#426 – Edward Gibson: Human Language, Psycholinguistics, Syntax, Grammar & LLMs
03:00:32 Share

The following is conversation with Edward gibson, or ted, as everybody calls him. He is a cycle linguistics professor. He heads the M. I, T.

Language lab that investigates why human languages looked the way they do, the relationship between cultural language and how people represent, process and learn language. Also, he should have a book titled syntax, a cognitive approach, publish by MIT press, coming out this fall. So look out for that.

And now a quick few second mention of each sponsor. Check them out in the description is the best way to support this podcast. We got yahoo finance for basically everything you've ever needed if you're investor listening for listening to research papers, policy genius for insurance, sharp fy for sun stuff online and eight sleep for nps.

SHE was wise in my friends. Also, if you want to work with our amazing team or just get in touch with me, get A X that can flash contact. And now onto the full ad reads, as always, no ads in the middle.

I try to make this interesting, but if you must skip friends, please to check out the sponsors. I enjoy their stuff. Maybe you will too.

This episode brought to you by yahoo finance, a new sponsor, and they got a new website. They should check out it's a website that provides financial management reports, information and newsroom investors. Yahoo itself has been around forever.

Yahoo finance has been around forever. I don't know how long, but I must be over twenty years. IT survived so much.

IT evolved rapidly, quickly, adJusting, evolving, improving all of that. The thing I use IT for now is there's a portfolio that you can add your account to. Ever since I had zero money, I used, boy, this is called T D M, A trait.

I still use that same thing, just getting a basic metro und, and I think T D metric got about. But Charles, swap or quired emerge. I don't know.

I don't know how these things work. All I know is that your finance can integrate that, and you show me everything I need to know about my corn or portfolio. I don't have anything interesting going on, but IT is still good to kind of monitor IT to stay in touch.

Now a lot of people I know have a lot more interesting stuff going on, investment wise. So all of that could be easily integrated in the hoo finance. And you can look at all that stuff, the charts ababa IT looked beautiful and sexy just helps.

You will be informed. Now that's about your own portfolio. But and also for the entirety are the finance information for the entirety of the world.

That's all there. The big news, the analysis of everything is going on, everything like that. And I should also mention that I would like to do more and more financial episodes.

I've done a couple of conversations with radi. A A lot of that is about finance, but some of that is about geopolitics and the bigger context of finance. Uh, I just recently did a conversation, bill very much about finance, and I did a serious of conversations on court to currency, lots of lots of billion people.

Michael saw someone, Charles and vitality, just lots of billion people, that bed thinking about the future of money, future finance. Anyway, you can keep track of all that with yahoo finance for comprehensive financial news and analysis. Got a ahoo finance that come the yahoo financed a com.

This episode. He has also brought you by listening an APP that allows you to listen to academic papers. It's the thing i've always wished existed, and I always come of.

Suspected is very dictate to pull off. But these guys pulled IT off. Basically, it's any kind of format attack brought to life through audio. Now for me, the thing I care about most, and I think that's the foundation of listening, is academic papers. So I love to read academic papers, and there's several levels of rigor in the actual reading process.

But listening to them, especially after I skim IT or after the a deep dive listening to them is just such a beautiful experience, is solidify the understanding IT brings to life all kinds of thoughts. And i'm doing this while, uh, cooking well, i'm running. I'm going to grab a coffee.

All the kind of stuff IT does require an elevated level of focus, especially the kind of papers I listen to, which are computer science papers. But you can load in all kinds of stuff. You can do philosopher aper.

You can do psychology papers like this very topic of linguistics, have listen to a few papers on linguistics, and when back the charms can listen to papers is great. Papers, books, P, D, S, web articles on the cost of even email, news letters in the voices that a pretty sexy is great, is pleasant to listen to. I think that's what's ultimately most important, is IT shouldn't feel like a chore to listen to IT like I really enjoy IT.

Normally you'd get a two week free trial, but listeners of this pockets get one month free. Got a listening that com slash legs that's listening that com slash legs. This episode brought you by policy genius, a marketplace for insurance, life, auto, home, simply all kinds of insurance. There's a really nice tools for comparison on big fan of nice tools for comparison. Like I have to travel to harsh conditions soon and I to figure out how I need to update my equipment to make sure is weather proof, waterproof even is just resilient to harsh conditions.

And it'll be nice to have the comparisons have to resort to like red post, foreign post, kind of debating different audio quarters and h cabling and microphones and water proof containers, all that kind of stuff I would love to be able to do al, like a rigorous comparison of them. Of course, it's going to uh, amazon get the reviews and IT doesn't actually really, really solid. I saw I think amazon has been the giant gift to society in that way that you cannot can lay out all the different options and get a lot of structured analysis of how good this thing is.

So amazon s has been greater that now what policy genes did is the amazon thing, but for insurance. So the tools, for comparison is, is really my favorite thing, is just really easy to understand the fool marketplace of insurance. With this genius.

You can find a life insurance policies that start at just two hundred and ninety two dollars per year for one million dollars of coverage. Had the policy genius that conflict looks, or click the link in the description to get your free life insurance quotes and see how much you can save. That's passions, ious, our conscious legs.

This episode also brought you by sharp fy, a platform designed for anyone to sell anywhere with a great looking online store. I'm not named dropping here, but I recently went on a hike with the CEO shop, but he's brilliant. I mean, a fan has for a long time long before uh, shop fy was a sponsor.

I don't even know if he knows that sharp fy sponsors this pocket. Now just to clarify, IT really doesn't matter. Nobody in this world can put pressure on me to have a sponsor, not to have a sponsor, or for sponsor to put pressure.

I mean, what I can, I can say, I want to wake up in the morning, feel completely free to say what I, anna say, and to think what I, anna think. I've been very fortunate that way in many dimensions of my life. And I also have always lived a fruit life in the life of discipline, which is where the freedom of speech in the freedom thought truly comes from.

So I don't need anybody. I don't need a boss. I don't need money. I'm free to exist this world in the way I right now. On top of that, of course, i'm surrounded by incredible people.

Many of who I disagree with have arguments, so i'm influenced by those conversations in those arguments. And i'm always learning, always chAllenging myself, always humbling myself. I have kind of intellectual humility.

I kind of suspect i'm kind of an idiot. I start my approach to the world of ideas from that place, assuming a immediate, and everybody has a lesson to teach me. Anyway, not sure why I got on off the attention a but the high, good, beautiful nature friends is beautiful.

Anyway, I have a sharp fied store like freeman to cons store is very minimal, which is how I like. I think most things, if you wanted set up a store that super easy, takes a few minutes. Even I figured out how to do IT sign up for a one dollar month trial period.

Sharp fied, that consulate? Lex, that's all lower case. Got a sharpie. Ed, that consulate? H likes to take your business to the next level today. This episode is also brought to buy asleep and spare three cover.

The source of my escape, the door, when opened, allows me to travel away from the troubles of the world into this a serial universe of comes a cold bed surface with a warm blankets, a perfect twenty minute nap. And IT doesn't matter how a dark to place my mind is, in a nap will pull me out. And I see the video of the world again.

Technologic is speaking basely bushes really cool. You can a control temperature, the APP is become such an intro part of my life that i've begun to take a for granted, typical human, so that they have controls the temperature. I said that currently i'm setting into A A negative five and is just super nice, cool surface.

It's something I really look forward to, especially i'm traveling. I don't have one of those IT really makes me feel like home check IT out and get special savings when you got to asleep, that comes lash legs. This is eleven in pocket. To support IT, please check on our sponsors in the description. And now your friends, here's Edward garson.

When did you first become fascinated with human language?

As a kid in school, when we had to structure sentences in english grammar, I, I, I found that process interesting. I found IT confusing what I was told to do. I didn't didn't understand what the theory was behind IT, but I found IT very interesting. So when you look a grammar.

you most think about like a puzzle, like most like a mathematical puzzle.

Yeah, I think that's right. I didn't know I was going to work on this at all at that point. I was really just, I was kind of a math geek person, computer scientist, I really like to your size.

And then I found language as A A neat puzzle to work on from an engineering perspective, actually. So as I sort of accidentally, I decided after I finished my undergraduate degree, which was computer science and math and canada and queens university, I decided to go to great school as I. That's what I always thought I would do.

And I went to cambridge, where they had a masters in masters programme in computational linguistic. And I hadn't taken a single language class before. All I taken with C S. Future science mass class is pretty much mostly as an undergrad. And I just thought this was an interesting thing to do for a year, because there was a single year program, and then I end up spending my whole life doing IT.

So fundamentally, your journey through life was one of a mathematician and computer scientists. And then you can discovered the puzzle, the problem of language, and approached IT from that angle to try to understand that. From that angle, almost like a mathematician, or maybe even an engineer .

is an engineer. I say, I mean, to be Frank, I had taken an A I class, I guess he was eighty three, eight, forty five, somewhere, eighty four in there a long time ago. And there was a natural language section in there.

And IT didn't impress me. I thought there must be worried resting things we can do IT didn't seem very IT seem just a bunch of, uh, hacks to me. IT didn't seem like a real theory of things in any way. And so I did. This was this seemed like an interesting area where there wasn't enough good work.

The java come across like the the philosophy angle of logic. So if you think about the eighties, they are the expert systems where you try to maybe side step the poetry of languages at the of the cent, acts in the gram and all that kind of, and go to the underline, meaning the language is trying to communicate and trying to somehow compress that in computer representable way. Do you ever come across that in your studies?

I mean, I probably did, but I wasn't as interested in. I was I was trying to do the easier problems first. And once I could thought maybe we're handle which is seems like the syntax is easier, like which is just the forms as opposed to the meaning like you're talking you started talking about the meaning.

That's very hard problem and IT still a really, really hot problem. But the forms is easier. And so I thought at least figuring out the forms of human language in which sounds really hard, but is actually may be more attractive.

so is interesting using there is a big divide. There's a gap. There's a distance between form and meaning exact the question you have discussed a lot with our lives because they're damn good at form yeah I think if we are good at .

this form exactly that's that's why they're good because they can do form meaning .

are do you think there's wow and that means to open question right? How close form and meaning are will discuss a bit. Add to me studying form, maybe the romantic notion gives you form is like the shadow of the the bigger meaning thing on the line.

Language, as I IT formers, language is how we communicate ideas. We communicate with each other as language. So in understanding the structure of that communication, I think you started to understand the structure of thought in the structure of meaning behind those thoughts and communication to me. But do you big gap?

yeah.

What do you find most beautiful about human language? Maybe the form of human language, the expression of human language.

What I find beautiful about human language is the um some of the generalizations that um happened across the human languages, within and across no language. So let me give you an example or something which I find kind of remarkable that is if like a language, if IT has um a word order such that the verbs tend to come before their objects.

And so that's like english does that so we have the the first the subject comes first in a in a simple certain so I say um you know the the dog chased the cat or mary kicked the ball. So the subject first and then after the subject, there's the verb and then we have objects. All these things come after an english, so it's it's generally a verb and most of the stuff that we want to say comes out for the subject comes it's the it's the objects is a lot of things we want to say to come after.

And there's a lot of languages like that, about forty percent of the languages the world like that there um suh subject verb, object languages. And then um these languages tend to have um prepositions, these markers on the nouns that that connect ounce nouns or nounce ed to verb. So I I so a verbally I saw proposition like in or on or of or about, I say I talk about something, that something is the object of that proposition that we have.

These little markers come also, just like verbs. They come before their their nouns. okay. And then so now we look at all the languages that like japanese or or hindi or some, these are so called verb final languages.

Those is about maybe little more than forty percent, maybe forty five percent of the world's languages, or more in fifty percent the worlds languages for final. Those tend to be um post positions, those markers the same. We have the states have the same kinds of markers as we do in english, but they put him after so uh h sorry, they put him a first and markers come first.

So you say instead of um you talk about a book, you say A A book about the opposite order there in japanese or in hindi, you do the opposite and and the talk comes at the end, so the verbal come at the end as well. So instead of um mary kicked the ball, it's mary a ball kicked and then it's mary kicked the ball too. John, it's john two the two little, the marker there, the proposition cut, is a post position in these languages.

And so the interesting thing, a fascinating thing to me, is that within a language, this order alliance is harmonic. And so if it's one or the other, if either verb initial or very final, but then you then you will have prepositions, prepositions or post positions. And so and that's across the languages that we can look at.

We ve got around a thousand languages, this around seven thousand languages around on on the earth right now. But we have information about a word order on around a thousand of those pretty decent amount of information. And for those thousand, what we know about um about ninety five percent fit that pattern. So they will have either urban is about it's about half and half half of urban initial like england and half of their final like um like japanese.

So just to clarify, verbal initial subject verb object that's rect final is still subject object that's correct.

Yeah the subject is generally .

first as so fascinating. I ate an apple or I apple eight yes. okay. In this first thing that there's a pretty even division in the world among those forty five percent yeah.

it's tty pretty. And and those two are the most common by far those two. Water is the subject tends to be first. There are so many interesting things, but these things are the thing I find so fascinating. Is there these generalizations within and across the language and anonyme, those are there isn't actually a simple explanation, I think, for a lot of that. And that is um you you're trying to like minimize dependencies between words.

That's basically the story, I think, behind a lot of why word order looks the way IT is, is you we're always connecting what is what is the thing i'm telling you? I'm i'm talking to you and sentences. You're talking to me and sentences.

These are sequences of words which are connected, and the connections are dependencies between the words and and IT. Turns out that what we're trying to do in a language is actually minimize those dependent ceilings is easier for me to say things. If the words that are connecting for the meaning are close together, it's easier for you in understanding if that's also true.

If they're far away, it's it's hard is to produce produce. It's hard for you understand and the languages of the world within the language and across the languages, you fit that generalization, which is you so know that turns out that having verbs initial and then having prepositions ends up making dependency shorter, and and having verbs final and having post positions ends up making dependently shorter than if you cross them. If you cross them, IT ends up, you just end up.

It's possible you can do IT within a language. Within a language you can do IT IT just ends up with longer dependencies then if you didn't. So languages tend to go that way. They tend to minimally, they say they call hermon's. So was observed the long time ago by uh without the excelsior by a guy called Josephin work because um famous typologies from stanford, he observed a lot of generalizations about how world order works and these are some of the harmonic generalizations that .

he observed harmonic journalizing about world world. There's so many things I want to ask you. Me just sometimes basic use matching dependencies of times. What you mean by dependencies?

Well what I mean is in um in language there's kind of three structures to three component the structure language one is sounds so cats at into in english I am not talking about that part. I am talking and there's two meaning parts and those are the words and and you're talking about meaning earlier or so. Words have a form and they have a meaning associated with them.

And so cat is a full form in english in that has a meaning associated with whatever cat and then the combinations of words uh, that we will all grammar syntax and uh that's like when I have accommodation like the cat or two cats, okay, so uh where I could take a two different words that put them together and I get a compositional mean from putting those two different words together and and so that's the syntax and in any sentence or alterations, whatever i'm talking to you, you're talking to me. We have a bunch of words and we're putting together in the sequence, the IT turns out they are connected so that every word is connected to just one other word. In that in that sentence.

And so you end up with called technically a tree, it's a tree structure. So where there's a root of that of that alterations of that sentence and then there's a bunch of dependence like branches from that route that go down to the words. The words are the leaves in this metaphor for a tree.

So trees also a mathematical .

construct.

uh, so in in the this fascine that you can break down a sense into a tree. And they want every word is hanging on to this.

depending on, and everyone agrees on. And so all link .

will agree with that. No one.

no one.

nobody sitting OK is sitting.

I think in every language, I think everyone agrees that all sentences are trees at some level. Can I place on that?

It's to me just as a laman is, uh, is surprising that you can break down sentences in many, most all languages into a tree.

I've never heard anyone disagree with that, as the details of the trees are what people disagree about.

Wh, okay, so what? Uh, what's that the root of which you how do you construct? How hard is IT? What is the process of constructing a tree .

from a sentence? Um well, this where you know, depending on what there's the radical notions i'm going to say this simplest thing, dependency grammar is like a bunch people invented this air was the first french guy back and I mean, the paper was published in one thousand nine hundred and fifty nine but he was working on the thirties and self so and and IT goes back to, uh, no phooey gist penni was doing this in ancient a india okay? And so you are doing something like this.

The simplest thing we can think of is that there's just connections between the words to make the the uterus and so just say, I have like two dogs enter the room okay, here's a sentence. So uh, we're connecting two and dogs together. That's like they're some dependency between those words to make some bigger meaning.

And then we're connecting dogs now to, uh, entered right and we connect a room somehow to entered and so I am going to connect to room and then room back to entered this that's the tree is that that the root is entered that that the thing is like an entering event. That's what we're saying here. And the the subject, which is whatever the dog is, is two dogs.

IT was and and the connection goes back to dogs, which goes back to that that goes back to two. I'm just, that's my tree IT IT starts at entered, goes to dogs down to two. And on the other side, after the verb, the object IT goes to room.

And then that goes back to the the determined article, I, anna, call that word. Uh, so so much of categories of words here noticing. So there are verbs.

Those are these things that typically mark, uh, they refer to events and states in the world and their nouns, which typically referred to people, places and things, what people say that that they can refer to. Other more, we think, for her to event themselves as well. They're marked by you know how they how the category the part of speech of a word is how IT gets used in language. It's like that how you decide what the the category of word is not not by the meaning.

but how is how IT gets you? What's usually the root is gonna be the verb that defines the event you.

Yes, yes, okay, yeah. I mean, if I don't see a verb, then there won't a verb until be something else.

What if you're messing? We talking about language like correct language. What if you do in poetry? Messing with stuff is IT, then then rules got the wind or right then is no you no, no, no. You're you're .

constrained by whatever language you're dealing with probably of other constrains in poetry such that you like. Usually in poetry, there's multiple constraints that you want to like. You want to usually convey multiple meanings is the idea and maybe you have like a rythm or a rimm structure as well and depending on so, but you usually are constrained by your the rules of your language for the most partner.

So you don't violate those too much. You can violate them somewhat, but not too much. So that has be recognizable as your language like an english. I can't say dogs to entered room.

Uh, I mean, I meant that, you know, two dogs entered a room and I can't mess with the order of the the articles, the articles and the now you can do that in some languages. You can, you can mess around with the order of words much more. I mean, you speak russian.

Russian has a much freer word order than english, and so in fact, you can move around words. And I told you that english has the subject, verb, object, word order, so is russian. But russian is much freer than english, and so you can actually mess around with the word order. So probably russian poetry is going to be quite different from english poetry because the word order is much less constrained.

Yeah, there is a much more extensive a culture of poetry throughout the history of the last hundred years in russia. And I also always wonder why that is. But IT seems that there is more flexibility in the way the language is used. There's more you are more female language easier, but altering the words, altering the order, the words messing with IT.

But you can just mess with different things in each language. And so in russian, you case markers so on the which you're just these endings on the nouns which tell you how IT connects each now connects to the right. We don't have that in english so when I say um mary kissed john, I don't know who the agent of the patient is except by the order of the words right in in russian, you actually have a marker on the end.

If you're using a russian name in each of those names, you'll also say is that the agent it'll be the you know of nominative which is marking the subject or an accused of will mark the object. And you could put them in the reverse order. You could put accused of first, as you could put subject, you could put the patient first, and then the verb, and then the, the, the subject.

And that would be a perfect, good russian sense. And I would still mean america. I could say john kissed mary, meaning markiss john, as long as I use the case markers in the right way, you can do that english and so .

I love the terminology of agent and patient and ah and the other ones you used those are sort of linguistic terms.

correct? Those are, those are for like kind of meaning. Those are meaning and in subject and object, or are generally used for position. So subject is just like the thing that comes before the verb.

And the object is when that comes after the verb, the agent is kind of like the thing doing, and that's kind of that means, right? The subject is often the person doing the action, right? The thing.

yeah. Okay, this is fast. So how hard is IT to form a train? General, as as our procedure toy, like if you look at different languages, is IT supposed be a very natural like is is automated? Or are some human genius involved?

Because I think it's pretty automatic at this point. People can figure the words are they can figure out the more themes, which are the technically more themes are the the minimal meaning units within a language OK. And so when you say eats or drinks, that actually has two more themes in english, there's the root, which is the verb.

And then there's some ending on IT, which tells you that's the third person. Third person, think more things are just the minimal meaning units within the language. And then the word is just the things we put spaces between english and and they're a little bit more. They have the mortality logy as well. They are the endings, this inflecting al mortality logy on the endings, on the roots.

They modify something about the word yeah. Adds additional al.

meaning we tell you, yeah, yeah, yeah. And so we have a little bit that anything was very little much more in russian, for instance. And and but we have a little bit and in english.

And so we have a little on on the nounce you can say it's a singular plural and and you can say the same thing for um for verbs like simple past tense. For examples like you notice in english, we say drinks, know he drinks, but everyone else is ee drink, you drink, we drink is unmarked in away. And then but in the past tense is just drink for everyone.

There is no myth ology at all for past tense. There is orthologic marking past tense. But it's kind of it's an irregular now.

So we don't even you know a drink to drank. You know it's not even a regular word. So in most verbs, many verbs, there's an E, D, we can add to walk to walk.

We add that to say it's the pastors that I just happened to choose in a regular as the high frequency word. And 呢 high frequency words tend to heavy regulars in english for what's in regular irregular is just， there's in a rule. So drink to drank is an an irregular drink drank.

Okay.

I supposed to walk, walked, talked.

And there's a lot of irregularities in english.

There's a lot of regulars in english, the frequent ones, the common words tend to be irregular though that there's many, many more low frequency words, and those tend to be those irregular ones.

The evolution of the regulars. fascinating. This is essentially slaying that's sticky because you're breaking the rules and then we use IT and doesn't fall the rules.

And they they say screw to the rules is faster. So he said that more themes, lots of questions. So method gies what the study of more fumes mort's .

hologic is the connections between the more themes onto the root, the roots. So in english, we mostly have sufficit. We have endings on the words, not very much, but little bit. And as opposed to prefixes, some words, depending on your language, can have you mostly prefixes, mostly suffixes, or mostly, or or both. And then even languages, several languages have things called in fixes where you have some kind of a general uh, form for the for the root and you puts up in the middle, you change the balls.

yes. So so in general, there's what two more fans per word, each one or two or three.

Well, in english it's one or two. In english tends to be one two. There can be more in other languages, you know a language like like finish, which has a very uh elaborate mortality logy. There may be ten more themes on the end of a road OK. And so there maybe millions be millions of forms of a given word OK OK.

I i'll ask the same question over over. But a how does just sometimes to understand things like more fumes? It's nice to just ask the question, how do these kinds of things evolve? So you have a great book starting of the how how, how the ordinary of processing, how languages for communication.

So the mathematical notion of how effective languages for communication, what role that plays in the evolution of language, but high level, like how do we how does the language evolve with where english is two more themes or one two more themes forward and then finish as infinity powered. So what how does how does that that happen? Is that just that's a .

really good question. Yeah that's a very good question. Is like why do languages have more morpho gy versus less portfolio? And and I don't think we know the answer to this.

I know I think there's just like A A lot of good solutions to the problem of communication. So like I believe, as you hinted, that language is an invented system by humans for communicating their ideas. And I think we IT comes down to, we label things we want to talk about.

Those are the the more themes and words, those are things we want to talk about the world. And we invent those things. And then um we put them together in ways that are um easy for us to convey to process.

But that's like a naive. And I don't I mean, I I think it's truly right. It's native.

probably right. I know that I think is simple, simple. Yeah I think nave is now even an indication and incorrect, somehow a trivial, too, too simple. I think I could very well be correct, but it's interesting how sticky IT feels like two people got together. IT just feels like once you figure out certain aspects of a language that just become sticky in the tribe forms around that language, maybe the language, maybe the try warms first, the language of all, and then you just kind of agree that you stick to whatever that is.

I mean, these are very interesting questions. We don't know really about how words, even words, get invented very much about. We don't really I mean, assuming they get invented, they we don't really know how that process works and how these things evolve.

What we have is kind of, uh, a current picture, a current picture of few thousand languages, a few thousand instances. We don't have any pictures of really how these things are evolving, really. And and then the evolution is massively confused by contact, right? So as soon as one language group, one group runs into another, we are smart. Humans are smart, and they take on whatever is useful in the other group. And so any kind of contrast which you're talking about, which I find useful, i'm going to i'm going to start using as well.

So I worked a little bit in um in specific areas of words, in in number words and in colour words and in color words that so we have in english we around eleven words that everyone knows for colors and and many more if you happen to be interested in color for some reason rather if you're a fashion designer and artist or something you may have many, many more words but we can see millions like if you have Normal color vision, Normal trio metric color vision, you can see millions of distinctions in collars. So we don't have millions efforts. The most efficient, no, as the most detailed color vocabulary have over a million terms to distinguish all the different colors that we can see. But of course, we don't have that.

So it's somehow it's been it's kind of useful for english to have have evolved in some way to eleven terms that people find useful to talk about, you know, black, White, red, blue, Green, yellow, purple, gray, pink and and I probably miss something there anyway so there's eleven that everyone knows yeah and and depending on your but you have different cultures um especially the non industrialized cultures and they'll many fewer so some cultures will have only two believe not that the dani in in papua ini have only two labels that the that the group uses for color roughly black and why they are very, very dark and very, very light which are roughly black and White and you might think o they're dividing the whole color space into lighten dark or something and that's not really true mostly just only label the light, the black and the White things. They just don't talk about the colors for the other ones. And so and and then there's other groups.

I worked with a group called the chemi down in um in bOlivia in south amErica and they have three words that everyone knows but there's a few others that are that that several people like many people know and so they have me is the kind depending on how you count between three and seven words that the group knows okay and again, they're black and White. Everyone knows those. And red, red is like that tends to be the third weird that everyone that that cultures bring in if there's a word is always read the third one.

And then after that, it's kind of all bets are off about what they bring in. And so after that, they bring in a sort of a big, blue, Green space, grew, grew. They have one for that. And then they, and then, you know, different people have different words that theyll use rather part of the space. And so anyway, it's probably related to what they want to start, what they know, what they know, what they see because they see the same colors as we see.

So it's not like they have they don't they have a week A A low color palate and the things they're looking at, they're looking at a lot of beautiful scenery OK, a lot of different colored flowers and barriers and things, and there's lots of things of very bright colors, but they just don't label the color in those cases. And the reason probably we don't know this, but we think probably what's going on here is that what you do, why you label something, is you need to talk to someone else about IT. And why do I need to talk about the color? Well, if I have two things which identical, and I want you to give me the one that's different and and the only way that varies is color, then I invent a word which tells you, you know, this is the one I want.

So I want the red sweater er, off iraq, not the Green sweater. R right? There's two. Until those those things will be identical because these are things we made and they are died and there there's nothing different about them.

And so in in industrial ized society, we have you know everything everything we've got is pretty much arbitrarily colored. Uh, but you go to the industrialized group, that's not true. And so they don't intently.

They are an interested in color. You bring bright color things to them. They like them just like we like them very colors are great.

They're beautiful. They but they just don't need to. How did talk about them?

They don't have. So probably colour words is a good example of how language evolves from sort of function when you need to communicate the use of south N I think that you kind of invent different variations and um and basically you can imagine that the evolution of the language has to do with what the early tribes doing, like what what they what kind of problems are facing them and their quickly figure out how to efficiently communicate the solution to those problems where there's a static or function, all that kind of stuff running away from a mam mother, whatever. But you know it's so I think what you're pointing to is that we don't have data on the evolution of language because many languages have formed a long time ago. So you don't get the chatter.

We've a little bit of like old english to modern english because there was a writing system and we can see how how old english looked. So the word order changed, for instance, in old english to midden english to modern english. And so, you know, we can see things like that.

But most language don't even have a writing system. So of the seven thousand, only, you know, a small subset of those have a writing system. And even if they have a writing system, they it's not a very modern writing system, so don't have IT.

So we just basically have for Mandate for chinese. We have a lot of a lot of evidence from for long time and for english and not for much else stuff from in german a little bit, but not for a whole lot of like long term um language evolution. We don't have a lot. We have snapshots is what we have got a current .

language and you get a single of that from the rapid communication and certain platforms. I can read IT. There's different communities and they'll come up with difference playing, usually from my perspective, darn by a little bit of humor or may be mocker whatever it's you know just talking shit in different kinds of ways.

And you could see the evolution of language there because um I think a lot of things on the internet you don't want to be the boring mainstream. So you like want to dv a from the proper way of talking. And so you get a lot of deviation, like rapid deviation than one community collide. You get like, uh, just a good said, humans adapt to IT and you can see that as a humor. I mean, it's a very difficult study, but you can imagine like a hundred years from now, well, if there's a new language born, for example, will get really highly olustee ata on .

I mean english changing english changes all the time. All languages change all the time. So you know is the famous um result. But the queens english. So the great if you look at the queens value, the queens english supposed to be no originally the proper way for the talk was defined by, however, the queen talked, or the king, whoever was in charge.

And and, and so if you look at them, how her voice changed from when he first became queen, one thousand nine hundred and fifty two and fifty three, when he was currently the first, I mean, that's queen lizbeth who was got who, who died recently, of course, until now, fifty years later, her voice change, her voice shifted a lot. And so that even in the sounds of british english in her, the way he was talking was changing. The walls were changing slightly.

So that's just in the sounds. There's change. I don't know what's i'm interested. We're all interested in what's driving or any these changes. The the world order of english changed a lot over thousand years, right? So I used to look like german IT looks used to be a very final language with case marking and IT shifted to a media language log contact. So our contact with french and IT became uh over media language with no case marking and so IT became this verbenas ally thing so and so that's volte IT totally evolved and so IT may be very I mean, you know doesn't involve maybe very much in twenty years this maybe what you're talking about. But over fifty and one hundred years, things change a lot.

I think will now have good data.

Which good can .

you talk to? What is sync and what is grammar? So you a book on syntax.

I did. You were asked me before about what you know, how do I figure out what the dependence structures is? I'd say the dependency structures aren't that hard to generally.

I think it's a lot of agreement of what what they are for almost any sentencing in most languages. I think people people agree on a lot of that. There are other parameters in the mix such that some people think there's a more complicated grammar than just a dependently structure.

And so like, no, he is the most famous linguist ever and he he is famous for proposing a slightly we're complicated syntax. So he he invented free structure grammar. So he's um well known for many, many things.

But in the fifties and early sixties, but late fifties, he was basically figuring out which called formal language theory. So and he figured out a framework for figuring out how complicated language, a certain type of language, might be, so called free structure grammar of language might be. So he, his, his idea was that maybe we can, we can think about the complexity of a language by how complicated the rules are.

okay. And the rules will look like this. They will have a left hand side and we'll have a right right hand side.

Something on the left side side will look band to the thing on the right hand side. So say, we will start with an and s, which is like the root, which is an a sentence. okay?

And then we're going to expand the things, uh, like a long phrase and a verb phrase is what he would say, for instance. okay? And s goes to an np in a VP is a kind of a free structure rule. And then and we figure out what an np is, an np, he is A A determiner in a non, for instance, and verb phrase is something else, is a verb, and another another phrase, and another m pace, for instance. Those are the rules of a very simple free structure.

okay? And and so he he proposed free structure grammar as a way to cover human languages and then he actually figured out that well, depending on the formalization of those grammar, you might get more complicated or less complicated languages and so could he said, well, you these are things called um context free languages that rule that he got no human languages are tend to be what he calls context free languages um and but there are similar languages which are so called regular languages and they have a more a more constrained form to the rules of the of the free structure of of these particular rules. So he, he basically discovered and kind of invented ways to describe the language and the, and those are, those are phrase structure, a human language. And he was mostly interested in english and initially, and he is work in the fifty.

So a cool questions around all. So formal language theory is the big field of just studying language formally. yes.

And IT doesn't have to be human language that they would have computer languages, any kind of system which is generating a uh uh um some set of um expressions in a language. And those could be like the the you know the statements in a in a computer language, for example. So a formal I could be that work could be human language.

So you can study programing.

yes, in heaven heavily studied using this formalism. There's a big of programmer languages within the formal language.

Okay, then the free structure grammar is this idea that can break down language into this S M P V P type of things.

It's a particular formalism for describing language OK so and charms. He was the first one. He is going to figure that stuff out back in the fifties and but he and that's equivalent actually the context programme was actually is kind of equivalent in the sense that that generates the same sentences as a dependency grammar wod as the defensive gramma.

R is a little simpler in some way. You just have a root and IT goes like we don't have any of these. The rules are implicit. I guess we just have connections between words. The free structure grammar. Is that kind of a different way to think about the the depends y grammer? So it's slightly we're complicated, but it's kind of the same in some ways.

So to clarify dependency, grammar is the framework under which you see language. You make the case that this .

is a good way .

to the language and no joke is watching. This is very upset right now. So it's just kind. But um what's the difference beam? Where is the place of disagreement um between free structure grammar and dependency grammar there they're very close.

So free structure grammar and dependency grammar aren't aren't that far apart. I like dependancy grammar because it's more prosperous. It's more transparent about representing the connections between the words.

It's just little harder to see in frame structure grammar the place where chomsky sort of devolved there are went off from from from this is he also thought there was um something called movement okay. And so so and that's where we disagree. Okay, that's the place where I would say we disagree and and I mean, maybe we will get into that later.

But the idea is, if you want do you want me to explain that? I would explain movement in o you think, okay, so here's the movement is charms. He basically sees english and he says, okay, I said you so we had that sentence early like he was like two dogs enter the robot to change a little bit, to say two dogs will enter the room and he notices that, hey, english, if I want to make a question, I just know a question from that same sentence.

I say, instead of two dogs will into the room, I say, will two dogs enter the room? Okay, there's a different way of to say the same idea and it's like, well, the xxi arive of that willing, it's at the front as opposed to in the midnight. okay.

And so and he looked, you know, if he look at english, you see that that's true for all those model verbs and for other kinds of exiles verbs in english, you always do that. You always put in excEllent verb at the front. And and what when he saw that? So now I say, um I can win this bet.

Can I win this bet? right? So I move a can of the front. So I see that's a theory. I I just gave you a theory there. He talks about IT as movement, that word in the decline things to declarative is the root is deserve default way to think about the sentence and you move the eagerly verb to the front.

That's a movement theory, okay, said and he just not that was just so obvious that IT must be true that that there's nothing more to say about that, that this is how auxiliary verbs work in english. There's a movement rule such that you're move like to get from the declarative to the interrogative of you're moving the excEllent to the front. And it's a little more complicated as seems as you go to simple, simple present and simple past because if I say, you know, john slept, you have to say, did john sleep not slept john, right?

And so you you have to somehow get an axilla verb. And I guess underlying ly, it's like slept, it's a little more complicated than but that his idea there's a movement. okay.

And so a different way to think about that, that isn't. I mean, then he ended up showing later. So he proposed the theory of grammar, which has movement. There's other places where he thought there's movement, not just exiling verbs but things like the past of an english and things like um a question wh questions venture places where they thought there's also a movement going on.

And in each one of those these thinks there's words while phrases and words are moving around from one structure to any which you call deep structure to surface structure. Mean there's like two different structures in in his theory okay um there's a different way to think about this um which is there's no movement at all. There's A A mexico copying rule such that the word will or the word can, these exiles ary verbs.

They just have two forms, and one of them is the declarative, and one of them is interrogative. And you basically have the declarative one. And oh, I form the interaction, or I can form one from the other, just matter which direction you go.

And and I just have a new entry which has the same meaning, which has a slightly different argument structure. Argument structure is deffand cy word for the ordering of the words. And so if I say the the dogs, two dogs can or will enter the room, there's two forms of will.

One is will declarative and and then, okay, i've got my subject to the left that comes before me and the verb after me in that one. And then the will interrogative is like, oh, I go first interrogative will is first, and then I have the subject immediately after, and then the world after that. And so you just you can just generate from one of those words another word with a slightly different argument structure with different ordering.

and is just lexicon copies, not necessarily moving for one. There is a romantic notion that you have like one main way to use a word, and then you can move IT around, which is essentially what .

movement is. The lexicon copping is similar. So do lexicon copying for that same idea that maybe the declaration is the source, and then we can copy IT. And so an advantage there's multiple ada vantage of the electric copying story is not my story. This is like, uh, ivan saw linguis bunch of linguis have have been proposing these stories as well, you know intend them with the movement story.

okay? No, is he died a while ago, but he was one of the proponents of the non movement for the legs of coffee story and so that is that um a great advantage is well, choy really famously in one thousand nine and seventy one showed that the movement story leads to learn ability problems IT leads IT leads to problems for for how language is learned. It's really, really hard to figure out what the underlying structure of the languages. If you have both free structure and movement, it's like really hard to figure out what came from what there's like a lot of possibilities there if you don't have that problem learning, the learning problem .

gets a lot this year. Just there's copies. We say the learning problem. You mean humans learning a new language?

Yeah just learning. go. So baby is lying around and listen to the crib listening to me talk and know how are they learning english or you know, maybe it's a two year old who's learning interrogatives and suffer when you how are they doing that? Are they doing IT from? Like are they figuring out.

Or like so charm, he said, is impossible to figure out. Actually he said it's actually impossible, not not hard but impossible. And therefore that's that's where universal grammer comes from is that he has to be built in.

And so what they're learning is ah there's some built in movement is built in in his story is absolutely part of your language module and and then you are you're just setting parameters you're said depending on english is just sort of a variant of the universal grammar and you're figuring out, oh, which orders does english do these things that the non movement story doesn't have this it's like much more bottom up. You're learning rules. You're learning rules one by one.

And oh, is this this word is connected to that word a great advance. Another advantage is learn able. Another advantage of IT is that IT predicts that not all exceler ies might move like IT IT might depend on the word depending on whether you and that turns out to be true.

So there's words that um that don't really work as a zilla. They work in declared ative and not an interrogative. So I can say um i'll give you the office at first if I can say aruna invited to the party. okay. And and and that's an an interrogative form, but it's not from I aren't invited the party there is no, I aren't right.

So that's that's interrogative only and and then we also have forms like um out uh I I ought to do this and and I guess some british old british people can say exactly doesn't sound right, does IT for me that sounds ridiculous. I don't even think odd is great. But I mean, I totally recognized, I want to do ideas.

Not too bad. I can say I have to do this. That sounds .

really good yeah.

something I don't know. I just sounds completely I anyway, so the variants here and if all of these words just work in one versus the other and and that's like fine under the lexicon copying story, it's like, well, you just learn usage. Whatever the usage is, is what you is what you do with this, we do with this word. But um IT doesn't it's a little that harder in the movement story, the movement story like that's an advantage, I think, of legs copying in all these different places. There's all these usage variance, which make the movement story little bit harder to work.

So one of the main divisions, here's the movement story, the last story that I rewind to the face, structure grammar versus dependency grammar.

There's a equivalent, in some sense, in that for any dependency grammar I can generate a dependent, a free structure grammar, which generates exactly the same sentences. I just, I just like the dependency grammar formalism, because that makes something really silent, which is the the length of dependencies between words, which isn't so obvious in the free the free structure. It's just kind of hard to see it's in there. It's just very, very.

uh, technically I think face structure grammar is map able to dependency grammar or yeah but there is like this, like little labels, S M P, V P yeah.

For a particular defensive grammar, you can make a free structure grammar which generates exactly those same sentences and vice versa. But there are many free structure grammar which you can really make a dependency grammar. I mean, there you can do a lot more and a free structure grammar, you get many more of these extra nodes. Basically, you can have more structure. And there and some people like that, maybe there's value to that don't like IT .

well for you. So we should clarify. So dependency grammar, it's just one one word depends on only one word. And you form these trees, yes, and that makes IT really puts priority on those dependencies, just like as as a tree that you can then measure the distance of the dependency for one word to the other.

They can then map to uh, the cogniac processing of the of the sentences, how well, how easy is to understand all that kind of stuff. So I just puts the focus on just like the mathematical, i'm a distance of dependence between words. So that gives you a different focus.

absolutely. Just continue on a thread of josey because it's really interesting because IT, as you're discussing disagreement to the degree there's disagreement, but you also in the history of the study of language, which is really awesome. So the image context reverses regular. Does that distinction come into play for the fences grammar?

No o not at all. I mean, regular, regular languages are too simple for human languages. They are, they is a part of the hierarchy.

But human languages are in in the free structure world, are definite, at least context free. Maybe a little bit more, a little bit harder than that. But so there's something called contact sensitive as well.

Where you can have like this is not just the form language description in a context free grammar. You have one. This is like a bunch of lake formal language there.

We're doing here. I love IT. okay. So you have you have a left and side category and you're expanding anything on the right is is a uh, that's a context free.

So like the idea is that, that category and the left expands in independent of context to those things, whatever they are on the right doesn't matter what. And a context sensitive says, okay, I I actually have more than one thing on the left. I can tell you, only in this context maybe you have like a left in the right context, or just the left context, or right context.

I have two earth more stuff on the left tells you how to expand that those things in that way. okay? So it's contact sensitive.

A regular language is just more constrained. And so IT IT doesn't allow anything on the right. IT allows very IT a visit. It's a one very complicated rule is kind of what of A A regular language is and so IT doesn't have any um to say long distance depends is that doesn't allow recursion for instance there's no recursion a recursion is where which is human language have recursion they have ebel ding and you can well he doesn't vow center ebel ded, recursion which human languages have which .

is what center embedded recurs within a sense so he were .

going to get to that but now the formal language stuff is a little aside of which chomsky wasn't proposing IT for human languages even he was just pointing out to that human languages are context free. And then he was most for human because that was kind of stuff he did for formal languages. And what he was most interested in was human language.

And that's like the movement is where we we we where where he said to set off on, I would say very interesting, but wrong foot IT was a kind of interesting it's A I agree, it's a very interesting history. So there's proposed this multiple theory is in fifty seven and sixty five there they all have this framework that was free structure plus movement, different versions of the of the free structure. And the movement in the fifty seven is the famous original bits of chance key's work.

And then seventy one is when he figured out that those lead to learning problems, that that there are cases where a kid could never figure out which rule, which set of rules was intended. And so and then he said, well, that means it's innate. It's kind of interesting.

He just really thought the movement was just so obviously true that he couldn't he didn't even entertain giving IT up. It's just obviously that's obviously right. And IT was later where people figured out that there's all these like simple ways in which things which look like generalizations aren't to generalizations.

And they across the category, their words specific and and they and work, but they don't work across the various other words in the category. And so it's easier to just think of these things as lexicon hobbies. And I think he was very obsessed.

I don't know. I'm like guessing that he just he really wanted this story to be simple in some sense, and language is a little more complicated in some sense. So he didn't like words um he never talks about where he likes her, about combinations of words and words.

Look up the dictionary. There is fifty senses for a common word, right? The word take will have thirty or forty cents in IT.

So a so many different senses for common words. And he just doesn't think about that or doesn't think that's language. I think he doesn't think that language. He thinks that words are distinct from combinations of words. I think that the same. If you look at my brain and in the scanner, while i'm listening to a language I understand and you compare, I can localize my language network in a few minutes, and like fifteen minutes.

And where you do I listen to language? I know I listen, you know, maybe some language I don't know, or I listen to muffled speech, or I I read sentences that I read, non words like I do anything like this, anything that sort of really like english, and anything is not very like english. So I got something like IT and not, and I got to control and the boxes, which just just, you know, the three d in in my brain that are responding most are are the is a language area and and that's this left lateralization area in my head.

And and whatever I look in that network, if you look for the combinations versus the words, it's there everywhere. It's the so it's hard to find. There are no that we know. I mean, that's it's a overstate right now at this point.

The technology isn't great, it's not bad, but we have the best, the best way to figure out what's going on in my brain when i'm listening, reading language is to use fmri, functional magnetic presence imaging. And that's a very good localization method. So I can figure out where exactly these signals are coming from.

Pretty real, down to millimeters. Cubic millimeters are smaller and okay, very small. We can figure very well.

The problem is the when okay ah it's it's measuring um oxygen OK and oxygen kes a little while to get to those cells on the order of seconds. So I talk fast, I I probably listen fast and I can probably stay things really fast. So a lot of stuff happens in two seconds.

And so to say that we know what's going on, that the words right now in that network, our best guesses, that whole network is doing something similar, but maybe different parts of that network are doing different things. And and and that's probably the case. We just don't have a good methods to figure that out right at this moment.

And so since we're kind of talking about the history of the study of language, what other interesting disagreements in your both at M I, T. Or war for a long time, what kind of interesting disagreements, their attention of ideas are their between, you know, joskin. We should say that no was in the linguistics department and you are, I guess, for time we're filed there, but primarily a brain and cogent science department. This is another way of studying language and you've been talking about I M my eye so like what is there something else interesting to bring to the surface about the disagreement between the interview or other people in .

the yeah I mean, i've been my tea for thirty one years since one thousand hundred and ninety three, and he chops spend her much. So I I met him. I knew him. I I met when i've first got there, I guess.

And and we would interact every now and then i'd say that so I tell are our biggest difference is our methods and so um that's the biggest difference between me and no um is that I gather data from people, I do experiments with people and I gather corpus data, whatever whatever cripps data available. And we do quantitate tive methods to evaluate any kind of hypotheses we have. He just doesn't do that.

So you know no, he has never once been associated with any experiment or cruise work ever and so it's all thought experiments. It's his own intuitions. So I just don't think that's the way to do things. Um that's that's a you know, across the street, they are across the street from a kind of difference between brain coxe and linguistics. S I mean, not all in some of the linguis, depending on what you do, more speech orientated, they do more quantity stuff, but in the in the meaning of words and wealth, it's combinations of words and taximeter. S they tend not to do experiments and uh and corporation.

all success probably well. But the message is symptom of a bigger approach, which is of a psychological philosophy. And knowing for you, it's more sort of data driven. So almost mathematical approach yeah I mean.

i'm psychologist, so I would say we're in psychology, you know I mean in brain cog sciences is my old psychology department. IT was the psychology private up until one thousand nine hundred eighty five and that became the brain cog science department. And so I, I mean, my training is in pi. I, an computer sciences, but I am a psychologist. I mean, I.

I mean, I know what I am, the psychos. Yeah, yeah.

you are and what I am, but I, but I I called the linguistic. I have have to be called the computer scientist. I have to called called just any of those things.

But in the actual I call them, manifest itself outside the methodology is like these differences, these of differences about the movement store versus the lexicon copy story.

Yeah, those are theories, right? So the there like the theory are I but I think that the reason we differ in part is because of how we evaluate the theories. And so I evaluate theories quantitatively and he name doesn't got IT.

Okay, well, let's let's explore the theory that you sporting your book. Let's return to this dependency grammar framework of looking at language. What's A A good justification? Why the dependence grammar framework is a good way to explain language? What's your intuition?

So the reason I like dependency grammar, as i've said before, is that it's very transparent about its representation of distance between words.

So it's like IT all IT is is you've got bunch of words, you're connecting together to make a sentence and A A really need insight, which turns out to be true, is that the further apart the pair of words are that you're connecting, the harder IT is to do the production, the harder IT is do the comprehension. Is that harder to produce? Harder understand when the words are far.

When they are close together, it's easy to produce and it's easy to comprehend. Let me give you an example. okay? So we have in any language, we have mostly local connections between words, but they're abstract. The connections are abstract between categories of words.

And so you can always make things further apart if you put your if you add modification, for example, after a known, so a known in english comes before a verb, the subject now comes before verb. And then there's an object after, for example. So I can say what I said, four years of the dog enter the room, or something like that. So I can modify dog. If I say something more about dog after IT, then what i'm doing is, indirectly, i'm lengthy the dependence, the dependence between dog and entered by adding more stuff to IT so I this make IT explicit here if I say, um the the boy who the cat scratched cried, we're gonna a mean cat here and and um and so what i've got here is like the boy cried IT we'd be a very short, simple sentence and I just told you something about the the boy and I told he was he was the boy who the cat scratched .

okay so the cry is connected to the boy at the end, connected to the body in the beginning.

right? And so I can do that. And I can say that that's a perfectly find english sentence. And I can say the cat which the dog chased ran away or something. Okay, I can do that, but it's really, so, really hard.

Now I i've got to whatever have you the boy who the cat now is, I try to modify cat OK. The boy who the cat, which the dog chased, scratched, ran away. Oh god, that's hard, right? I can.

I'm sorry, just working that through my head how to produce and how to it's pretty just horrendous to understand it's not so bad at least i've got a intention there to remark the boundaries but it's that's really complicated that's sort of english and away I mean that follows the rules of english but uh, so what interesting about that is, is that what i'm doing is nesting dependent here and putting one i've got a subject connected to a verb there and and then I modifying that with clause, another laws, which happens to have a subject in a verbal. And i'm trying to do that again on the second one. And what that does is that lengthens out the dependence, multiple dependence actually to get lengthened out there.

The dependency is get get longer, longer on the outside ones get long and even the ones are between get kind along and you just so what's fascinating is that that's bad. That's really horrendous in english, but that's horrendous in any language. So and no matter what language you look at, if you do just figure out some structure where i'm going have some modification following some head, which is connected some later head and I do again, IT won't be good IT guaranteed like one hundred percent that will be uninterpreted in that language .

in the same way that was uninterpreted is whenever the uh boy cried this um there is a dependence between two words and then you counting the number of what more themes between them.

That's a good question. I I just say words. Your words are more themes between.

We don't know that actually that's a very good question. What is the distance metric? But let's just say words and you're .

saying the longer the distance that dependence, the more no matter language of supplies even okay talk about OK um but that the people will be very upset that speak their language not upset but they'll either not understand IT there will be like this is that their brain will be working in overtime yeah they will have a hard time either producing you're comprehending .

IT they might tell you that's not their language you know it's sort of the language and is following like disagree with each of those pieces as part of the language. But somehow that combination will be very, very difficult to produce and understand.

Is that a chicken or egg issue here? So like i'm giving you explanation.

So I well, I mean, and i'm giving you two kinds of explanation. I'm tell you the center and bedding, it's nesting. Those are the same.

Those are cinnamons for the same concept here. And the explanation for what those are always hard, center betting and nesting are always hard. And I gave you an explanation for why they might be hard, which is long distance connections.

There's when you do center betting way, you think you always have long distance connections between the dependence. You just so that's not necessarily the right explanation. I can go through reasons why that's probably a good explanation, and it's not really just about one of them.

So probably it's a pair of them or something of these dependence that you get long that drives you to like we really confused in that case. And so what the the behavioral consequence there you I mean, we this is kind of methods like how do we get at this? You could try to do experiments to get people to produce these things.

They're going have a hard time producing them. You can try the experiments to get them understand them and see how well they understand them. Can they understand them? Uh, another method you can do is give people partial materials and ask them to complete them.

Know those those in bedded materials and and they fail. So i've done that. I've done all these .

kinds of things what I mean. So so central betting, meaning like you take a Normal sense that boy cried and inject your crap in the middle that separates the boy and cried, okay, that central bedding and nesting .

is set in top of that. No nesting is the same thing. Central bedding is a totally equivalent terms.

And so what your saying, there's a lot of different kinds of experiments you can do. I mean, I like to understanding one is like have more in bedding, more central, Better. Is that easier to and to measure the lever?

Understand you beautiful ways to do that. I mean, there's is the simplest ways. Just ask people, how good is the sound? How natural this sound? That's a very blunt, but very good measure.

Very, very reliable. People will do the same thing. And so is like, I don't know what that means exactly, but it's doing something such that we're measuring something about the confusion .

and the difficulty associated with. And those like those are giving a signal that why you can say what what completion .

central if you give them a partial sense, I say um the book which the author who and I ask you to now finish that off from, I mean either say yeah yeah but you can say it's written in front of you and you can just type in much print time as you want. They will even though that once not too hard, right? So if I say it's like the book is like how the book which the author who I met wrote was good, you know it's a very simple completion for that.

If I give that completion on online somewhere to a you um a crowd sourcing platform as people complete that, they will miss off of a very, very regulated at half the time, maybe two thirds at the time. They'll say they'll just leave off one of those other phrases even with that simple so to say the book uh which the author who and it'll say um was um they won you need three verbs so I need three verbs or who I met wrote was good and they'll give me to you will say. Who who is famous was good or something like that, don't just give me to.

And the'd happen about sixty percent of time. So forty percent, maybe thirty, you'll do IT correctly in correctly meaning. Theyll do with three verbs. Ys, I don't know what's correct or not know this is hard. It's a hard test.

Yeah, I actually i'm strugling in my head.

yes. Well.

it's easier you list. If you look at .

a little easier, then the listings is pretty tough because you have to because there's no traces of IT. You have to remember the words that i'm saying, which is very hard to charity. We wouldn't do IT this way.

Do IT written. You can look at IT and figure IT out is easier. Many dimensions in some ways, depending on the the person.

Easier to gather written data for. I mean, almost I C I work and cycle anguish s right, psychology of language and stuff. And so a lot of our work is based on written stuff because it's so easy together. Data from people doing kinds of task spoke and tests are just more complicated to administer and analyze because people do weird things when they speak and it's harder to analyze what they do. But they um they they generally point to the same kinds of things.

So okay so the universal theory of language yeah by tid gibson is that you can form dependency. You can form trees from any senses. You can measure the distance in some way of those dependent season. And then you can say that, uh, most languages have very short dependencies. All languages.

all languages, all languages have short dependencies. You can actually measure that. So a an next student of mind, this guys at university california, irvine, and Richard, if few trial at a thing of bunch years ago now where he looked at all the languages we could look at, which was about forty initially.

And now I think there's about sixty for which there are dependency structures like you. So there are meaning is going to be like a big text, bunch of text which have been passed further dependency structures and this about sixty of those which have have been passed that way. And for all of those, um you can what what he did was take any um any sentence in one of those languages and uh and you can do dependency structure and then started the route where we're talking about dependency structures.

That's pretty easy now and it's trying to figure out what a control way you might say the same sentence is in that language. And so we just just like, all right there's a root and I say as a sentences um we go back to you know two dogs entered the room as of entered is the root and entered has um two dependence is god dogs and IT has room okay? And what he is is like that scramble, that order that's three things the root and the the head and the two dependence and just some random, just random and then just do that for all the dependence and so now look, do IT for whatever what is two in dogs and for in room and and know this not is a very short sentence.

When sentence is get longer and you have more dependence, there's more scrambling that's possible. And when he found what, so that that's one, you can figure one scrambling for that sentence. IT is a hundred times for every sentence in every corp.

And every one of these texts, every corporate. And and then he just compared the dependency length in those random scrambling to what actually happened. The the english a of the french or the german was in the original language, or chinese or like eighty nine no sixty languages.

okay. And and the dependency length are always shorter in the real language compared that this this kind of a control and there's another that he is is a little more rigid his control. So um that the way I described IT, you could have crossed dependencies like that by scrambling that way you can scrambling anyway at all.

Um languages don't do that. They can not cross dependencies very much like so the pencil structure, they they tend to keep things uh non crossed as in this you know they is a technical term they call that projective but it's just noncircular is all that is projective. And so if you just constrain the the scrambling so that IT only gives you projective non crossed is the same thing holds so so that you still still human languages are much short uh then this this kind of a control.

So there's like IT, what he means is that that that were in every language we're trying to put things close in relative to this kind of a control like IT doesn't matter about the word, some of these verb final, some of them these verb media like english, and some are even verb initial. There are a few language of the world which have VS o world, word, order, verb, subject, object. Language is haven't talked about those at ten percent of. And even even .

in those languages are still short dependences .

short dependencies is rules.

Okay, so how what what are some possible explanations to that that for why while languages have evolved that way so that that's one of the um supposed disagreements, you might have a charley. So you consider the evolution language in in terms of information theory, yeah. And for you the purpose of languages, ease of communication, right?

In processing, right? So I mean, the story here is just about communication IT. IT is just about production, really. It's about easy production is the .

story we say production.

oh x me. Each of language production is easier for me to say things when the come what i'm doing, when ever i'm talking to you. So somehow i'm formulating some idea in my head and putting these words together.

And it's easier for me to do that to put to say something where the words are close, closely connected and independent y, as opposed to separated, like by putting something in between and over and over again. I it's just hard for me to keep that in my head like that. That's the whole story.

Like this story is basically, I like that the dependency grammar server gives that to you like just like long, long as badge or good. It's like easy. You're keep in mind because you have to keep that in mind for probably for production know probably matters and comprehensive as well.

like also matters and comprehension.

But I guess it's probably all for production producing. Is that it's easier for me to say that ends up being easier for you also, that's a very hard to disentangle. This idea of who's IT for is IT for me, the speaker, or is IT for you, the listener? I mean, part of my language is for you, like the way I talk to you is going to be different from how I talk to different people.

I'm i'm definitely angling what i'm saying to who i'm saying, right? It's not like i'm just talking the same way to every single person. And so I am sensitive to my audience, but how does that does that, you know work itself out in the in the dependently link differences? I don't know. Maybe that's about just the words that part .

you know which words I select my initial tuition is that you optimize language for the audience yeah but but this is just kind of like messing with my head lobed to say that some of the optimization might be and may be the primary objective of the optimization might be the ease of production.

different senses. I guess i'm i'm like very selfish and I don't like I don't think it's like it's all about me. I'm like I am just doing this easier for me.

I don't want I like all I mean, but I have, of course, choose the words that I think you're gonna know. I'm not going to choose words you don't know if I am going to fix that. I so there is but maybe for for the syntax, for the combinations. It's just about me. I feel like it's I I don't .

know though it's but the purpose communications to understood is to convince others in one. So like the selfish thing .

to be understood about the living OK.

right? I mean, like ea production.

how can me be understood then? I don't .

think .

it's circular. So I the primary I think .

the primary objective is to be understand about the listener because otherwise if you're optimizing to for the east production, then you you you're not going to have any the interesting complexity of language like you're trying to like .

to control for what IT is. I want to say i'm saying let's control for the thing the the message control for the message .

to sage need to be understood.

That's to go ah but that's the meaning so i'm still talking about the form, just the form of the meaning. How do I frame the form of the meaning is all i'm talking about you're talking about harder thing I think is like how like trying to change me. I can't let's keep the meaning constant.

Like you have to keep the meaning constant. How can I phrase whatever is I need to say? Like I got to pick the right words and i'm going pricking order so that it's so easy for me. That's what I think .

is probably I think i'm still tying me and form together in my head. But you're saying if you keep the meaning of what you say cause that what the optimization yeah IT could be the primary objective that optimization is the for production. That's interesting. I am struggling to keep constant meaning it's just so a man i'm such on the human right so for me the form without having introduced honest, the form of the meeting are tied together like deeply because i'm a human like for me when i'm speaking, could haven't thought about language like in a rigors way about the form of language.

Look for any event, there's there's an unbounded I I don't want to see infinite, but sort of a Better ways of that. I might communicate that same event. These two dogs entered the room.

I can say in many, many different ways. I can say, hey, there's two dogs. They entered the room. Hey, the room was entered by something. The thing that was entered was two dogs.

I mean, there's I mean, it's kind of awkward and we are, but those are all similar messages with different forms. The different ways I might frame, and of course, I use the same words there all the time. I could have referred to the dogs as, you know, a domination and a poodle or something know there.

I could have been more specific or less specific about what they are. And I could have said been more abstract about about about the number there's like so i'm trying to keep the meaning, which is this event constant and then how am I onna describe that to get that to a kind depends on what you need to know, right? And what I think you need to know, but i'm like turning let this get control for all that stuff and not in the i'm just like choosing, but i'm doing something simple other than you're doing, which is just forms. And yes.

what is to use specifying the species, the breed of dog and. Whether the cuter not is changing the meaning that might be yeah.

yeah that would be changed or that would be changed. The mean for sure.

right? You're just mop that's .

changing the mean. But so even if we keep that constant, we can still talk about what's easier heart for me, right? The listener in the which free structures I use, which combinations which .

this is so fascinating, and just like A A, A really powerful window into human language. But I wonder still throughout this, how vast the gap between meaning and form I just I just have this like maybe romanticized notion that they're close together, that they evolve, close the hand in hand, that you can just simply optimize for one without the other being in the room with us. Like it's what kind of an iceberg form is, the typical iceberg and the rest the meeting is the iceberg. You can't like sup.

but I think that's why these large language models are so successful because that they're good at form and form is in that hard in some sense and meaning is tough still. And that's why they're they don't understand what they're who we talk about that later maybe, but like we can distinguish in our forgetting about large line which moves like humans, where maybe we will talk about that later.

Two is like the difference between language, which is a communication system, and thinking, which is meaning. So language a communication system for the meaning. It's not the meaning. And so that's what I mean. And there's a lot of interesting evidence we can talk about relevant to that.

Well, I mean, that's a really interesting question. What is the what is the different a language written, communicated versus thought? What to use the difference between them?

Well, you or anyone cast to think of a task which they think is is a good thinking task. There's lots and lots and tasks which should be good thinking tasks. And whatever those tasks, let's say, it's no playing chess or as a good thinking task or playing some game or doing some complex puzzles.

Uh, maybe maybe remembering some digits that's thinking. Remember meering some lot of different tasks you might think maybe just listening to music is thinking or there's a lot of different task we might think of his thinking. This is woman in my department of federal and she's done a lot of work at on this question about what's the connection between language and thought and and so SHE uses, I was a french earlier to mr.

Ifm ri, that's her primary method. And so SHE is been really fascinated by this question about whether what language is okay. And so is, I mentioned there, you can localize my language, your languages, in a few minutes, okay, like fifteen minutes.

I can listen the language, listen to non language or backward speech or something. And we will find areas left later. Ized network in my head, which is specially, which is very sensitive to language, as opposed to whatever that control was. Okay.

can specially me by language, like communicating language that .

is like just sentences. You know, i'm listening to an english of any kind story or can read sentences, anything at all that I understand. If I understand IT, then I will activate my language network. So right now, my language network is going like crazy when i'm talking and when i'm listening to you because we're both were .

communicated and that's .

pretty stable yeah it's incredibly stable. So I I haven't to be mary third woman have rego. So i've been scanned by her over and over and over since two thousand and seven or six or something.

And so my language network is exactly the same in a month ago as IT was back in two thousand. And it's amazingly stable, astounding with it's it's a really fundamentally cool thing. So my language neuk is is like my face. okay? It's not changing much .

over time inside my at which point in, as you grow up from baby to adult, does that stabilize? We don't know.

That's a very hard question. They're working on that right now because of the problem scanning. The kids like doing the trying to do trying to do the the localization on little children in in the scanner lying in the FM ice.

And that's the best way to figure out where something going on inside our brains and the scanners loud and you're in this tiny little area, your cluster phobic and IT doesn't bother me at all. I can go sleep in there. But some people are bothered by and little kids don't really like IT and they don't like to lie still.

And you have to be really still because you move around that that mess up the coordination of where where everything is. And so you know, try to get you know your question is how and when our language developing, know how we did. How does this left literalist system come to play? Where is you know, and it's really hard to get a two year old to use, but you can maybe where there is starting to get three and four ign five rules to do this task for short periods. And IT looks like it's there pretty early.

So clearly, when you lead up to you like a baby's first words before that, there's a lot of fascine turmoil going on. Figure out what what are these people saying yeah and you're trying to like, make sense hot is that connect to the world now that kind of yeah that might be just fascinating development that's happening there. That's yeah hard to respect.

Anyway, you say we were back to this scanner and I can find my network in fifteen minutes, and now we can ask, we can ask, find my network, find yours, find twenty other people do the task of, and we can do some other tasks. Anything else you think is thinking of some other thing? I can do a speciaal memory task.

I can do a music perception task. I can do programing task if I programme, okay, I can do what I can like, understand computer programs. And none of those tasks tap the language network at all, like at all.

There's no overnight they do. There are highly activated other parts of the brain. There's a there's a bilateral network, which I think you chance to call the multiple demands and network, which does anything kind of hard.

So anything is kind of difficult in some ways will um activate that multiple demands network. I mean, music will be in some music areas, you know there's music specific kinds of areas. And so but there but but none of them are activating the language area at all unless there's words you look. So if you have music and there's a song and you can hear the words many, then then you get language.

We talking about, speaking and listening. But are or we also talking about reading?

This is all comprehension of any .

kind and this .

network doesn't make any difference if it's written or spoken. And so the thing that he calls federico calls the the language network is a high level language. So it's not about the spoken the spoken language and its not about the written language is about either one of them.

And so so when you do speech, your sort of this, you either you're listen to speech and you you need to track away some language you don't understand and see which and or use to track away backbone wards speech, which sound sounds like speech but isn't and and then so you you take away the sound part altogether and so and then if you written, you get exactly the same network. So for just reading the language versus reading some of don sense words or something like that, you will find exactly the same network. So this is about high level um language.

You in this case and the same thing happened, production a little harder run the scanner. But the same thing happens in production, you get the same network. So productions little hard way. You have to figure out how do you run a task you know in the network such as you're doing some kinds production and I can't remember what they've been, a bunch of different kinds of tasks there where you get people to produce things yeah figure how to produce and the same network goes on there exactly the same place.

So if you way wait so if you read random .

words yeah things .

like gish yeah .

yeah Lewis, Carol, twice. Billy, very Walker, right? They call that ever Walker speech.

The word doesn't get up .

and not as much. There are words in there. It's words, stuff. So it's a vo, yeah yeah. So there's like visit.

The more language like that is the higher IT goes in the language network and that network is there from when you speak from as soon as you learn language and and is there like you, you speak multiple languages, the same network is going for your multiple languages. So you speak english, you speak russian, than both of them are hitting that same network. If if you're fluent in those languages .

for programming.

not at all is not amazing, even if you're a really good programmer that is not a human language, is is not conveying the same information. And so IT is not in the language network.

And so that is mine below as I think that's pretty is amazing.

So that's like one set of day. This hers like shows that but you might think is thinking is is not language. Language is just the just this conventionalist system that we've worked out in in human languages. Oh, another fascinating little bit to bit is that even if they're these constructed languages like clean on or I don't know the languages from game with thrones, am sorry, I don't remember those languages.

And a lot of people there's .

people that speak those languages, they really speak those languages because the people wrote the languages for the shows. Um they did an amazing job of constructing on something like a human language and those that that lights up the language area that's like because they can speak you know pretty much arbitrary thought in a human language. It's not a it's a constructed human language that probably is related to human languages because the people that were constructing them, where we're making them, like human languages in various ways. But IT also activates the same network, which is pretty really cool, anyway.

decided to go into a place where you maybe a little bit photo official, but is impossible that this area of the brain is doing some kind of translation into a deeper set of, almost like a concepts mister has .

to be doing. So is doing in communication, right? IT is translating from thought. Whatever that is is more abstract. And it's doing that. That was doing like IT is that is kind of what IT is doing. It's kind of a meaning network.

I guess yeah a translation network yeah. But I wonder what is that the core at the bottom of IT like what our thoughts are, they are thoughts to me, like I don't thoughts and words, are they neighbors or are is a one turtle sitting on top of the other? Meaning, like a very deep set of concepts that we there's connections right between .

the what these things mean, and then the probability products, the brain, that what these things mean. And so. You know, when I am talking about whatever IT is I wanted talk about, if it's it'll be represented somewhere else. That knowledge of whatever is will .

be represented somewhere else. I wonder if there is pressed encoding of meaning. I don't know that support from language that link you know, I guess I guess the implication here is that, that we don't think in language that's .

correct is cool and that's so interesting. So people, I mean, this is like hard to do your experiments on, but there is this idea of inner voice. And a lot of people have an inner voice.

And so if you do a poll on the internet mask, if you you hear ourself talk when you're just thinking whatever about seventy, eighty percent of people will say, yes um most people have an inner voice. I don't and so I always find this strange. So when people talk about an inner voice, I always thought this was a metaphor.

And they here, I know most of you, whoever listening this thing, things I crazy. Now I I don't have an inner voice and I I just don't know what you're listening to. I I just did sounds so kind of annoying to me, but that have this voice going on what you you're thinking. But I guess most people have that and I don't have that and I don't we don't really know what .

that that connects to. I wonder if the invoice active?

I don't know. Let me, the T, V speeches, right? So that's like you hear.

Do you have any voice? Think so. Oh.

a lot of people have the sense that they hear other, they hear themselves and then say they read someone eating and life for people tell me that they hear that other other person's voice when they read other people's in emails. And i'm like, wow, that sounds so disruptive.

I do think I like vocalized what i'm reading, but I don't think I hear a voice.

Well, let's be probably have an inner world. People have an inner voice. People have the strong percept of hearing sound in their heads when they're just thinking.

I refuse to believe that the majority of people.

majority absolutely was. It's it's like two thirds or three quarters. It's lot. I never asked class and and I went internet. They always say that so are unity.

IT could be a software port flaw. IT could be, you know what i'm reading? Yes, inside my head, i'm kind of like saying the words, which probably the wrong way to read, but I don't hear a voice.

There's no percept of a voice. I refuse to believe the majority people have anyway, is a faster, the human brains faster, but is still blew. M my mind that the language that does appear, comprehension does appear to be separate from thinking.

So that's one set, one set of data from federico's group is that um no matter what task you do, if he doesn't have words and combinations of words in IT, then IT won't let up the language network. You know you could itto be active somewhere else, but not there. So that's one.

And then this other piece of evidence relevant to that question is the IT. Turns out there are these this group of people who've had a massive stroke on the left side and wiped out their language network and IT as long I didn't wake out everything on the right test flaw. In that case, they wouldn't be no cognitive ly functionality.

But if they've just wiped out language, which is pretty tough to do because it's it's very expensive on the left. But if they have, then there are these these patients like this uh called took called global of physics who um can do any task just fine, but not language. They can, they can talk to them, I mean, they don't understand you, they can speak, can write, they can read, but they can do, they can play chess, they can drive their cars, they can do all kinds of other stuff.

Do math, you all like. So math is not in the language area. For instance, you do rithmetic itself. That's not languages. It's cut symbol.

So people are set of confused some kind of symbolic processing with language, and symbol processing is not the same. So there are symbols and they have meaning, but it's not language is not a you know conventionalized language system. And so language math isn't there. So they can do that. They do just as well as their control age matched controls.

And all these tasks is rose verely over in university college london who has bunch patients who who SHE shown us that they are just um so that that sa combination suggests that language isn't necessary for thinking IT IT doesn't mean you can't think in language. You could think in language because language allows a lot of expression, but it's just you don't need IT for thinking. It's IT suggest that language .

separate is a separate system log. I'm trying to load that yeah because that has implications for large language model.

This does, and they've been working on that well.

sick. A stall there. You wrote at the best current theory of human language are arguably large language models so .

this has to do with form is the kind of a big theory and um but the reason is argued the best is that IT does the best predicting what's english for instance it's it's incredibly good know is Better than any other theory so but we don't know this is not sort there's .

not have detail so pic like it's not .

you what is like box. But I IT is a theory.

the definition of a theory because this is gigi gigi c backwards, you know a very large number parameter controlling to me. Theory usually requires A A simplicity, right? Well, I don't know.

Maybe i'm just being loose there. I think it's not a great theory, but it's a theory. It's a good theory in one sense that IT covers all the data again. Then you want to say in english, IT doesn't.

So that's why that's how it's arguably the best is that no other theory is as good as a large language modelinia s in predicting exactly what's good and what's bad in english. You know you're thing is a good theory. Well, probably not, you know, because I I want a smaller theater than that. It's too big, I agree.

You could probably construct mechanism by which you can generate a simple explanation or particular language like a set of rules, something like, could you could could generate A A dependency grammar for a language, right? Yeah, you could probably a you will probably just ask .

you about IT. So you know that what I mean, that presumes there there is some evidence for this that that some large language models are mention something like dependancy grammar inside them.

And so there's work from a guy called Chris manning and colleagues over IT um stanford in natural language and they looked at, I don't know how many large language model types, but certainly bird and some others where and and where you do some kind of fancy math to figure out exactly what the what kind of abstractions of representations are going on and they and they were saying that look like dependency structure is is what they are constructing IT doesn't like. So it's actually a very, very good map. So because they are constructing something like that, um does that mean that you know that they're using that for meeting? I mean, probably but we don't know you write .

that the kinds of series of language that analysts are closest to are called construction base theory can explain what construction these series are.

It's just a general theory of language such that um there's a form and a meaning pair for um for a lots of pieces of the language. And so it's primarily usage based is a construction river is trying to deal with the things that people actually say, actually say and actually right and so that's it's a usage based idea.

And what's the constructions of constructions? Either simple words so of like a more females its meaning or a combination of words is basic combinations of words like though the rules. So but it's it's it's unspecified as to what the form of the grammar is er underlying ly and so I I would I would argue that the dependency grammar, maybe the the right form to use for the types of construction grammar. Construction grammar typically um isn't kind of formalized quite and so maybe the formalization a formalization of that IT might be in defenced grammar. Ah I mean I would think so, but I mean it's up to people, other researchers in that area if they agree or not.

So what do you think that large language models understand language? Are they myrick language? I guess the deeper questioners, are they just understanding the surface form? Or do they understand something deeper about the meaning that then generates the form?

I mean, I would argue they're doing the form. They are doing the form of doing really, really well. And are they doing the meaning? No, probably not.

I mean, there's lots of these examples from various groups showing that they can be tricked in all kinds of ways. They really don't understand the the meaning of what's going on. And so there's a lot of examples that he and other groups of given which which show they don't really understand what's going on.

So you know the monty hall problem is this silly problem, right? Where, you know, if you have three door, let's make a deal, is this old game show. And there's three doors and there's a prize behind one and there's some junk prizes behind the other two and you're trying to select one. And if you you know he knows monty, he knows where the target item is. The good thing you know everything is back there and you're supposed he gives you a choice.

You choose one of the three and then he opens one of the doors and it's some junk prize and then the question is, should you trade because the other one and and the answers is, yes, you should trade because he knew which ones you could turn around and so now the odds are two thirds okay? Um and then you just change that a little bit to the large language large of language to seen that that explanation so many times that is just if you change the stories a little bit but IT make IT sound like it's somebody helper on what it's not you just say, oh um there are three doors in one behind them is a good prize. There's two bad doors.

I haven't know it's behind door number one. The good prize. The car is behind dorm number one so i'm going to choose their number one monty hall opens door number three and shows me nothing that should I trade for the other two even though, no, the good prize and donor want in the large language mall say, yes, you should trade because he just goes through the the, the, the form that seen before so many times on these cases where IT, yes, you should trade because, you know, your odds are shifted from one and three now to two hundred and three to being that thing IT doesn't have any way to remember that actually you have one hundred percent probability behind that do not want you know that that's not part of the. Of the the scheme that has seen hundreds and hundreds of times before. And so you can even if you try to explain to IT that is wrong, that they can do that, it'll just keep giving you back the the the problem.

But it's also possible the large language model will be aware of the fact that there is sometimes over a representation of a of a particular kind of formulation, and it's easy to get trick by that. And so you could see if they get larger and larger models be a little bit more skeptical. So you see over representation.

So like you IT just feels like form can training on form can go really far in terms of being able to generate a things that look like the thing, understands deeply the underlying world, world model of the kind of mathematical world, physical world, psychological world that would generate these kinds of sentences. IT just feels like you keeping close to the meaning part easily, food on this kind of stuff, but that humans too. So IT just seems really impressive. How often IT seems like a understands concepts.

I I mean, you don't have to convince me that I I am very, very impressed. But does does do mean you're giving a possible world where maybe someone's going to train some other version such that it'll be somehow abstracting away from types of forms. I I mean, I don't think that happened.

And so what? No, no, no. I am saying that I think when you just look at ana, go to examples and just showing a large number of them where IT doesn't seem to understand yes and easily fall yes, that does not seem like a scientific the data driven like analysis of like how many places is a damn impressive, no terms of meaning and understanding and how many places is easily food and like.

That's not the inference. So I don't want to make that the inference I don't I wouldn't want to make with that in the inference i'm trying to push es just that is IT is IT like humans here. It's probably not like humans here.

It's different. But humans don't make that air if you explain that to them, they're onna make that you know they don't make that error. And so that's something is doing something different from humans they're doing in that case.

what what's the mechanism by which humans figure out that is an air?

I'm just saying the error there is like if I explain to you there's one hundred percent chance that the cars behind this case, this door will you you want to trade, feels like no but this thing will say yes because it's so true that trick is so wound up on the form that is that's an error that a human doesn't make, which is kind of interesting .

less like to make, I should say yeah less the very oh yeah. You're asking, you know, you're asking humans to you're asking a system to understand percent like asking some mathematical concepts. And so like.

look, the places where a large language models are, the form is amazing. So let's go back to nested structure, center meted structures. Okay, if you ask a human to complete those, they can do IT neither kind of a large language model.

They're just like humans in that. If you ask, if I ask a large .

language mode, central, and that's not trained, so they do exactly so. So that is a similarity. So but it's that's not meaning, right? This is form. But when we get into meaning, this is where they get kind of messed up when you started to saying, oh, what's behind the door oh, it's, you know is the thing I want he wants don't mess that up as much. You know here the form is it's just like the form of the match is amazing, is similar without being trained to do that.

I mean, it's trained in the sense that is getting lots of huge data, which is just like human data, but it's not being trained on um you know bad sentences and being told what's bad. I just can't do those. It'll actually say things like those are too hard for me me or something which is kind of interesting actually kind of how does not know that I don't know.

but IT really often doesn't just completely, they get off very often sense stuff this true and sometimes that stuff is natural. And almost always the form is great yeah but it's still very surprising that was really great form and able to generate a lot of things that are true based on was restrained up. Yes, it's not it's not just form that is ating. It's mimicking true statements .

from the .

internet. I I guess lying idea there is that on the internet, truth is over represented yeah versus false with I think .

that's the right here.

So but the fundamental thing is trained on you're saying .

is just form yes. Yeah.

I think so. Ah well, that's a sad if that's to me that's still a little bit of open question. I probably lean agreeing with you, especially now blown my mind that there's a separate module in the brain for the language versus thinking. Maybe there is a fundamental part missing from the large language model approach that lacks the thinking, the reason and capability.

Yeah that's what this group are used. So the the same group, uh federal coast group, has a recent paper arguing exactly that guy called ka moo who's here Austin, texas actually he's an old student might but he's a fact and linguistic texas and he was the first author on that.

This first thing, still to meet an open question, what you are the interesting limits .

of them you know I I don't see any limits to their form. Their form is impressed. Ah yeah yeah.

it's pretty I mean, close to what you said, ability to complete central and beldings yeah.

it's just the same as humans. IT seems the same.

but it's not perfect, right? You should be good.

No, but I want to be like humans. I tried to, I want a model, humans.

but, but we also perfect, is as close to humans. I got IT. yeah. But if you should be able to, if you're not human, you like your superhuman, you should be able to central Better sense, right? I mean that .

the the mechanism is if it's modeling something, I think it's kind of really interesting.

That is really .

interesting. I think it's potentially underlying modelling, something like with the the way the form is processed.

the form of human, the way how and how humans process language.

I think that's plausible.

And how they generate language, process language, general language. So in that sense, that is perfect if you can just linger on the center embedding thing that's hard for alone produce, and that seems really impressed because that's hard for humans to produce. And how does that connect to the thing we've been talking about before, which is the dependency grammar framework, in which view language and the finding that, uh, short dependencies seem to be a universal part of language. So why is IT car to complete center and bettington?

So what I like about dependently grammar is IT makes the cognitive cost associated with long, longer distance connections, very transparent bases there, some there IT. Turns out there is a cost associated with producing in comprehending connections between words which are just not beside each other, the further apart they are, the worse that is the, the and according to, well, we can measure that. And there is a cost associated with that.

Or can you just linger on, what do you mean by cognitive costs?

And how do you measure? Oh, you can measure in a long ways of the simplest is just asking people to say whether you know how good a certain sense we just ask is one way to measure. And you try to like triangulate then across sentences, across structures to try to figure out what the source of that is.

You can look at um reading times in controlled materials, you know in certain kinds of materials when and then we can select measure the dependency distances. There we can there's a recent study which looked at where we're talking about the the brain. Here we can look at the language network, okay, we can look at the language network and we can look at the activation in the language network and how big the activation is depending on the length of the defense.

And turns out into random sentences that you're listening to if you're listened to, turns out there are people in the stories here. Uh, and the bigger the longer the dependence the penny is, the the the stronger the activation in the language in the language network. And so there's some measure, a different one of of different measures we could do that, that kind of a neat measure actually of natural activation, activation in the brain .

so that you can somehow, in different ways, covered to a number. I wonder if there is a beautiful equation connecting the cost equals I A square kind of thing.

Yeah, it's complicated, but probably it's doable. I I would guess it's doable. I try to do that a while ago and I was reasonably successful.

But some for some reason, I stop working on that. I do I agree with you that I would be nice to figure out. So it's like some way to figure out the the, the cost. I mean, it's complicated. Another issue you raised before was acai measure distance is IT words is IT probably isn't. Is the part of the problem is that some words matter than more than others and probably you know meaning like knows might matter depending and then maybe depends on which kind of now is that now we've already introduced or known that already been mentioned as IT a pronouns, sus a name, like, like all these things probably matter. So probably the simplest thing to do, just like I always forget about all that and just think about words for things for sure.

But there might be a kind like there might be some insight in the kind of function yeah yes, that fits the data. Meaning like like what I think .

it's an expensive al. So we think is probably an exponential such that the longer the distance, the less IT matters. And so then it's the sum of those is my that was our best guess a while ago. So you've got a bunch of dependency es, if you've got a bunch of them that are being connected at some point that at at the ends of those the cost is the is some expense title function of those is my guest.

But because reasons is probably an expansion, al is like it's it's not just the distance between two words because I can make a very, very long subject, global dependency by adding lots and lots of non phrases and prepositional phrases and IT doesn't matter too much. It's when you do nested, when I have multiple of these, then things get go really bad. Go self .

probably memory.

Yes, that's probably the function of the memory. Here is the access is trying to find those earlier things. It's kind of hard to figure out what was referred red to earlier those of those connections that that's the sort of notion of working as supposed to historical thing.

But trying to connect, uh, we retrieve those earlier words depending on what was in between. And then. Then we're talking about interference of similar things in between.

That's the right theory, probably has that kind of notion and is an interference of similar. And so i'm dealing with an abstraction over the right theory, which is just you know it's count words. It's not right, but it's close.

And then maybe you right out there is some sort of um an expansive or something on on the to figure out the totals so we can figure out a function for any for every any given sentence in any given language. But you know it's fun. You know people haven't done that too much, which I do think is i'm interested that you find that interesting. I really find that interesting and and a lot of people haven't found that interesting. And I don't know why I haven't got people to want to work on that.

I really like that too. The beauty and the underlying idea is beautiful, that there's a coconuts that corals with the link to dependency, they just IT feels like, because a deepmere language is so fundamental, the human experience. And this is a nice, clean theory of language. Well, yeah, it's like, wow, okay so like, we like our words close together, dependent words close together.

And that's why I like you too. It's so simple.

yeah. The only simple .

and yet IT explains some very complicated phenomenon. If you if I write these very complicated senses, it's kind of hard to know why they're so hard. And you can like or nail IT down, I can do, I can give you a math formula for a why each one of them is bad and wear. And that's kind of cool. I think it's very need.

Have you gone to the process? Is there like you take a piece text and then simplify sort like there's an average uh linked the dependency and then you like uh you know uh reduce IT and see comprehension on the titan. I just single sense like you know you go from James joyce to hemingway .

or something no, no. I simple answers know that there's probably things you can do. And in that kind of direction, it's fun. We might we going to talk about legalism at some point, and may be we will talk about that kind of thinking. We apply to legal ism .

because you mention as an exception, we taking attention upon attention. That's an interesting you give him as an exception is an exception that you say that most natural languages, as we've been talking about, have local dependency es with one exception.

Legal is right.

So what is legal is first all oh well.

legal is you think IT is it's just any other language.

I mean, like I actually know very little about the kind of language of lawers use.

so i'm just about language in laws and language contract. So the stuff that you have to run into, we have to run into every other day, every day and you skip over because IT reads poorly and or you know partly is just long, right? There's a lot of text there that we don't really want to know about.

And so but the thing i'm interested in, so i've been working with um this guy called eric Martinez who is say he was a lawyer who was taking my class. I was teaching a cycle anguish CS lab class and have been teaching for a long time at mt. And he he was a lost harvard, and he took the class because he had done some linguistics as an underground, and he was interested in the problem of why.

Sounds hard to understand, know why. So why is IT hard understand? And why do they write that way? If IT is so heart to understand, IT seems a parent that it's hard to understand.

The question is why is IT? And so we didn't know and um we did uh, an evaluation of buch contracts actually, which I took a bunches of random contracts because I don't know you know there's contracts and laws might not be exactly the same, but contracts are kind of the things that most people have to deal with the most trying. So this kind of the most common thing that humans have, like arguments that that adults in our industrialized society have to deal with a lot.

And so so that's what we pulled and we we didn't know what was hard about that, but IT turns out that the way they're written is is very central. Ted has nested structures in them. So IT has low frequency words as well as, not surprising, lots of texts have lower IT does have surprising a slightly lower frequency words than other kinds of control text, even sort of academic text legal.

This is even worse. That is the worst that we were being. Just you just .

revealed the game that lords are playing.

They are optimizing a different really interesting that's that's now you're getting and why. And so and I don't think it's your thing. It's they are doing intentionally.

I know they're doing intentionally. And yeah, yeah, yeah, look, get to that. Look, get to that. And so we wanted to see why we should see what first as well. So because IT turns out that we're not the first to observe the legal like back to nicks on had a plane language act in by in one thousand nine and seventy and an obama had one and have boy, a lot of these, a lot of her presidents, they said how we've gotten simplify legal language must simplify. But if you don't know how it's complicated, it's not easy to simplify that.

You need to know what that is you're supposed to do before you can fix IT, right? And so you do you need to cycle english to analyze the text and see what's wrong with IT before you can like fix IT. You don't know how to fix that, how I supposed to fix something, I don't know what's wrong with that.

And so what we did was just that we did. We figured out, let's okay, we just a bunch contracts had people uh and we encoded them for the the a bunch of features. And so another features, the people, one of them was the center bedding.

And so uh that is like basically how and a um a clause would intervene between a subject and a verb, for example. That's one kind of a bedding of a clause. okay.

And um turns out there massively center bit like so I think in random contracts and in random laws, I think you get about seventy percent or eighty seven seventy percent sentences have a center Better cause which is insanely high. Know if you go to any other text it's down to twenty percent or something. It's it's so much higher than any control you can think of including you think of people think oh technical um academic text.

No people don't write center and bedded sentences in in technical academic text I mean to do a little bit but much it's on the twenty percent, thirty percent realm as opposed to seventy. And and so there's that and and there's low frequency words and then people oh, maybe it's passive. People don't like the passive passive for some reason the passive voice in english has a bad wrap and i'm not really sure where that comes from. Um and there is a lot of passive in um the there is much more voice in in the legalities .

and there is in other is no, no, no.

No .

drop.

The judgment is just like these are frequent, these are things which happen in legal list text. Then we can ask the dependent measure is like how well you understand those things with those feature? yeah.

okay. And so then and IT turns out the past, that makes no difference. So IT has zero effect on your comprehensive ability, on your recall ability, not did nothing at all.

That makes no effect. The words matter a little bit. They do no frequency. Words are gna hurt you in recall and understanding. But what really, what really hurts the center Better, that kills you, that is like, that slows people down, that makes that make some very poor red understanding, that makes them they can't recall what was said as well, near as well. And we we did this not only on lay people, we did a lot of lay people.

We went on one hundred lawyers, and we recruit lawyers from from a wide range of of sort of different levels of law firms and stuff. And they have the same pattern. So they also like I, when when did this? I did not know how how that maybe they could process.

They they used to legal is say they proceed just as well as IT was Normal. No, no, they they they're much Better than lay people. So they're much they can much Better recall, much Better understanding, but they have the same main effects as as as late as lay people exact the same.

So they also much prefer the non center. So all we constructed non center embed ded versions of each of these reconstructed versions, which have um higher frequency words in those places. And we we did we on passive ed, we turn them into active version. The passive active made no difference, the words made little difference, and the unset ebel ding may makes big differences in all the population.

Unsentient betting. How hard is that process, by the way, don't question, but how hard is IT detect.

center betting detect .

just not dependent or so there's .

automatic parsons for english, which are pretty good and .

they can detect something or I guess you .

so you're not just looking for .

long dependencies, you're just literally looking for something.

Yeah, we are in this case. And but long parents.

they're highly correlated. Center betting is is a big bomb you thrown inside of a sense that just grows up that that makes .

can I read a sense for you from these things? I I mean, this is just like one of the things that this is just my eyes.

my glaze over middle midsentence no, I that I mean legal, so hard they go .

because in the event that any payment are benefit by the company, all such payments and benefits, including the payments and benefits under section three a year of being here, here and after referred to as a total payment, would be subject to the exercise tax, then the cash severance payments shall be reduced. So that is something we pulled from a regular text from from a contract wall and the Better.

But there is just for some reason, there's a definition. They throw the definition of what payments and benefits are in between the subject in the verb, let's how about to do that? How would put the definition somewhere else as opposed to in the middle of the sentence.

And so that's that's very, very common, by the way, that is that's what happens. So you just throw your definitions, you use word, couple words and then you define IT and then you continue to sentenced like just don't write like that and you ask so when we ask lawyers, but that mean be lawyers like this layers don't like this. They don't like this. They don't want, they don't want write like this. We ask them to rate materials which are with the same meaning with with uninterpreted and the, and they much preferred the unbid version on the .

complete on the reading side.

And we asked them, we asked them, would you hire someone who writes like this or this? We ask them all kinds of questions. Always preferred the less complicated version, all of them. So I don't even think they .

want IT this way. yeah. How did that happen? how?

That's a very good question. And the answer is we still don't know, but I have .

some theories.

Well, our best theory the moment is that there's there's actually some kind of a performative meaning in the center embedding in the style which tells you its legal. We think that that's the kind of a style which tells you its legal is like that is a reasonable guess. And maybe it's just so for instance, of year, like it's like a magic spell.

So you can call this and magic spell hypothesis. So when you give them when you tell someone to put a magic spell on some what what did you they you know people know what a magic spell is, and they they do a lot of rimming. You know that's that's kind of what people who will tend to do.

They'll do rimm. They'll like some kind of poetry. And maybe that there is a syntactic sort of reflects here of of a magic spell, which is center bedding.

So that's like, oh, it's trying to like tell you this is like this is something which is true, which is what the goal of law law is right, telling you something that we want you to believe as certainly true, right? That's that's what legal contracts are trying to enforced on you, right? And so maybe that's like A A form which has this is like an abstract, very abstract form sent from bedding, which has A A has a meaning .

associated with that. Well, do you think there's an incentive yeah, for lawyers to generate things that are hard to understand?

That was one of our working hypotheses.

We just couldn't find any evidence of lawyers don't understand. Would that mean you ask in a communist soviet union, the individual members, uh, their software report is not going to correctly reflect what is broken about the gigantic bureaucracy that leads to turnover or something like this. Um I think the incentives under which you Operator, not always transparent to the members within that system.

So IT is feels like a strange coincidence that that there is benefit if you just zoom out, look at system, is a just asking that makes something hard to understand, is going to make a lot of people money yeah like there's going you're gonna need a lawyer, uh, to figure that out, I guess, from the perspective of the individual. But then that could be the performance of us IT IT could be as opposed the incentive driven to be complicated, to be performative to where we lawyers speaking in this sophisticated way and you regular humans don't understand. And so you need to hire lawyers. I don't know which one IT is, but it's suspicious, suspicious that it's hard to understand and the everybody is eyes glaze over and .

they don't read suspicious as well. I'm still suspicious. And here you're saying that could be kind of no individual and even average individuals that could just be a few bad apples in a way which are driving .

the effect in some way fluently. Al, yeah, that really looks up to whatever they're like in central figures.

And you know that turns is IT is kind of interesting among our hundred lawyers. They did not what they really didn't like IT.

So they were Better at than regular people comprehending IT where they were an average Better. But they had the same different.

the exact same different. So they but I they wanted IT fixed. So they they also, and so that that gave us hope that because IT actually isn't very hard to construct A A material which is unset in bad has the same meaning, it's not very hard to do just basically the next situation.

Just putting definitions outside of the subject verb lation in the predictive example, that kind of that's pretty general. What they're doing is just throwing stuff in there, which you didn't have to put in there. There's extra words involved typically um you may need a few extra words sort of to refer to the things that you are defining outside in some way as because if you only use IT in that one sentence then there's no reason to introduce tra extra terms. But so we might have a few more words but it'll be easier to understand. So I may I have hope that now that may maybe we can make legal is less less convoluted.

Maybe the the next present in the next days can in generic things, say exactly I am center on bettins and make to the the language zar of Martinez .

is the guy you should really put in. But .

batting the bad thing to have this, you get rid of that .

they'll do a lot of IT that .

to fix a lot fascine. Yeah, that is so fascinating. And IT is really fascinating, many fronts that humans are just not able to do with this kind of thing.

And that language, because of that involved in the way that faster. So one of the mathematical formulations you have when talking about languages communication is the idea of noisy channels. What's the noisy channel?

So that's our communication. And so this is going back to shanon. So shanon clock.

Shen was a um uh student M I T in the forties, and so he wrote this very influential piece of work about communication theory or information theory. And he was interested in human language. Actually he was trying.

He was interested in this problem of communication, of gaining A A A message from my head to to your head and and so and he he was concerned or interested in um what was a robust way to do that and so that assuming that we both speak the same language, we both already speak english, whatever no whatever languages we speak, that what is a way that I can say the language so that is most likely to get the signal that I want a to you. And so and then the problem there in the communication is the noisy channel is that there's I make there's a lot of noise in the system. I don't speak perfectly.

I make errors. That's noise. There is background noise. You you know that like litter, literal background noise, there is like White noise in the backgrounds are some other kind of noises, some speaking going on that you or just you're the party that background noise, you trying to hear someone, it's hard to understand them because there's always other stuff going on in the background and and then there's noise on the communication on the country, on the receiver side so that you have some problem may be understanding me for stuff this is internal to you in some way.

So you've got some other problems whatever with uh understanding for whatever reasons. Maybe you if you had to wish to drink, you know who knows why you're not able to pay attention to the signal. So that's the noisy channel.

And so so the language, if its communication system, we are trying to optimize in some sense, the passing of the message from one side to the other. And um so IT hurt. I mean, one idea is that maybe.

You know, aspects of like word order, for example, might have optimized in some way to to make language a little more easy to be passed from speaker to listener chance. The guy that did the stuff way back in the four, very interesting there was historically he was interested in working in linguistics. He was in mat, and he, this was his masters thesis of all things know, it's crazy how much how much he did was masters theses in one thousand hundred and forty eight.

I think you're forty thousand something. And and he wanted to keep working in language. And I just wasn't a popular communication as a as a reason source for what language was wasn't popular at the time. So romsey was become IT was moving in there. He he just was not able to get a handle there, I think so and so he moved to bell haps and worked on communication a from a mathematical point point of view and was you uh did all kinds .

of amazing work and so he's just more signal side hi, I would have been interesting to see if you proceed the lingua side yeah .

that's really interesting. Yeah he was interested in that as examples in the forties. Kind of like their very language, like like things. We can kind of show that there's a noisy channel process going on in. When you're listening to me, you can often to guess what I meant by what I what what you think I meant given what I said. Um and I mean, with respect to sort of why language looks the way that does, we might there might be served as eluded might be ways in which word orders is somewhat optimize for for because of the noisy channel in some way.

I think that's really cool to sort of model if you don't hear certain parts of a sentence or have some probability of missing part, how are the idea .

that idea and you can .

saying like the world order and the syntax of language, the dependency length, are all hopeful.

Yeah, well, the pencils link. This is really about memory, really. I think that's like about sort of what easier, harder to in some way.

And these other ideas are about our robust to communication. So the problem of potential loss lost signal due to noise. So there may be aspects of word order, which is somewhat optimized for that.

And know we have this one guess in that diary. These are kind of just so stories I have to be no prety Frank. They're not like I can show this is true.

All we can do is like look at the current languages of the world. This is we can sort of see how languages change or anything because we're got these snapshots of a few know one hundred or a few thousand languages. We don't we don't really.

We can do the right kinds of modifications to test these these things experimentally. And so so just take that, this with the grain soul, okay, from here, this, this stuff that depends the stuff I can, and much more sol IT on. And like, here's what the length are and here and here's what's hard.

Here's what's C, Z. And this is a reasonable structure. I think i'm pretty reasonable.

Here's like why you know why does the word order looked? The way that does is for now into shaky territory. But it's kind of cool .

we're talking about, just to be clear, we talking about maybe just actually the sounds of communication like you and I sydney bar very loud and you you a model with noy channel, the loudness, the noise and we have the signal is coming across though. And you're saying water, water might have something to do, the optimizing that present. I am interesting.

I mean, to me, it's interesting how much you can load into the noisy channel. Like how much can you begin? You said like cardinal load on the receiver.

And we think that those are there three at least three different kinds of things going on there, and we probably don't want to treat them all as the same. And so I think the right model, a Better model of a noisy channel would treat we have three different sort sources of noise which which are background noise, you speaker speaker um inherent noise and listener inherent noise. And those are not those .

are all different things. But underneath that there's a million you like receiving. I mean, I just mentioned coat load up both sides. Then there's like speaking a speech in paramus everything, a world view. I mean in the meaning we started to creeps into the meanwhile m of work, we have different world views.

Have I just form still the like, you know, like, so how well you know the language? And so if its second language for you versus first language and and how maybe what are the languages you know, these are still to form stuff and that's like potentially very informative and and you know how old do you are? These things probably matter, right? So like a child learning a language is is, you know as a noisy representation of english grammar, uh, you know, depending on all they are, thank you. When there are six, they are perfectly formed.

But uh, you mention one of the things is like a way to measure the a languages learning problems. So like what's the correlation which in everything we've been talking about and how easy is to learn al language? So is a like short dependencies corner ability to learn a language.

Is there some kind of or like the defenced grammar, there's some kind of connection there. How easy is to learn yet? Well.

all the languages in the world's language, none is right now, we know, is any Better than any other. With respect to sort of optimizing dependency links, for example, they are all kind of do IT well, they all keep low. So I think of every human language is some kind of an of an optimization problem, a complex optimization problem to this communication problem. And so like you've solved, didn't know they're just serve noisy solutions to this problem of communication. There are just so many ways you can do this.

So they're not optimize for learning .

the and learning. So yes, one of the factors which, yes, so learning is messing this up a bit. And so so for example, if IT worked just about minimizing dependency links and and that was all that matters, and then we so then we might find grammar, which didn't have regularity in their rules, like, but languages always have regularity in their rules.

So so what I mean that is that if I wanted to say something to you in the in the optimal, to say that was really matter to me, all that mattered IT, was keeping the defensive as close together as possible. Then I, then I would have a very lack set of free structure or or dependency rule. I ouldn't have very many of those.

I would have very little that. And I would just put the words as close, the things that refer to, the things that are connected at right beside each other. But we don't do that like there are water, water rules, right? So they are depending on the language, they're more and less strict rates.

You speak dressing, they're less strict than english. English is very rigid world order rules. We order things in a very particular way. And so why do we do that? Like that's probably not about um communication. That's probably about learning mean that we're talking about learning proud easier to learn a regular things, things which are very predictive all and easy to so so that's that's probably about learning is my is our guess because .

I about community can be just noise. Can you be just the the messiness of the development of a language?

Well, if if we're just a communication that we we should have languages which have very, very free word order. We don't have we have free or but not free, like there's always well.

no but what I mean by noise is like cultural like sticky cultural things, like the way the way you communicate just that is a sticky story, that is it's an imperfect it's a noise afternoon is the cash yeah the the function over which you optimizing is very noisy yeah so uh because I don't IT feels weird to say that learning is part of the objective function because some language is a way harder to learn than others, right? Or that that's not true. That's interesting. I mean, that's the public perception.

right? Yes, that's true. The for a second language, second, but that depends on what you started with, right? So so IT really depends on how close second language is to the first language you've got. And so yes, it's very, very hard to learn arabic. C if you started with english or with heartache harder or japanese, if you have started with think of chinese, I think is the worst in the there's like defense language institute in the in the states has like a list of of how hard IT is to .

learn what you from english. Don't care. No.

no evidence that there's anything harder, easier. But any baby, any language learn like three or four, they speak that language. And so there's no evidence of any anything harder, easier. But any human language.

there are all kind equal. To what degree is language? This is a returning to chop scare little bit.

This is a nate. You said that for to sky, he used the idea that language is some aspect. The language need to explain why, certain things that observed. But how much are we born with language at the core of our mind brain? I mean.

you know the answers, I don't know, of course, but the I mean, I like to be an engineer hurt. And yes, and I sort of think it's fine to postulate that a lot of is learned. And so i'm guessing that a lot of is learned.

So I think the reason chomsky went with the nightlife is because he, he, he hypothesized movement in his grammar. He was interested in grammar and movements. Hard to learn.

I think he's right. Movement is a harder, is a hard thing. Learn to learn these two things together and how to interact. And it's like a lot of ways in which you might generate exactly the same sentences and it's like really hurt. And so he's like, oh, I guess it's learned so I guess is not learned ed in ate.

And um if you just throw the movement and just think about that a different way, you know then you get some matinees but the matinees is human language, which is exactly fits Better. That messiness isn't the problem. It's actually uh uh it's a valuable asset of of uh of the theory.

And so so I think I don't really see a reason to populate much and ate structure. And that's I think these large language point are learning so well is because I think you can learn the form, the forms of human language from the input. I think that's like it's likely to be true.

So that part of the brain that lights up when doing all the comprehension that could be learned and .

that could be just, yes, you don't know. So like lots of stuff is. Modular in the brain that learned IT doesn't have to be. So there's something called the visual word form area in the back.

And so it's in the back, your head near the you know the visual cortex, okay? And that is very specialized language, very specialized brain area, which does um visual word processing. If you read, if you read okay, if you don't read you don't have in okay.

Guess what you spend some time learning to read and you do well that that brain area which does exactly that. And so these the modulation is not evidence for an atis. So the martian ation of a language he doesn't mean we're born with that we could have easily learned that we might have been born with that we just don't know at this point.

We we might very well have been born with this left lateralization area. I mean, there's like a lot of other interesting components y the features of this kind of argument. So some people get a stroke or something goes really wrong on the left side where the left, where the language area would be.

And that and that isn't there. It's not not available and IT develops just fine. The right. So it's no it's so it's not about the left.

IT goes to the left like this is a very interesting question is why is the why are any of the brain areas the way that they are? How did they come to be that way? And, uh, there's these natural experiments which happened where people get these no strange events in their brains at very Young ages, which wipe out sections of their brain and and they behave totally Normally, and no one knows anything was wrong.

And we find out later, because they happen to be accidentally scanned for some reason, like what what happened here. Left hemisphere is missing. There is not many people miss their whole left hemisphere, but theyll be missing some other section of their left or right.

And they behave absolutely Normally, would never know. So that's like a very interesting, you know, current research know this is another project that this person infected is working on. She's got all these people contacting her because she's scan some people who have been missing sections.

One person missing missing section of her brain that was scanned her lab and and SHE and SHE haven't be a writer for the new york times. And there was an article in new york times about about a the a just about the scanning procedure, about what might be learned about by side of the general process of MRI in language, in the language. And because she's running for the new your time, and all these people started writing to her. We also have similar, similar kinds of deficits because theyve been, you know, accidentally, you know, to stand for some reason and and found out they are missing some section. And they in this, they vote teer to be scant.

These natural .

experiments at ural experiment.

they're kind of messy, but natural best cool, first few hours, days, months of human life. But inside the room actually like that development, that machinery, whatever that is, seems to create powerful humans. They are able to speak comprehensive, all that kind of stuff no matter what happened, not no matter what robust to the different ways that um um the the brain might be damaged so well that's that's really that's really interesting. But what what would chomsky say about the fact the thing you're saying now that the languages is seems to be happening separate from thought because as far as understand, maybe incorrect. Me, he thought that language under pins.

Yeah, he thinks so. I don't know what you say.

He will be surprised because for him the idea that language, it's a through the foundation of thought.

that's right, absolutely.

And it's pretty. My belongs to think that I could be completely separate from thought.

But so you know he's basically a philosopher, philosopher of language in a way, thinking about these things. It's a fine thought. You can test IT in his methods.

You can't do thought experiment to figure that out. You need a scanner, you need brain damage people. You need something. You need ways to to measure that.

And that's what you know F R I offers is and you know patients er a little messier FM I is pretty ambiguous. I'd say it's like unvitiated. There is no way to say that the language network is doing any of these tasks.

There's like you look at those data, it's like there's no chance that you can say that there those networks are overlapped. They are not overlap ping. They're just like completely different.

And so so no, you can always make it's only two people. It's four people or something for the ate the patients. There's something special about them, we don't know, but these are just random people and and with lot of them and you find always the same effects and it's very .

robust and fast effect. Uh what's the invention believe a um what's the connecting culture and language? You've you've also mentioned that you know much of our study of language comes from W E I R D, weird people, western educated, industrialized, rich and democratic. So when you study like remote cultures such as around the amazon jungle, what can you learn about language?

So that term weird is from joe Henry. He is at harvard. He's a harvard evolutionary biologist, and so he works on lots of different topics and he basically was pushing that observation that we should be careful about the inferences we want to make when we're talking in psychology or so ah most in psychology, I guess about humans if we're talking about you know undergrads and M I T and harvard are the same, right? These aren't the same things.

And so if you want to make inferences about language, for instance, you there is a lot of very a lot of other kinds of languages in the world, then english and french in chinese. And so maybe for a language, we care about how culture, because cultures can be very amen. Of course, english in chinese cultures are very different.

But you know, hundred gathers are much more different in in some ways. And so if culture has an affecting what language is, then we kind of want to look there as well as looking. It's not like the industrialized cultures. Res are interesting, of course they are. But we want to look at non industrialized cultures as well.

And so I I worked with two I work with a chemi which are in um bOlivia and in the amazon, both in the amazon these cases and there are so called farmer forgers, which is not hunter gathers um instead of one up from hunter gathers and that they do a little bit of farming as well, a lot of hunting as well, but a little bit farming in the kind of farming they do is the kind of farming that I might do if I ever were to grow like tomatoes or something in my backyard. This is that it's not like so it's not like big field farming. It's just A A farming fur family.

A few things you do that. So that's what that's the kind of farming they do. Um and the other group I worked with her the Peter hm which are in uh also in the amazon and happened to be in brazil and that's with um I called dan ever IT who was a lingual anthropologist who actually lived and worked in the I mean he was a missionary actually initially back in the seventies working with trying to translate languages so they can teach them the byo teach them Christiani .

what working you say about .

that yeah so the two groups I worked with, the much money in the pitter, have both islet languages, meaning there's no known connected language at all. I just like on their own ah IT is a lot of those. And in most of the isolates occur in in the in the amazon or in papua ini and these places where the world has sort of states still for a long enough and there are hat like so there aren't earthquakes.

There aren't um was still in earthquakes in the members jungle. And and and the climate isn't bad, so you don't have drought. And so you know, in africa, you've got a lot of moving of people because there's drought problems.

And so so they ve got a lot of language contact when you have, when people have to, if you gotta move because you you got no water, then you got get going, and then, uh, then you run into contact with other, other tribes, other groups in the amazon on. That's not the case. And so people can stay there for hundreds and hundreds, probably thousands of years, I guess.

And so these groups have the the chance I and the phar both isolates in that and then I guess they just lived there for ages and ages with minimal contact with other outside groups. Um and so I I mean i'm interested in them because they are I mean in this case, i've interested in their words and I would love to study their syntax, their orders of words. But I mostly just interesting how languages you know are connected to um their their cultures in this way.

And so with the Peter har, so are most interesting. I was was working on number there, a number information. And so the basic idea is, I think languages invented, right? Just so I get from the words here is that I think language is invented.

We talked about color earlier as the same idea, so that what you need to talk about with someone else is what you're going to invent words for. okay? And so we invent labels, four colors that I need now that I, that I can see, but but the things I need to tell you about, so that I can get objects from you, or get you to give me the right object. And I just don't need a word for till or or a word for aco marie.

And in the in the amazon jungle, for the most part because I don't have two things which differ on those colors, so I just don't have and so and so numbers are really are another fascinating source of information here where um you might you know naively I certainly thought that all humans would have words for exact counting uh and the pillow I don't okay, so they don't have any words for even one. There is not a word for one in there in their language. And so there's still no word for two, three or four. So so that kind of blows people's mind yeah, that boy.

mind. That's pretty weird how you can ask. I want two of those.

You just don't. And so that's just not a thing you can possibly ask in the it's not possible that is there is no word for that. So here's how we found this out okay, so so he was thought to be a one to many language.

There are three words for quantifies, for sets but um in the people have thought that those meant one, two and many um but what they really mean a few, some and many many is correct. It's few, some and many and so and so the way we figured this out uh and this is kind of cool, is that um we gave people. We have set of objects.

Okay, these were having the spools of thread does not matter what, they're identical objects. And and I start off here. I just give, you know, give you one of those and say, what's that? We see the european hospital and you tell me what IT is and and then I gave you two and say, what's that and and nothing's changing in the set except for the number.

okay. And then I just ask you to label these things. We just do this for a bunch of different people and and Frankly, it's I did this .

task and it's .

a weird it's a little bit weird. So you they say the word that they thought that we thought was won its few, but for the first one and then maybe they say few, or maybe they say some for the second, and then for the third of the fourth, they start using the word many for the set.

And then five, six, seven, eight, I got the way, ten, and it's always the same word, and they look at me like, i'm stupid because they told me what the word was for six, seven, eight and i'm going to a continue asking them at nine and ten and a guy and they understand that I want to know their language. That's the point of the task, is my pride in their language. And so that's okay.

But that does seem like i'm a little slow because I they already told me what the word for many was five, six, seven. And I keep asking. So it's a little funny to do this task over and over. We do this with that. I called that then was the our translator.

He's the only one who really speaks Peter ha fluently he's a good by lingo um for bunch of languages but also in english and Peter a and then when I called mike, Frank was also a student with me down there and I did these things. And so you do that, okay? And everyone is the same thing they all all all do.

You know, when we ask like ten people and they all do exactly the same labelling for one up, and then we just do the same thing down on a grand m order. Actually, we do some of them up, some of them down first, okay? And so we do.

Instead of one to ten, would you turn down to one? And so so I give him ten, nine and eight. They starts saying the word some.

And then down when you get to, for everyone is saying the word for few, which we thought was one. So it's like IT, the context determined what word, what that quantify they used was. So so it's not a count word.

There are account word. They're just approximate word. And there are going .

to be noise when you interview a bunch of people that what the definition of few, and this can be a thresh hold in the context.

and yeah, I don't know, means that's going to be turned on the context. English too, right? You ask an english person what a few is, I mean, depended completely on the context and that .

might actually be, at first hard to discover yeah cause for a lot of people that jump from one to two will be few, right?

So this is just, yeah that might be there.

I mean, that's fascinating. Fascinating numbers don't present there.

aren't there? And we these other things, well, if if they don't have the words, can they do exact matching kinds of task? Can they even and do those tasks? And and the answer is so yes and no.

And so yes, they can do them. So here's the test that we did. We put out those schools a thread again. okay. So we put like three out here and then um that we give them some objects and those happen to be an inflated red balloons. IT doesn't really matter what they are.

It's just a bunch of exactly the same thing and IT was easy to put down right next to these um um schools of thread OK. And so then I put out three of these and your task was to just put one against each of my three things and they could do that perfectly. So I mean, we actually do that.

This was a very easy test to explain to them, because I have, I would did this with this guy, mike Frank, and he would be my, I be the experiment, or telling him to do this and showing him to do this. And then we just like to do what he did. You'll copy him all we had to I didn't have to speak you how, except or know what copy him like.

Do what he did is like all we had to be able to say and and then they would do that just perfectly and it's always move IT up. We do some sort of random number of items up to ten, and they basically do perfectly on that. They never get that wrong.

I mean, that's not accounting task. great. That is just a match you just put one against that doesn't matter how I don't need to know how many they are there to do that correctly and and they would make mistakes, but very, very few and no more than MIT grads just say like no is low stakes. So you know you make mistakes.

counting not required to complete?

Not at all. okay. And so and and so that's a control. And this guy had gone down there before and said that they couldn't do this task.

But I just don't know what he did wrong there, because they can do this test perfectly well. And you, I can can train my dog to do this task. So I of course, they can do this task. And so you know it's not a hard task.

But the other task that was third more interesting is like, so then when you bunch a test where you need um some way to encode the set, so like we one of us just i've just put A A um um OPEC sheet in front of the of the things I put down, a bunch a set of these things, put no pig sheet down and so you can see them anymore. And I tell you, do the same thing you are doing before, right, you know? And easy if it's two or three is very easy, but if I don't have the words for eight, it's a little harder. Like maybe you know with practice, when one of because .

you have to close for us.

it's easy because we we just count them. It's so easy to count them, but but they don't they can't count them because they don't count. They don't have words for this thing. And so they would do proximate is totally ly fascinating.

So they would get him approximately right after after four or five, you know, because you can basically always get four, right three or four, you that looks that something we can visually see, but but after that, you kind of have it's approximate number. And so then and there was a bunch of task we did and they all failed as I mean failed. They did approximate after five on all those tasks and kind of shows that the words ah you kind of need the words you know to be able to do these these kinds of .

time is a little bit of a chicken and egg there because if you don't have the words that maybe they'll limit you in the kind of like a little baby einstein, there won't be able to come up with the accounting task you know mean like a the ability to count enables you to come up with the interesting things probably. So yes, you develop counting because you need IT. But then once you have counting, you can probably come up with a bunch of different inventions, like, I don't know .

what kind .

of thing they do, matching really well for building purposes, building some kind of hut or something like this. So it's interesting that language is A A limit on what you're able to do.

Yeah, here's language is just is the words here is the words like the words for exact count is the limiting factor here. I just don't .

have a yes and yes, but that's what I mean. The that limit is also limit on the society what they're able to build.

That's going to be true. yeah. So it's probable. I mean, we don't know this is one of those problems with the snapshot of just current languages is that we don't know what causes a culture to discover slash, invent accounting system. But the hypothesis that guess out there is something to do with farming.

So if you have a bunch of goats and uh, you want to keep track of them and you save seventeen goats and you go to bed at night and you go get up in the morning, boy, it's easier to have account system to do that. And I and that's an abstract strache over a said so that they don't have like people often asked me when I talk to me about this kind of work, and I say, wow, these Peter, how do they have kids? Don't have love children.

And I like, they have love children, and they do they. They often have families of three, four, five kids and and they go, then they need the numbers to keep track of their kids. And and I always like ask the person who says they like, do you have children? And the answer always no, because that's then how you keep track of your kids.

You care about their identities. It's very important to me. When I go, I say five children doesn't matter which ah doesn't matters which five.

It's like if you replaced one with someone else, I I would care goat maybe not right. That's the kind of point. It's an abstraction, something that looks very similar to the one.

Wouldn't matter to me probably, but if you care about goats, you gonna a know them actually, individually, also mean, go, if that's the source of food, milk.

But, but, but i'm saying this is an abstraction such that you don't have to care about their identities to do this thing fast. The hope this is not mine from ethnologists are guessing about where words for counting came from is from farming maybe.

yeah. Do you have a sense why universal languages like a toronto have .

not taken off like we have .

all these different languages?

Well, my guesses is the function of a language is to do something in the community. And and I mean, unless there's some function to that language in the community, it's not going to survive, it's not going to be useful. So here's a great example.

So what I like, language death is super common. okay? What language is her dying all around the world and here's here's why they're dying like, yeah I see this is not happening right now and either the thermal or the or the pieh, but IT probably will.

And so there's a neighboring group called most on, which is I said that it's a isle is actually there's a dual, there's two of them. okay. So it's just two languages which are really close, which are most one and um and chemi which are unrelated to anything else and mostly is unlike chemi and that IT has a lot of contact with spanish.

And it's dying so that language is dying. The reason is dying is there's not a lot of value for the local people in there native language. So there's much more value in doing spanish like because they want to feed their families.

And how do you feed your family? Learn spanish so you can make money, so you can get a job and do these things, and then you can, then you make money. And so they want spanish things they want.

And so so most of time is is in danger and is dying and that's Normal. And so basically the problem is that people the reason we learn languages to pick and we need we use IT to to make money and to do whatever IT is to feed their families. And if that's not happening um then IT won't take off.

It's not like a game or something. This is like something we you like. Why is english so popular? It's it's not because it's an easy language to learn. Maybe IT is I don't really know, but doesn't not why it's popular.

but because of the united states of gigantic economy.

Therefore, economies do this. It's all about money. And that so is a motivation to learn and Mandarin, there's a motivation to a spanish, there's a motivation to learn english.

These languages are very valuable to know because there's so, so many speakers all over the world, there's lesson of value economically. It's like kind of what drives this. It's not it's not a you not just for fun.

There are these groups that do want to learn language just for the language of sake and they want and then there's something you know to that. But those are rare. Those are rarities in general. Those are few small groups that do that. Not most people don't do that.

Well, that was a primary drive. Other than everybody was speaking english or speaking one language, there's also attention that, well.

languages are .

you are right. Maybe maybe this is slow, but that's what we're moving. But there is attention.

You seeing a language, the fringes. But if you look at geopolitics and superpowers, IT does seem that there's another thing. Attention, which is a language, is a national identity.

Sometimes are you certain nation? I mean, that's the the world ukraine language, ukrainian languages, a symbol of that war in many ways, like a country finding for its own identity. So it's not merely the convenience.

I think those two things are attention is the the communities of trade and the economics and be able to uh communicate with neighboring countries and the trade more efficiently with you. In I I D, the group completely agree as languages is the way for every community. Like dialects that emerge are kind of identity for people, yes, sometimes away for people to say, F, U, to the more powerful, yeah, people is interesting. So in that way, language can be used as that all.

yeah. I completely agree. And there's a lot of work to try to create that identity. So people want to do that speak, you know, as a cognitive ciencias in language expert, I hope that continues because I don't want languages to I want languages to survive because I um because there is so interesting for for so many reasons. But I mean I find the faster just for the language part, but I think the there's a lot of connections to culture as well.

which is also very important to have hope, uh, for machine translation that can break down the barriers of language. So while all these different diverse languages exist, I guess was many was asking this question. But basically, how hard is to IT to translate in an automated way from one language to another?

There's going to be cases where it's going to be really hard, right? So there are concepts that are in one language and not another, like the most extreme kinds of cases are these cases of number information. So exactly like good luck translating a lot of english in to Peter ha.

It's just impossible. There is no way to do IT because there are no words for these concepts that we're talking about. There is probably the flip side, right?

There's probably stuff in Peter hah, which is gonna hard to translate into english on the other side. And so I just don't know what those concepts are. I mean, you know, the space, the world space is is different from my world space. So I don't know what like that.

The things they talk about, things you know, it's going to have to do with their life is supposed to, you know, my industrial of life, which is gonna different and and so there's going to be problems like that always um you know there's like it's not maybe it's not so bad in the case of some of these spaces and maybe it's gonna harder. Uh, so it's pretty bad in number. It's like, you know extreme i'd say in the number space, you know exact number space, but in the the color dimension, right? So that's not too bad. There's I mean but it's a problem that that you don't have ways to talk about the concept and there .

might be entire concept missing. So to you is more about the space of concept. First is the space of form. Like form you can probably map.

yes, yeah. But so you are talking earlier about to translation and about how translations. There's good and bad translations.

Now we're talking about translations, the form, right? So what makes a writing good, right? It's music and right. It's not just the content, it's how it's written and translating that know that sounds difficult.

So we we should say that there is like I know I hesitate to say meaning, but there's a music anna rid M M to the form when you look at the broad picture like the street, the industry as and told uh or have you know having way because James joy, I can mention there's a beat to IT. There's an edge to IT that like is in the form we can .

probably get measures of those yeah I I don't know. I am optimistic that we could get measures of things and so may be that .

on I would like traction to I mean way is probably the lowest. I would like to see different authors, but the average per sentence dependency like for heavy way, is probably the surest.

That's your sense. It's simple sentences which short yeah, short yeah, yeah yes.

I mean, that's if you have really long senses even if they don't have .

like they can have longer connections.

long N S.

I don't have to. You can have a long, long sentence with a much of local words yeah but but IT is much more likely to have the possibility of long with long senses.

Yeah ah I met a guy named as a asking who who is a lot of stuff when he brilliant works with trees on hair and a bunch of stuff but he was talking to me about communicating with animals he cofounded earth species project where you're trying to find the common language between whales, cows and humans and who's saying that there is a there's a lot of promising work that even though the signals are very different, like the actual um if you have a bedding of the languages, they're actually trying to communicate similar type things and um is something you can comment on that like where is a promise to that in everything? You've seen a different cultures, especially remote cultures, that this is a possibility. You know, they can talk to wales.

I would say yes, I think it's not crazy at all. I think it's quite reasonable. There's the sort of weird view well, odd of you.

I think that to think that human language is somehow special. I mean, IT is maybe IT is uh, we can certain ly do more than any other species. And so and maybe maybe our language system is part of that.

It's possible. But but people do have often talk about how human, like chomsky, in fact, is talk about how human, only human language has this this compositionally thing that he thinks is sort of key in in language. And it's the the problem with that argument is he doesn't speak .

whale and he .

doesn't speak uh co he doesn't speak monkey. You know he like they say things like, well, they're making a bunch of grants and squeaks and and that reasoning is like that's bad reasoning like you know, i'm pretty sure if you asked a whale what we're saying, they say, well, i'm making a bunch weird and so it's like, this is a very odd reasoning to to be making that human language is special because we are the only one to have human language like, well, we don't know what those other we just don't know. We can't talk to them yet. And so there is probably a signal in there and IT might very well be something complicated like human language. I mean, sure, with a small brain in in, in lower in lower species, this probably not very good communication system, but in these higher, higher species where you have, you know, what seems to be, you know, abilities to communicate something, there might very well be a lot more signal there than where, uh, then we might have otherwise thought.

But but also we have a lot of body. Here is somebody formally from my my t area, oxygen, who had my very much has talked a lot about his work done communicating with plants. So like, yes, the signal there is even less than but like it's not out of the real possibility that all nature has a way of communicating and it's a very different language. But they do develop a kind of language through the chemistry uh to some way of communicating with each other. And if you have enough humility about that possibility, I think you can I think it'll be a very interesting in a few decades, maybe centuries, hopefully not um a humble in possibility of being able to communicate I just between humans effectively, but between all of living things on earth.

Well, I mean, I think some of them were not to have much interesting to say. So we don't know. We certainly don't know.

I think I I think for a humble, there could be some interesting trees out there.

Well, well, they're probably talking to other trees, right? They are not talking us. And so to the extent they're talking, they are saying something interesting to some other you you know can't specific as the lost rate.

And so there probably is there maybe some signal there. So there are people out there actually pretty common to say that, that human language is special and different from any other animal communications system. And I I just don't think the evidence is there for that claim. I think it's not obvious um and do we don't know because we don't speak these other communication systems. And until we get a Better you know I do think there there are people working on that, as you pointed out, people working on whale speak, for instance, like that's really fascinating.

Let me ask you a wild out there, safe. I question, if we make contact with an intelligent aliens, civilization, and you get to meet them, how hard do you think? Like how surprised you be about their way of communicating? Do you think you would be recognizable. Maybe there are some parallels here when you go to the remote tries.

I mean, I would want then ever with me he is like amazing at learning, uh, foreign languages. And so heat like this is amazing feet, right? To be able to go this a language to piao, which has no translators before him.

I mean, there were, he was really like that. Well, there was a guy that had been there before. He wasn't very good.

And so he learned the language far Better than anyone else had learned before him. He's like good at he's just he's a very social person. I think that's a big part of IT is being able to interact. So I don't know what kind depends on these these the species is from space, how how much they want to talk to us.

There's something to say about the process he follows like what how do you show up to a tribe and socialize I mean, I guess colors and counting as as one of the most music thing to figure out.

Yeah you start that you actually start with like objects yes, and just say you just throw stick down and say stick and and you say when you call this, do the future and then they'll say the word whatever and he says a standard thing to do is to throw two sticks and two and then he learned pretty quick that there weren't any count towards this language because they didn't know this was an interesting, he was kind of weird theyd say someone or something the same.

We're over over again. And so but that is a standard thing. You just like try to but you have to be pretty out there socially like willing to talk to random people, which these are a really very different people from you. And he was he he is very social. And so I think that's a big part of this is like that's how you know a lot of people know a lot of languages that they're willing to talk to the people.

That's a tough one. We just show up knowing nothing. Yeah oh so beautiful that human will connect in that way yeah you've had an incredible career expLoring this fascinating topic. What advice would you give to Young people um about how to have a career like that or a life that that they can be proud of? When you see .

something interesting, just go and do IT like I do. I do that like that, something I do, which is kind of unusual for most people. So like when I saw the people, and like Peter hu was available to go and visit us, like, yes, yes, i'll go.

And then when we could not go back, we had some trouble with the brazilian government, is some corruption. There was very difficult to get go back in there. And so I was like her, I got to find another group.

And so we searched around, and we were to find the change because I wanted to keep working on this kind of problem. And so we found the tremont just go there. I didn't really have.

We don't have. Can we have little, little contact and brought someone and and that was, you know, we just you just kind of just try things. I say it's like a lot that's just like ambition. Just try to do something that other people haven't done. Just give IT a shot is what I I mean.

I do that all time. I I love IT, and I love the fact that your pursuit of fun has landed you here talking to me. This was an incredible conversation that you're are a you just the faster human being. Thank you for taking a journey through human language with me today.

This is a some thank you very much like x pleasure.

Thanks for listening to the conversation with award gibson to support this podcast, please check out our sponsors in the description and now let me leave you some words from we can stand the limits of my language, mean the limits of my world. Thank you for listening, and hope to see you next time.