that's, it's very speculative. Moreover, if you actually look at, remember how we talked about the high level language system being separable from like the low level perceptual speech regions and speech production also like articulation regions. If you look at the lateralization of those earlier functions, they seem to be less lateralized than the higher level language system. So that makes it very hard. Like, they're much more bilaterally present, like your speech cortexes. It does show a small bias, but it's much more bilateral compared to the language system, which is very, very heavily biased to the left. So that makes it really hard to say, like, oh, some early sound perception bias or motor production bias shifted that system. Because if they're not even as strongly lateralized, it just kind of doesn't the pieces don't fall together into a coherent story yet. But it's a very interesting thing to very important thing to solve, because one thing that you see in many brain disorders, including even ones that don't specifically have to do with language, is that the language system manifests more bilaterally. So for example, in individuals with autism and individuals with schizophrenia and individuals with just epilepsy, like any kind of brain disorder you take, you see this more bilateral manifestation. So something is obviously going differently in cases where the brain is not quite working in a typical way that leads to this manifestation. And so it's really, really important to try to understand why does it go to the left in most common, typical individuals, and what's the functional importance of that, and why does it lead to problems if it doesn't happen this way? We're working on this. A lot of other groups are working on this, so maybe there'll be answers in next 20 years. Yeah, so
Nick Jikomes 58:47
we don't know, but, but it does happen reliably, this lateralization, that's very interesting. It presumably is happening. For some reason, it sounds like one idea could be that it has to do with things like computational processing speed, maybe because I have to parse so many things with on a scale of seconds, you know, in terms of making my speech, hearing what you're saying, understanding the context all this, perhaps adding the latency that would come from plugging one side of the brain into the other just makes it much less efficient to do that or something. That's
Ev Fedorenko 59:19
right. But that's, again, just a bias for having a within hemisphere system, as opposed to bilateral, and not for left versus right, right, because, you know, make the same arguments for, you know. So it could have been like maybe in half of the individuals who would end up on the left and then half on the right, right. It's like a random little choice point, but yeah, that always goes to the left. Almost always goes to the left. So something is special about the left hemisphere for
Nick Jikomes 59:43
interesting and so when people learn multiple languages, so so like yourself, if we put you in a brain scanner, to what extent would would each language light up the same language network in exactly the same way, versus be sort of using. At least somewhat different chunks of the brain.
Ev Fedorenko 1:00:01
It's the same system. We've done this now with individuals who speak multiple languages, up to, like, you know, 50 languages. We haven't tested them in all their 50 languages, but we've tested them in a bunch, and they all go to the same system. Now, of course, there has to be some differentiation at a finer scale, but it's not the case like, you don't find kind of course, differences like, let's say, you know, your French region is the left frontal one, and your English one is the left all the whole system gets engaged in the processing of all these different languages. You can use something like fine grain activation patterns to reliably discriminate which language you're listening to. So I can take your brain when you listen to a bunch of English sentences and say French sentences if you're English, French bilingual, and I can reliably classify which language you're listening to from this fine grain patterns of activation. Other ways in which there are differences is in how strongly that system is working when you're processing these languages. And there it's basically two big factors that matter. One is how well you can understand the language. So the more you can understand, the higher the response, which, again, an example I often use when I teach language acquisition is there's this old cartoon from the New Yorker about a man talking to a doggy ginger, and he gives her all these like, sophisticated sentences, like, ginger, you've done this thing again. I told you so many times, blah, you know, ginger. Why are you doing this? And then in the dog's thought bubble, it's like, you know, ginger, blah, blah, blah, blah, blah, blah, blah, Ginger, right? So the only thing that the dog can get from that is that single word, which is her name. And similarly, when you're learning a new language, and I give you information that language, you may be able to kind of grasp little bits and pieces of language, but you can't quite put it together into a structured, complex representation. And so basically, the level of response in the language areas scale such that the more you can understand the language, the stronger the response, except for your native language, which is a little bit lower than you would expect based on the proficiency alone. So if I take like for me, right? Well, I'm not a good example. I'm actually, I don't show this effect. But for somebody who is similarly proficient in two languages, and one language is their native language. One is non native. You sometimes see a little bit of a low response to the native language, presumably because it's a little bit less effortful. I just don't I'm like, a heritage speaker, what they call like, I don't use Russian pretty much at all anymore. And so for me, I actually we see similar levels of response to English and
Nick Jikomes 1:02:39
Russian. So one of the things I want to ask you about too, it relates to this idea that language is primarily about communication. You know, when you compare, when you compare, like the word or symbolic systems for language compared to other symbolic systems that people construct, like like math and algebra and stuff like that. One of the differences that jumps out is, you know, when we do mathematics, the symbols are extremely, extremely precise. That's sort of the whole point of when we do mathematics, we're being extremely, extremely precise with these abstract symbols. But with language, there's lots of redundancy and ambiguity. The same word can often have different meanings when it's used in different contexts. Can you talk a little bit about the information processing there in the sense of whether that whether, whether that sort of slop that's part of language is actually a feature rather than a bug when it comes to communication,
Ev Fedorenko 1:03:29
that's a very interesting hypothesis. I have not quite heard this hypothesis articulated in recent years for the key difference between language, sorry, between like natural languages and systems like math. I mean, I don't know whether, okay, so one way that you could test this is, I guess there is a lot of meanings in language, a lot of words that do convey very precise meanings, like, it's not like language isn't capable of doing this. It's true. There's still probably more ambiguity. But so you could imagine looking at the responses to words that convey precise meanings versus not. It's all going to happen within the language system, but we do see, we do have some evidence that the language system cares about more abstract meanings, but that's not quite the same dimension. I mean, so we in a recent line of work, we actually wonder, like, what are the features of language that make it do different things than like a system like math or programming languages, both of which load on on a different system in the brain? And we examined the case of constructed languages like Esperanto and Klingon, which shares some features with like programming languages. On the one hand, they're like artificially constructed. They're very evolutionary, recent. They're typically made by a single person, right? Like designing that system. And we looked at whether constructed languages would. Pattern with kind of programming languages or with natural languages, and they pattern with natural languages, and we made an argument that this is because of the kinds of meanings that are conveyed, but not along the dimension you mentioned, which is like the level of precision of what meanings they pick out, but more in terms of the fact that the natural languages typically convey meanings about the kind of the inner and outer worlds, right? We refer to objects, to, you know, entities, to actions, whereas math and programming languages convey a large range of meanings, but it's all kind of relational meanings. Like it's kind of like an internally very restricted system. It's very abstract kind of relations that those systems, formal systems, can express. And so that's kind of the story that we told. But it may very well be the case that kind of the precision with which these forms pick out their meanings plays into this as well. So it's a there should be ways to empirically distinguish those possibilities, including perhaps with like constructed languages that vary on that dimension. That's a very interesting idea. Yeah.
Nick Jikomes 1:06:04
And I guess, you know, another, another sort of related piece there that I was thinking of was that, is there some tie in here with, like, learnability? So, so the fact that, you know, obviously, you know, when you talk to a linguist, they'll often say things like, well, they'll talk about things like recursion, and you can make a sentence that's arbitrary. Arbitrarily long and whatever. But of course, obviously we don't learn an infinite list of words and meanings and things. It's ultimately finite. Is there something here to do with learnability, whereby the fact that there's lots of redundancy, a fair amount of redundancy in language, that there's different words can mean different things, different contexts? Does that actually help be more learnable faster than otherwise, if you had to add words or something? Yeah,
Ev Fedorenko 1:06:46
very likely. So there is evidence suggesting that it actually is very helpful. So reusing form. So in fact, Chomsky and a few others have made this argument. Well, if language was for communication, it wouldn't have ambiguity in it, and you can mathematically show that it's actually really useful for a system to have ambiguity, because it allows you to reuse forms that are, you know, short and easy or whatever have other properties that are useful for other things, like learning or ease of production and something like that. So it's actually and because exactly for the reason you say that context will typically disambiguate. It's not like we're confused all the time when we use language. Of course, occasionally that may happen, but it's very uncommon, actually. And so reuse of forms to mean different things in different contexts is actually quite useful for an efficient communication system. And
Nick Jikomes 1:07:32
I want to go back to this distinction between language and thought and ask and actually like come at this from something that you mentioned briefly, and I didn't plan for this, but I thought it was really interesting when interesting when you brought it up, I brought up schizophrenia, and you mentioned that you thought that was a disruption, not of the language faculty per se, but of thought at a different level. Can you unpack that a little bit more for
Ev Fedorenko 1:07:55
me? Yeah. So I mean, schizophrenia is, you know, construed by many as a thought disorder. So, and it's a very interesting condition. I mean, it's a very, you know, devastating condition, but it's very interesting in that it manifests, it, it. It has similar manifestations across very different cultures. Like, there are certain themes in the kinds of thoughts that people get which take slightly different force, but they all have many shared features, like, you know, somebody controlling you, right? Like you losing, not losing, being, this ability to be an autonomous agent, or these kind of conspiracy theories, right, where somebody has orchestrated a bunch of very sophisticated kind of, you know, causal chains that lead to certain outcomes, and so on and so forth. And there's a lot of work on studying linguistic patterns and schizophrenia, because it's, again, that's a cool window into the thought process, right? And people have reported things like, you know, there is less continuity. Like, if you have an individual of schizophrenia generate language, there's often like these, like weird associative jumps from kind of like one word may trigger something that's like a really weird Association, as opposed to kind of relating to different deeper underlying structures, or something like that, where you would have much more continuity. But again, if you look at their the form of their language, they don't have a language problem, like they produce perfectly grammatical sentences. They use words in the right context, but they make these weird leaps sometimes in how they reason, how they put pieces of information together into kind of larger causal structures. But their language in its kind of surface seems to be totally fine.
Nick Jikomes 1:09:38
And you know, one of the things that you've shared with us is that, you know, obviously, language is a very important aspect of humans, and our humanness, obviously it's an important way we're largely distinguished from other species. But what's, what's sort of implied by some of the things you've said, is that it's, it's not the it's not like languages. The only thing that's making us as successful as widespread. As we are as a species, we have other cognitive abilities that are also either absent from other species or just, you know, quantitatively, better, more developed. What are, I mean, to what extent do you study those other aspects of cognition? And what can you tell us there? Yeah, we've
Ev Fedorenko 1:10:14
done, we've done some work on this, and are actually actively working on it now. I mean, so there's a bunch of systems that have been implicated in what you may call high level reasoning, the three systems, well, the two systems that I've spent some work on are, one is a system of kind of formal abstract reasoning. So that's a system on which mathematical reasoning and understanding computer languages loads on. But it's also a system that is engaged in, you know, what people call basically goal directed action, like, when you're focused on a problem at hand, you're planning your next steps. Like, that's a system that allows you to do that kind of thinking and planning. And then there's a very separate system which supports social reasoning. So that's a system that researchers like Rebecca Saxe at MIT has studied a lot. So this is a system that we use to understand how, how others form and rely on their beliefs in their behavior. So it's, it's a system that allows us to what's called, you know, mentalize, right? Think about, you know, what is Nick thinking about this thing, and how does that affect, you know, his behavior, for example. And that system is very different from the system that supports, like, mathematical reasoning. Again, as you know, maybe very obvious, if you think of maybe some friends you have who are, like, really nerdy, smart kids, but sometimes not so good socially, or people who are socially really good at, like, you know, reasoning about others minds, but may not be so good at formal, kind of abstract thinking anyway, and so, so those are two systems, and both are totally distinct from the language network. And then there's a third system that is less well understood, and there's more debate about kind of what exactly it's doing, but it's linked to what people refer to as episodic projection. So it's kind of imagining yourself in the future or remembering yourself in the past, right? So it's putting yourself somewhere on a extended time trajectory. That's also some also have linked that system to spatial cognition in some ways, but again, that's also separate from the social and the formal reasoning system. And so those are three things that exist in animals to some degree. It's not like any of them are totally de novo in humans, but one thing that has changed in humans relative to other animals is this massive expansion of the so called Association cortex, which is parts of our brain that don't do perception and motor control, so that basically do cognition. And all of these systems live in this association cortex. So given this massive expansion of those chunks of the brain, all of these systems increased in size, and presumably, you know, strength and connectivity and all those other measures that you may look at in cross species comparisons. There's other parts of the association cortex that are not parts of those systems that we still don't know what they do. And like I said, that's an active area of research that's ongoing, including in some groups, here, in my group, and in other groups and in many other labs across the world. It's kind of like trying to understand the structure of thought, right, which kinds of mental processes are distinct and which ones are supported by the same machinery.
Nick Jikomes 1:13:25
And so just taking one example there, you talked about social reasoning being something that's separate from, say, mathematical ability or or the language faculty when you're studying this and determining those things. How do you so specifically for the social reasoning one that one's a little hard for me to think about, how do you actually do this? How do you actually dissociate it from language? How do you test someone's social reasoning abilities without talking to them? Yeah, so that's
Ev Fedorenko 1:13:51
a good question. Sami, so you can't use linguistic stimuli. You just have to match them for linguistic properties. So in fact, a very common contrast to identify these regions is a contrast between short vignettes that involve other people's mental states and vignettes that are similar, linguistically in terms of like their linguistic complexity, but talk about kind of a similar mental structure, but don't involve mental states. So I'll give you an example just to concretize it. So, for example, you know, Nick was, you know, came home and put his car in the garage, but later his brother came home and he moved the car to the driveway right. And then you may ask question like, Where will Nick look for the car in the morning? Or you can have a very similar scenario, like, for example, you know, I took a picture of my house last year. Since then we added a second floor, you know, or whatever, right? Or like, an extension, or something like this. And then you may ask a question, like, does the what does the picture look like, right? You know? And so in both cases, you're contrasting two states of reality, right? Except in one case, it's somebody's mental state and reality. In another case, it's a physical represent. Presentation of outdated
Nick Jikomes 1:15:00
reality. And you can match that for linguistic complexity, of the description for linguistic
Ev Fedorenko 1:15:04
complex, you can match it for cognitive complexity for like, how hard it is to do this task. And that contrast very robustly activates this network of regions that seem to support social reasoning. And then you can also find it with non, non linguistic stimuli. So it's those systems are very, abstract in that, however, that information is conveyed, which can be in simple patterns of like geometric shapes moving around in ways that suggest complex social interactions. And these regions will care deeply about that kind of information, even if it's coming, you know, in very, very kind of visually simple representations.
Nick Jikomes 1:15:45
When we talk about, obviously, a hot topic today, in the very recent past, is language as it's handled by machine intelligence and AI systems. People are using chatgpt in these large language models. I'm not a technical expert on these things, but conceptually, you know, I know how they work. I simply want to ask you, as like a neuro linguist, to what extent are these systems actually using and comprehending language in the way that a human would, versus using, you know, being trained on text and using it in impressive ways, but, but nonetheless, without the actual sort of semantic knowledge that a human mind has,
Ev Fedorenko 1:16:26
yeah, yeah. I mean, I think a lot of these questions will bottom out and like, how broadly you define language, if you define language to include the systems of thought that often work together with language during kind of real world complex scenarios, then, yeah, I don't think these models have those abilities. But you can ask this question. I mean, so you can we have a conceptual piece on this question where we basically just try to say, in thinking about these models and their abilities, it's very helpful to separate the formal ability of knowing how to use language appropriately, how to build grammatical structures, how to use words in appropriate context. And reasoning abilities, which you know you can use linguistic stimuli to probe, but the just the mastery of the forms of language does not entail that you should be able to do the reasoning parts right. You can just learn the linguistic patterns, and you kind of, you know, even in kind of generative ways. So, like, I think the extreme version of a story where these models just mimic whatever they've seen in their training. I don't think that's right. They generate new linguistic structures. They use language creatively in all sorts of ways that, you know, I think would have been hard to predict a priori, but that all said, I think their thinking capacities are pretty limited. Still, they've gotten much further than I would have predicted, and that tells you that there's a lot of structure that's packed in linguistic patterns that you kind of use to, you know, kind of do some form of reasoning. But I think at least at present, it's still very easy to break these models on these reasoning tasks. So you know, you may see them succeed on a whole bunch of things, but then you make some small perturbations to the wording of the problem, and they fail in ways in which humans would never fail, like we're very robust to kind of the formulation of a problem, as long as we learned how to solve a particular class of problems, we'll be able to do no matter what form or shape it comes in. But these models seem to not be able to generalize in quite the same ways. And there's a host of interesting questions of kind of how far language can take, if you just keep scaling with more data and more compute power, how far that can take you in being able to at least behave in ways that look like reasoning, even if it's done not in the same ways as humans. But there's also interesting questions of you know, are you fundamentally lacking some pieces of cognitive machinery to get you to look more human like in the way that you learn and process and use information.
Nick Jikomes 1:19:03
What? Where are you guys at in the lab today, in terms of some of the big questions that you're trying to address empirically in the lab, about the neural basis of language?
Ev Fedorenko 1:19:13
Well, there's a whole bunch of things. So one we've touched on some of them. So one big question is development, like, how the language system comes about? Like, what is it made of before it becomes a specialized language system? One big bucket of questions is, what is the structure of thought, which kinds of pieces of mental computations are dissociable in the mind and brain a whole other system is to try to understand, how is it that the language system does what it does? How are we so efficient at using language like what is to the extent that you know, as we said, it's extended set of brain areas, and we and others don't seem to see this structure at the regional. Level, it's not like, Okay, this region stores words, and this region, like, does syntactic computations, or something like that, like the whole system, every region shows both of these kinds of processes. But how is how exactly is it implemented? What are the kind of organizing axes of the representations of language that allows us to use language in these kinds of ways. And then there's a lot of questions about how language interfaces with systems of reasoning. Because, of course, in many real life complexity, situate complex situations. We have to we use language beyond ways that just you know deals with constructing grammatical sentences, right? We tell each other about other humans we know. So we have to update our social representations. We teach each other about the world. We build social relationships we know. Construct legal contracts. I mean, there's all sorts of ways in which language interfaces with almost every aspect of our life, and there is a host of questions of how the information has to change in its representations, to go from whatever format it exists in the language system, when you extract the meaning, to ways in which it is used by these downstream reasoning systems to do other stuff with it, like update world knowledge, or, you know, draw on some other information that we've acquired before to produce a response, or something like that. And so there's a lot of ways in which we're trying to tackle these questions, using both neuroscience tools and language models. Yeah, and yeah. We also have a cool question on a cool, cool project on people with very different brains. We didn't talk too much about anatomy, but there's also people who have very different brains from you and me, and their cognition is totally fine. In fact, a lot of these people kind of find out by accident that, like, they're missing almost half of a hemisphere due to, like, a massive cyst, or like a severe case of hydrocephalus that you know just went undetected. How is it that we can implement a typical like mind in this very different set of structural constraints, and which kinds of things, including things like lateralization or specialization of function, which of these things are preserved in spite of this very different physical instantiation, and which are kind of soft or viable constraints on how To build a mind? So there's, there's a lot going on. We're kind of covering a lot of ground. And, yeah, yeah,
Nick Jikomes 1:22:26
interesting, yeah. I mean, one of the things, one of the themes here that I'm getting from you, is there's good evidence that that the language faculty is not all of our cognitive faculties. There are other non language ways that we engage in cognition, and that actually, you know, once you sort of see that or accept that, that makes sense of a lot of anecdotes, you often hear people talk about, like, some people will say, you know, well, my internal monolog says this, this and this, and it's going on all the time. And someone else will say, What are you talking about? The voice in your head? And like, well, I don't have a voice in my head. I don't think it words. And they're like, What are you talking about? Like, it's, it's, we probably all have sort of a different suite of a cognitive tools that are happening. That's very true what you know, you know, we've covered a lot of ground today, but just sort of like to summarize and wrap up here over your career in I'll just call it neuro linguistics. So the empirical study of language and and how it works in the brain, you know, staying squarely within the realm of, you know, empirically grounded areas of study in this realm, not necessarily the philosophical stuff that disagreements people get in. What's been like, one of the things that you say, you would say, like, over the course of your career, we've really like, learned or had insight on that was unresolved before. What are some of the big questions that we've made progress on that were unanswered? I don't know, 510, 15 years ago?
Ev Fedorenko 1:23:52
Yeah. I mean, there's a few. I mean, I think the Language and Thought distinction is a big one, and it's quite interesting sociologically, because the patient evidence has existed before, but somehow, without the complementary brain imaging evidence, people were unwilling to take this evidence and draw the strong conclusion that I think the data actually suggests. And you would always kind of people would say, well, maybe there's some small bit of the language system that's left untouched by the stroke, and you're using that, you know, and you're still able to think because of that. But looking at brain imaging evidence where, like, Okay, here's the language system. We know that it does language. We can show that it does language in all these different ways. And then that system, just like, doesn't show any activity when you're doing math or whatever, somehow those two pieces together, I think, brought it into a story that became more compelling, or somehow was more compelling for more people. And then I think showing that this separation between words and syntax is not on the right track, like a lot of people have made proposals about different regions within the language system, doing different parts of language, and that just empirically ends up not. Not finding much support, and I think that has helped clarify quite a few, quite a few things. And, yeah, and I mean, I mean, I think we didn't talk much about it, but our group was also among the first to show that representations that you get from large language models like chatgpt Share representations with that we can obtain from neural recordings of people processing language. So this is this emergent field of neuro AI, where you, instead of just kind of saying, oh, language models do language, and humans do language, and there are some similarities, we can actually directly take the internal model activation weights and try to use them to compare them to the kinds of structures that we see in the representations of language in humans, and we find some similarities, similarities, yeah, and that's really cool, because that opens doors to all sorts of new things. So that's another big part of our unfolding Research Program, is trying to leverage these models as kind of precise computational hypotheses how some pieces of language may be instantiated, and we're pushing on that front pretty strongly. So yeah, lots of exciting things to do yet.
Nick Jikomes 1:26:15
And one final question I'll ask you. This is not a question I ask often, but I don't know for me, I love this is sort of like an area I've always found fascinating, even though it's not an area of expertise. For me, I think a lot of people are interested in language, and you know how it works, where it comes from, from many different angles. What are one or two books that you think are really a good starting point for someone to read if they want to get a reasonably up to date view on what we understand around language that isn't just going back to some of the ideas that have been around for decades. That you good question.
Ev Fedorenko 1:26:49
So my collaborator and husband, Ted Gibson, just has a book coming out in MIT Press called the cognitive basis of syntax, or syntax the cognitive approach, one of those titles, and that's a very nice summary of a lot of the like cognitive behavioral work. And he and I are actually maybe starting to work on the language in the brain book, which deals with the language kind of thought relationship more. But you know, a lot of like in terms of books written by serious linguists, I would say, you know, Ray jackendoff's books are highly recommended with a lot of questions related to language, in this relationship to cognition, and Adele Goldberg, who is a linguist at Princeton, who also has written some very compelling Books, yeah, I think those would be good starting points.
Nick Jikomes 1:27:43
Excellent. Well, thank you again for your time. I really appreciate it. This is a fascinating area. There's a lot that we covered and jumped around on and a lot we didn't even get to but this is a big, rich area of research. Any final thoughts you want to leave people with? Anything you want to reiterate before we sign off? I don't know.
Ev Fedorenko 1:28:04
We're always, I'm always trying to inspire young people to come and join the academic field. I think there's a lot that happens when young, fresh, young minds with fresh ideas join the field. And there's a lot of complaints about academia and all sorts of infrastructural problems, but it's also such an incredibly privileged life that we lead where we can think for a living and try to figure out how the mind works. And it still remains one of the kind of biggest mysteries. And so for those even kind of young listeners who are interested in these questions, I would certainly not be dissuaded by the current criticisms of academia. I think there's a lot of positive still and a lot of people who really are passionate about finding the answers to how our minds work, which I certainly have had a lot of fun doing, and will probably continue to do for as long as I'm around.
Nick Jikomes 1:28:59
Alright, well, Professor Fedorenko, thank you very much for your time.
Ev Fedorenko 1:29:02
My pleasure, of course, thanks.
Share this post