Latest / Wellness, Actually with Emily Oster & Perry Wilson, MD / Learning to Speak: Understanding the babbling black box

Transcript

Learning to Speak: Understanding the babbling black box · Wellness, Actually with Emily Oster & Perry Wilson, MD

• 0:00
  Speaker 1: This is parent data. I'm Emily ost.
• 0:06
  Speaker 2: Hm hmmmm mm hmmm, hig duck ba Mama. The book
• 0:33
  Speaker 2: Hi j k N then a peak to wos w
• 0:40
  Speaker 2: x Y and c why.
• 0:49
  Speaker 3: When I was a baby, my mother was at a
• 0:52
  Speaker 3: party and she met a woman named Catherine Nelson who
• 0:55
  Speaker 3: was an expert in child language development. And as my
• 1:00
  Speaker 3: mom used to tell it, she explained to Catherine that
• 1:03
  Speaker 3: I like to talk to myself in the crib more
• 1:06
  Speaker 3: or less NonStop.
• 1:08
  Speaker 1: And Catherine asked her, what would you be willing to
• 1:10
  Speaker 1: record that? And so my mom.
• 1:12
  Speaker 3: Recorded with a tape recorder under my bed me for
• 1:16
  Speaker 3: two years, almost every night when I talk to myself.
• 1:20
  Speaker 3: And she turned all of these tape recorders over to
• 1:23
  Speaker 3: Catherine and a bunch of researchers use them to try
• 1:27
  Speaker 3: to understand how kids develop language. And I remember when
• 1:33
  Speaker 3: I was nine getting the book that resulted from this
• 1:36
  Speaker 3: research and looking through it and thinking, who would read this?
• 1:39
  Speaker 1: It's incredibly boring.
• 1:42
  Speaker 3: But as an adult, as a parent myself, I have
• 1:44
  Speaker 3: gone back to read that book and I no longer
• 1:46
  Speaker 3: find it boring. Not because I'm so interested in my
• 1:50
  Speaker 3: own babbling, although who isn't, but because the question of
• 1:55
  Speaker 3: how our kids develop language as parents is so fascinating
• 2:00
  Speaker 3: because it's so hard to understand.
• 2:03
  Speaker 1: At one moment, your kid.
• 2:05
  Speaker 3: Is just babbling, and then all of a sudden, it
• 2:09
  Speaker 3: feels like you wake up two weeks later and they're talking.
• 2:12
  Speaker 3: And those first moments of talking in a way you
• 2:15
  Speaker 3: can understand are so powerful because now instead of screaming
• 2:19
  Speaker 3: and throwing things, they can say milk and you can understand, ah, milk,
• 2:26
  Speaker 3: I should have guessed. So child language development is exciting,
• 2:32
  Speaker 3: is interesting.
• 2:33
  Speaker 1: It can also be fraught.
• 2:34
  Speaker 3: It is one of the first things that gives us
• 2:36
  Speaker 3: a moment to say, was my kid behind? Or they
• 2:38
  Speaker 3: have enough words? What is enough words? How do I
• 2:41
  Speaker 3: understand whether that even is a word? My guest today,
• 2:46
  Speaker 3: Michael Frank researches language development. He's a professor at Stanford
• 2:51
  Speaker 3: and he studies how kids learn to speak and how
• 2:56
  Speaker 3: we can understand how kids learn to speak. And we're
• 2:59
  Speaker 3: going to talk today about what's cool about studying this,
• 3:03
  Speaker 3: what we know about it, how we should think about
• 3:05
  Speaker 3: the difference between understanding and speaking, and think about when
• 3:10
  Speaker 3: you should be worried and should you be worried, and
• 3:12
  Speaker 3: how to think about the stresses that come with us
• 3:16
  Speaker 3: after the break Professor Michael Frank. So, I'm delighted to
• 3:24
  Speaker 3: welcome Stanford's psychology professor Michael Frank to the podcast.
• 3:27
  Speaker 1: Mike, thanks for joining.
• 3:28
  Speaker 4: Me absolutely, thanks for having me so to.
• 3:31
  Speaker 3: Introduce people to you, I actually want to read language
• 3:34
  Speaker 3: from your website because I think it's the simplest way
• 3:36
  Speaker 3: to explain why I'm excited about this conversation. So here's
• 3:40
  Speaker 3: from your website, how do we learn to communicate using language?
• 3:44
  Speaker 3: I that's you study children's language learning and how it
• 3:48
  Speaker 3: interacts with their developing understanding of the social world. I'm
• 3:51
  Speaker 3: interested in bringing larger data sets to bear on these
• 3:53
  Speaker 3: questions and use a wide variety of methods, including I tracking,
• 3:57
  Speaker 3: tablet experiments, and computational models. The work in my lab
• 4:00
  Speaker 3: has focused on data oriented approaches to development. So for
• 4:04
  Speaker 3: parent data, you've pretty much just put together our favorite keywords.
• 4:08
  Speaker 3: There's data, big data, children, methods, development, it's the best,
• 4:12
  Speaker 3: And so I was I was struggling with it where
• 4:14
  Speaker 3: to start, And we're going to try to touch on
• 4:16
  Speaker 3: a bunch of those pieces, but I think we're going
• 4:18
  Speaker 3: to start with language development. So without asking you to
• 4:22
  Speaker 3: teach me like five semesters worth of PhD classes. Can
• 4:26
  Speaker 3: I just ask you to talk a little bit about
• 4:28
  Speaker 3: this very basic question of like, how does my kid
• 4:31
  Speaker 3: learn to talk?
• 4:32
  Speaker 1: Because it seems like magic.
• 4:34
  Speaker 4: Yeah, I mean I think it seems like magic to
• 4:36
  Speaker 4: a lot of us. Actually, no, it's really amazing. Right,
• 4:39
  Speaker 4: So you have this trajectory over not that many years
• 4:43
  Speaker 4: where you go from just not being able to talk
• 4:45
  Speaker 4: to being able to do pretty well at making your
• 4:48
  Speaker 4: way through the world and conversation. So back in the day,
• 4:52
  Speaker 4: the starting hypothesis was I say things to my kid,
• 4:55
  Speaker 4: they repeat it back, they imitate. And that's in some
• 4:59
  Speaker 4: sense not a terrible way to start thinking about language development,
• 5:02
  Speaker 4: because you know, in different languages around the world, there's
• 5:05
  Speaker 4: a different word for dog, and so you've got to
• 5:08
  Speaker 4: learn that word by hearing it and hearing it matched
• 5:11
  Speaker 4: with something in the world, ideally like a dog. What
• 5:14
  Speaker 4: if you were a stuffed dog or a picture of
• 5:15
  Speaker 4: a dog in a book or so forth. But language
• 5:18
  Speaker 4: is more than imitation. So you're hearing these things, but
• 5:21
  Speaker 4: you're also hearing how they go together in sentences. And
• 5:24
  Speaker 4: it's actually very very quick to see kids starting maybe
• 5:28
  Speaker 4: even eighteen months two years to combine words in ways
• 5:31
  Speaker 4: that they haven't heard before, so starting to generalize and
• 5:34
  Speaker 4: produce new forms. And when that gets super obvious to
• 5:37
  Speaker 4: parents is when they start to make errors, especially when
• 5:40
  Speaker 4: they say, like, oh, I goed to the store like this.
• 5:43
  Speaker 4: That's them combining that ed from you know, other verbs
• 5:47
  Speaker 4: that they know with a verb that it doesn't go with.
• 5:52
  Speaker 4: So basically some combination of learning from the environment from
• 5:56
  Speaker 4: the people around them to get those word forms and
• 5:57
  Speaker 4: then starting to combine them into novel expressions, trying to
• 6:01
  Speaker 4: make new sentences and get their message across in whatever
• 6:04
  Speaker 4: way they can.
• 6:05
  Speaker 3: How much we sort of dialed into the like brain
• 6:08
  Speaker 3: part of this, how much do we understand about like
• 6:12
  Speaker 3: I don't know the neuronal connections of how that happens,
• 6:15
  Speaker 3: Like do we actually know what part of the brain
• 6:18
  Speaker 3: is turning on or off when we're doing that or
• 6:20
  Speaker 3: is that still kind of a black box?
• 6:22
  Speaker 4: I would say basically it's still a black box in
• 6:25
  Speaker 4: terms of kids. So when we're studying adults, there have
• 6:28
  Speaker 4: been a bunch of really nice studies recently that have
• 6:32
  Speaker 4: really uncovered some of the neural structures that are are
• 6:35
  Speaker 4: supporting language. I mean, there's there's a long history here,
• 6:38
  Speaker 4: of course, dating back to you know, people looking at
• 6:40
  Speaker 4: folks with brain injuries. But there's been some.
• 6:43
  Speaker 3: Exciting Phineas Gauge, right, I have my younger kid is
• 6:46
  Speaker 3: named Phineas, So this is sometimes people with sayle like
• 6:49
  Speaker 3: Phineas Gauge.
• 6:50
  Speaker 4: Well, well he didn't lose language. He lost according to
• 6:52
  Speaker 4: the Resi version self control. But you know, I'm broken
• 6:56
  Speaker 4: and Wernickey, these neurologists back in the nineteen sex we're
• 7:00
  Speaker 4: looking at, you know, what happened to language when you
• 7:04
  Speaker 4: had brain injuries. But it's really modern neuroscience with decoding
• 7:09
  Speaker 4: methods and even computational models that's starting to give us
• 7:12
  Speaker 4: insight into how language is working in the adult brain.
• 7:15
  Speaker 4: But there's lots of challenges when you try to scale
• 7:17
  Speaker 4: that to the kid's brain, most notably that our best
• 7:20
  Speaker 4: technique for this is functional MRI, which requires you to
• 7:23
  Speaker 4: sit still for many minutes at a time, which is
• 7:26
  Speaker 4: not a strength of toddler's learning language. So we're actually really,
• 7:30
  Speaker 4: you know, I think in the next couple of years,
• 7:33
  Speaker 4: going to learn a lot as folks like some of
• 7:35
  Speaker 4: my colleagues here at Stanford persuade toddlers to sit still
• 7:39
  Speaker 4: for just a little bit longer and get a little
• 7:41
  Speaker 4: bit faster at functional MRI scanning, and maybe we start
• 7:44
  Speaker 4: to apply some of those insights from the adults to
• 7:46
  Speaker 4: the kids. But right now, the brain data on kids
• 7:49
  Speaker 4: learning language is actually much more limited than you might expect.
• 7:53
  Speaker 3: So when my kid is little, I mean, this is
• 7:55
  Speaker 3: something I think parents wonder, is like when does something
• 7:57
  Speaker 3: become language? Right, so your kids starts, they're making the noises,
• 8:02
  Speaker 3: the bu mum, you know, the noises, and then slowly
• 8:07
  Speaker 3: it turns into something which maybe it seems to them
• 8:11
  Speaker 3: like a word, but certainly isn't something that hears.
• 8:15
  Speaker 1: Should we view those as words? Or is this all
• 8:17
  Speaker 1: sort of a vague?
• 8:19
  Speaker 3: Is there a moment when it's like now they understand
• 8:21
  Speaker 3: the idea of the mapping or is that just a
• 8:22
  Speaker 3: very slow process.
• 8:25
  Speaker 5: Yeah.
• 8:25
  Speaker 4: Part of the amazing thing about languages, it does have
• 8:28
  Speaker 4: this gradual emergence at the beginning. So my daughter Madeline,
• 8:32
  Speaker 4: around nine months she said something like bra bra, which
• 8:35
  Speaker 4: I think meant brown bear, you know from that classic
• 8:38
  Speaker 4: brown Bear brown Bear when you see book. And I
• 8:41
  Speaker 4: was like, oh my god, it's the first word. We're here,
• 8:42
  Speaker 4: We're in communication. It's language is great, and then it
• 8:45
  Speaker 4: went away and we didn't hear anything else told about
• 8:48
  Speaker 4: thirteen months, so that was a kind of a long
• 8:50
  Speaker 4: dry spell. And first words are kind of frustrating like
• 8:53
  Speaker 4: that they come and they go. They're sometimes communicative, sometimes
• 8:57
  Speaker 4: they just feel like you're trying things out. Way that
• 9:00
  Speaker 4: my father in law, who's actually a speech pathologist, describes
• 9:03
  Speaker 4: this process is that kids are playing the instrument. They're
• 9:06
  Speaker 4: just trying stuff out with their voices. And you know,
• 9:10
  Speaker 4: you actually see this with signing kids too. They'll babble
• 9:12
  Speaker 4: with their hands, they'll try out different gestures. But I
• 9:15
  Speaker 4: think the reason here that this is a complicated question
• 9:18
  Speaker 4: is because there's no language, not language distinction. It's all
• 9:22
  Speaker 4: communication all the way down. That's my viewpoint is even
• 9:26
  Speaker 4: very very early on, kids are really communicating to other people.
• 9:30
  Speaker 4: And you see that actually well before language and most
• 9:32
  Speaker 4: kids because they start to point, and that emergence of
• 9:35
  Speaker 4: pointing is a huge milestone. It's really I want to
• 9:38
  Speaker 4: share this external thing with you. And a lot of
• 9:41
  Speaker 4: people have argued that that's pretty unique to humans. That's
• 9:44
  Speaker 4: that big declarative, like let's share this thing that's outside
• 9:49
  Speaker 4: of the two of us. But let's share it between
• 9:51
  Speaker 4: the two of us, that triadic idea, that that's human
• 9:55
  Speaker 4: unique and really predictive of later language. It's really a
• 9:58
  Speaker 4: powerful moment. So I actually remember that moment with both
• 10:00
  Speaker 4: of my kids that like, fingers out, we're looking at
• 10:03
  Speaker 4: something else and you want to tell me about it,
• 10:05
  Speaker 4: and that's really special. So it's like, you know, walking
• 10:08
  Speaker 4: down the street in New York City with my son
• 10:10
  Speaker 4: Jonah when he was about ten eleven months old at
• 10:13
  Speaker 4: Thanksgiving visiting my parents, and he was like dah ah
• 10:16
  Speaker 4: and I was like, yeah, that's that's the moment.
• 10:19
  Speaker 1: Yeah.
• 10:19
  Speaker 3: My mother in law and sis that my husband said
• 10:21
  Speaker 3: the word fishy when he was six months old, and
• 10:24
  Speaker 3: that like that was his first word.
• 10:25
  Speaker 1: I mean, I've sort of always thought about that.
• 10:27
  Speaker 3: As just a thing she completely fabricated, which is almost
• 10:30
  Speaker 3: certainly true, but maybe maybe not.
• 10:34
  Speaker 4: You know, my grandparents say that my first word was
• 10:36
  Speaker 4: like rhinoceros, which is hilarious for so many reasons, one
• 10:39
  Speaker 4: of which is that the sounds and rhinoceros are actually
• 10:42
  Speaker 4: extremely hard to say, right, but also how would you
• 10:46
  Speaker 4: have been exposed to so first words are really notoriously difficult.
• 10:52
  Speaker 4: It is very interesting though there's a bunch of work
• 10:54
  Speaker 4: with more sensitive measurements that has shown that six month olds,
• 10:57
  Speaker 4: seven month olds, nine month olds do know a teen
• 11:00
  Speaker 4: little bit about language. So there are detectable traces of
• 11:04
  Speaker 4: understanding a little bit of words under there, under the surface.
• 11:07
  Speaker 4: So that's like what that means is like if you
• 11:09
  Speaker 4: gave them a choice between ball and dog, when you
• 11:12
  Speaker 4: say dog, they're gonna look at the dog, like, uh,
• 11:16
  Speaker 4: fifty three percent of the time.
• 11:18
  Speaker 3: This is these eye tracking experiments right where you watch
• 11:20
  Speaker 3: where they're looking to see whether they understand what you're talking.
• 11:23
  Speaker 4: About exactly, and you can find the first traces of
• 11:27
  Speaker 4: success in that early period, but it's not really until
• 11:31
  Speaker 4: about the first birthday where language comprehension really kind of
• 11:34
  Speaker 4: leaps above the surface and you start to see, oh, yeah,
• 11:37
  Speaker 4: you get this. You're understanding what I'm saying in some
• 11:39
  Speaker 4: kind of more consistent way.
• 11:41
  Speaker 3: So that's actually good segue to what I think is
• 11:44
  Speaker 3: there's the part of this it's so exciting that it's
• 11:47
  Speaker 3: just so cool, but then there's a part of it
• 11:49
  Speaker 3: that's really scary because your kid isn't doing the thing
• 11:53
  Speaker 3: that you thought. They're not pointing, they're not making the words.
• 11:56
  Speaker 3: They're not developing in the way that we expect. And
• 11:59
  Speaker 3: I get a lot of quotes about this, and I
• 12:01
  Speaker 3: is just source of a source of tremendous stress for parents,
• 12:05
  Speaker 3: and so I'm not even sure I have a question
• 12:08
  Speaker 3: here as much as you know, can you comment on
• 12:11
  Speaker 3: that topic?
• 12:13
  Speaker 4: Yeah, it's totally stressful. So I think there's there's a
• 12:16
  Speaker 4: couple things to say here. The first is that language
• 12:20
  Speaker 4: is really variable. Early language is scary and frustrating and
• 12:23
  Speaker 4: complicated for parents, in part because the variability is very high.
• 12:28
  Speaker 4: So most kids in the US and actually around the
• 12:31
  Speaker 4: world walk within a month or two of their first birthday.
• 12:35
  Speaker 4: There's some variation, but it's it's not so big. But
• 12:37
  Speaker 4: if you take a milestone for early language, like saying
• 12:40
  Speaker 4: fifty words or one hundred words or combining words, it
• 12:43
  Speaker 4: varies by a lot of months. And that's within typically developing,
• 12:47
  Speaker 4: healthy kids who are going to have good language outcomes.
• 12:50
  Speaker 4: So if there's anything that I've you know, kind of
• 12:52
  Speaker 4: discovered from digging into data with lots and lots of
• 12:55
  Speaker 4: kids that was I think not really appreciated or known,
• 12:59
  Speaker 4: it's just how much the variation between kids is actually
• 13:02
  Speaker 4: consistent or universal across the samples we look at. No
• 13:06
  Speaker 4: matter where you're looking in the world, kids are all
• 13:08
  Speaker 4: over the place with respect to language.
• 13:10
  Speaker 3: So we're talking about like, when we look at sort
• 13:12
  Speaker 3: of a typically developing eighteen month old, you can have
• 13:15
  Speaker 3: these ranges between you know, a few words, maybe almost
• 13:18
  Speaker 3: no words, up to kids who kind of seem I mean,
• 13:22
  Speaker 3: I had a friend who's kid at that age. We
• 13:24
  Speaker 3: showed up at a at.
• 13:26
  Speaker 1: Like a restaurant, and she was like, look, there are umbrellas.
• 13:29
  Speaker 1: I was like, what are you kidding me?
• 13:32
  Speaker 4: No, exactly right. It's really really remarkable how different you
• 13:37
  Speaker 4: walk into a preschool class, how different these language abilities
• 13:41
  Speaker 4: are across the class. So that's within the typically developing range.
• 13:45
  Speaker 4: And that's very anxiety provoking for parents who always want
• 13:47
  Speaker 4: to be at the very top of the range, or
• 13:49
  Speaker 4: or worry that it.
• 13:50
  Speaker 3: Is predictive, right, I mean, I think it's sort of
• 13:53
  Speaker 3: The worry is is in some ways not so much
• 13:56
  Speaker 3: that people are worried that their kid won't learn to talk,
• 13:59
  Speaker 3: but they're something that this is denoting that is going
• 14:04
  Speaker 3: to be true in.
• 14:05
  Speaker 1: The long term.
• 14:06
  Speaker 4: So one really important thing about that is that when
• 14:09
  Speaker 4: you talk as a little kid is not just related
• 14:12
  Speaker 4: to how much you understand, but also how easy is
• 14:15
  Speaker 4: it is for you to move your mouth in a
• 14:17
  Speaker 4: particular way and move your you know, your entire articulatory apparatus,
• 14:22
  Speaker 4: Your vocal cords need to hum and buzz the right ways,
• 14:25
  Speaker 4: you need to push air through your entire system. So
• 14:30
  Speaker 4: it's there's a lot of coordination that has to happen
• 14:33
  Speaker 4: in order for talking, and especially for combining words and
• 14:36
  Speaker 4: too complex sentences to happen right. Evolutionarily speaking, there's a
• 14:41
  Speaker 4: lot of mechanics that human beings have that other primates don't,
• 14:45
  Speaker 4: for using your vocal apparatus and your breathing and your
• 14:49
  Speaker 4: vocal cords and your articulators so flexibly to be able
• 14:52
  Speaker 4: to do this incredibly fast dance that we do when
• 14:54
  Speaker 4: we talk and so like. For example, my son Jonah
• 14:57
  Speaker 4: actually was by many standards what's called a late talker.
• 15:01
  Speaker 4: His productive vocabulary was relatively low at age two. I
• 15:06
  Speaker 4: wasn't worried at all about him. Well, I wasn't worried
• 15:08
  Speaker 4: very much about him. How can you? You are a parent,
• 15:10
  Speaker 4: so yeah, gotta worry, But I wasn't really very worried
• 15:15
  Speaker 4: from a rational perspective, because I knew he understood a
• 15:17
  Speaker 4: tremendous amount language was working, but moving his mouth in
• 15:22
  Speaker 4: this particular way was pretty hard. And his early vocabulary
• 15:26
  Speaker 4: was hilarious. He was like, had some numbers, some abstract
• 15:29
  Speaker 4: you know, colors, you know, a bunch of kind of abstract,
• 15:32
  Speaker 4: complicated stuff that made me feel like, Okay, your cognition
• 15:35
  Speaker 4: is fine. But this whole moving the mouth thing and
• 15:37
  Speaker 4: lining that up with the vocal cords, that was really
• 15:39
  Speaker 4: pretty tough for him.
• 15:41
  Speaker 3: It's interesting because I think you're Another perhaps simpler way
• 15:44
  Speaker 3: to say this is just that we think of productive
• 15:48
  Speaker 3: language as a metric of cognition, as a sort of
• 15:52
  Speaker 3: pure like it is your brain that is that is
• 15:55
  Speaker 3: doing this, and in fact it is a combination of
• 15:58
  Speaker 3: cognition and something that's a more basic physical milestone, which
• 16:02
  Speaker 3: is the production and the way. We don't think about
• 16:04
  Speaker 3: walking as of something that is about, you know, your
• 16:08
  Speaker 3: cognitive development. We just think about it as a physical thing.
• 16:11
  Speaker 3: But there's this physical component to this as well.
• 16:14
  Speaker 4: That's exactly right. And so when you go to the
• 16:17
  Speaker 4: pediatrician or the speech pathologist when your child at age
• 16:19
  Speaker 4: two two and a half isn't saying that many words.
• 16:22
  Speaker 4: One of the first things they look into is going
• 16:24
  Speaker 4: to be, well, are they having trouble understanding language as well,
• 16:27
  Speaker 4: or is there understanding relatively intact, but there's something speech
• 16:31
  Speaker 4: motor that's giving them difficulty. So there is huge variation.
• 16:36
  Speaker 4: That's not to say that there shouldn't sometimes worry. There
• 16:39
  Speaker 4: are places where we do need to go to the pediatrician,
• 16:44
  Speaker 4: ask for for all, maybe go to a speech language specialist,
• 16:46
  Speaker 4: try to figure out what's going on. And that's typically
• 16:49
  Speaker 4: when a child is in that third year, there's two
• 16:52
  Speaker 4: two and a half and they're still not saying much,
• 16:54
  Speaker 4: and especially when they're having trouble understanding, or even earlier
• 16:59
  Speaker 4: if if they're really not communicating, so they're not doing
• 17:03
  Speaker 4: those communicative gestures like pointing, they're not responding to their
• 17:08
  Speaker 4: own name consistently. These are the kinds of indicators that
• 17:13
  Speaker 4: we look at where it's good to chat with a
• 17:15
  Speaker 4: professional to try to figure out what the next steps are.
• 17:17
  Speaker 4: But I think what you want to look for for
• 17:19
  Speaker 4: that reassurance is the comprehension, is the understanding and the
• 17:23
  Speaker 4: community and the desire to communicate, Those are going to
• 17:25
  Speaker 4: be your big kind of easy to spot markers. If
• 17:28
  Speaker 4: there's comprehension and desire to communicate, it's probably a speech
• 17:32
  Speaker 4: motor thing. If there's not, you know, you might be
• 17:34
  Speaker 4: looking at either a general language issue or something like
• 17:37
  Speaker 4: an autism spectrum disorder.
• 17:40
  Speaker 3: Yeah, and there is, of course all of these things
• 17:42
  Speaker 3: at top the top on the variation that are predictable,
• 17:44
  Speaker 3: Like if your kid is learning multiple languages, the productive
• 17:48
  Speaker 3: vocabulary in any single language is smaller because there's many languages,
• 17:53
  Speaker 3: and it can be hard for people to perceive their
• 17:56
  Speaker 3: kid expressing as much because of those differences.
• 17:59
  Speaker 4: That's exactly right, and this is one of the biggest
• 18:01
  Speaker 4: challenges that our field is dealing with from a practical
• 18:04
  Speaker 4: perspective now, is that when you have kids growing up
• 18:07
  Speaker 4: in very varied, multi lingual environments. You know, maybe they're
• 18:10
  Speaker 4: learning one language at school and the other at home,
• 18:12
  Speaker 4: Maybe they have one caregiver speaking one language. Maybe they've
• 18:16
  Speaker 4: got a whole distribution of languages and different circumstances. How
• 18:19
  Speaker 4: do you assess whether what looks like a smaller vocabulary
• 18:23
  Speaker 4: or small smaller comprehension ability even in one circumstance, whether
• 18:27
  Speaker 4: that's predictive of an actual language delay. So doing good
• 18:30
  Speaker 4: assessment in that kind of case is something that we're
• 18:33
  Speaker 4: working on and a bunch of other groups around the
• 18:34
  Speaker 4: world are working on because it's so critical to find out.
• 18:38
  Speaker 4: Is this child experiencing over a language delay where there
• 18:41
  Speaker 4: might be an intervention that we could do, or is
• 18:44
  Speaker 4: this just a natural consequence of Hey, like, there's a
• 18:46
  Speaker 4: lot of language input from different sources coming in. They're
• 18:48
  Speaker 4: different languages. I need to differentiate them. I need to,
• 18:51
  Speaker 4: you know, learn enough vocabulary to work with each independently.
• 18:55
  Speaker 3: So let me ask one other practical question before we
• 18:57
  Speaker 3: talk about data, which is one of the things parents
• 18:59
  Speaker 3: here a lot is you have to talk to your
• 19:01
  Speaker 3: kid all the time.
• 19:02
  Speaker 1: That the way to get your kid to.
• 19:03
  Speaker 3: Learn language is to narrate your entire life, beginning from birth.
• 19:07
  Speaker 3: You know, mommy is taking your diaper off. Now, oh,
• 19:09
  Speaker 3: you had such a big poopy in your diaper, Now
• 19:12
  Speaker 3: mommy is putting your diaper back on.
• 19:14
  Speaker 1: Et cetera.
• 19:15
  Speaker 3: That your whole life needs to be narrated. Then that
• 19:17
  Speaker 3: is how you get your kid to learn to talk.
• 19:20
  Speaker 1: Is that true?
• 19:20
  Speaker 4: Please tell me now. So that's that's the most extreme
• 19:25
  Speaker 4: version of it. So where that comes from, I'll tell
• 19:27
  Speaker 4: you you know, this is sort of a maybe a
• 19:29
  Speaker 4: good data story, right. Where it comes from is that
• 19:31
  Speaker 4: there have been studies looking at relations between kids language input,
• 19:36
  Speaker 4: and that could be how much they hear at home,
• 19:39
  Speaker 4: both in terms of the quantity and also the quality
• 19:41
  Speaker 4: of the input. And then there's a correlation with some contemporaneous,
• 19:45
  Speaker 4: some you know, outcome measure like how much they speak,
• 19:48
  Speaker 4: or maybe some later outcome measure like you know, how
• 19:52
  Speaker 4: big their vocabulary is, or maybe even how they do
• 19:54
  Speaker 4: eventually when they get to school. And so those correlations
• 19:57
  Speaker 4: do exist, their a distant size, in part because any
• 20:01
  Speaker 4: correlation between you know, parent and child has a lot
• 20:04
  Speaker 4: going on. There's measurement error, so forth. Then there's there's
• 20:08
  Speaker 4: also the question of how much those are forced by.
• 20:10
  Speaker 1: Are they necessarily causal? Yeah, not necessarily right?
• 20:13
  Speaker 4: Are they forced by genetics? Are there other aspects that
• 20:17
  Speaker 4: that are you know, that are causally hodogenous there? So
• 20:22
  Speaker 4: there's a lot going on. So there are some studies
• 20:26
  Speaker 4: where you can intervene on the parents to try to
• 20:30
  Speaker 4: get them to talk more to the kids. Not very many,
• 20:33
  Speaker 4: not very big, but there are some, and you do
• 20:35
  Speaker 4: find gains in the child's vocabulary Now the next question
• 20:40
  Speaker 4: is can we then estimate like how big an effect
• 20:42
  Speaker 4: that is? How it lines up. There's a lot of
• 20:45
  Speaker 4: you know, kind of complicated research questions about the causal
• 20:47
  Speaker 4: inference there that's good that are not well worked out.
• 20:50
  Speaker 4: So that's all to say there is some effect. Does
• 20:54
  Speaker 4: that mean that you then should go and intervene in
• 20:56
  Speaker 4: your own house in this kind of uncontrolled way, I
• 20:58
  Speaker 4: would say no. What you should do maybe is think
• 21:03
  Speaker 4: from first principles, like, okay, you know, what's a good
• 21:07
  Speaker 4: like high quality, fun, language, rich thing you can do
• 21:10
  Speaker 4: with your kid? You know, reading a book is one
• 21:14
  Speaker 4: kind of having a conversation or a kind of in
• 21:16
  Speaker 4: depth play session. So a lot of the ways that
• 21:18
  Speaker 4: we think of relating in a high quality way to
• 21:21
  Speaker 4: our kids actually provide really good what we call grounded language,
• 21:26
  Speaker 4: where we're talking about the here and now in some
• 21:27
  Speaker 4: way that they understand and maybe that those are typically
• 21:31
  Speaker 4: the moments that are most predictive also in these studies.
• 21:36
  Speaker 4: So it's not about constantly talking and jamming the Wall
• 21:39
  Speaker 4: Street Journal in their ear. It's about creating moments that
• 21:43
  Speaker 4: are high quality and fun and interactive and motivating. And
• 21:46
  Speaker 4: you know that's often what we encourage parents to do
• 21:49
  Speaker 4: even when we're not really thinking about language per se.
• 21:54
  Speaker 3: Okay, so I'm hearing that it's good to talk to
• 21:56
  Speaker 3: your kid, but you don't need toenerate the diaper change.
• 21:58
  Speaker 3: That's what I That's what I got, or you could
• 22:01
  Speaker 3: if it was enjoyable.
• 22:02
  Speaker 4: Yeah, yeah, if that's the way you relate your kid.
• 22:04
  Speaker 4: I mean, I you know, for me, it's kind of
• 22:07
  Speaker 4: more fun to chat with a baby or a kid,
• 22:09
  Speaker 4: you know, even if they're not chatting back. But yeah,
• 22:12
  Speaker 4: certainly there are probably some seiling effects there. You don't
• 22:16
  Speaker 4: need to narrate everything. And also the kind of language
• 22:19
  Speaker 4: that's most important for the kid is the kind that's
• 22:21
  Speaker 4: about what they care about, embedded in a fun and
• 22:24
  Speaker 4: interesting interaction. So that's the kind of language that's going
• 22:28
  Speaker 4: to be most impactful if you're really aiming in that direction.
• 22:34
  Speaker 3: So a lot of people have talked about the language
• 22:36
  Speaker 3: learning loss during COVID, whether it's because of laugh exposure
• 22:40
  Speaker 3: to other people in a young age, whether it's because
• 22:42
  Speaker 3: of masks. Do we how much do we know about
• 22:44
  Speaker 3: even whether there was a COVID effect on language development
• 22:48
  Speaker 3: for little kids.
• 22:50
  Speaker 4: Not so much. I would say, right, what we have
• 22:55
  Speaker 4: are some very incomplete observational data sets. And you know,
• 23:01
  Speaker 4: I was part of a big cross national consortium that
• 23:03
  Speaker 4: jumped into action and tried to study language effects during lockdown.
• 23:07
  Speaker 4: And maybe there were some tiny little effects, and maybe
• 23:10
  Speaker 4: they were related a bit to things that the parents
• 23:13
  Speaker 4: said they did or didn't do with the kids, Like
• 23:15
  Speaker 4: maybe they said there's more screen time and then they
• 23:17
  Speaker 4: reported less language. Methodological fans of your show or other
• 23:23
  Speaker 4: critics would note immediately that, and we said this in
• 23:26
  Speaker 4: the paper. I think we're very upfront about it, Like
• 23:28
  Speaker 4: if your kid is more on screens during the pandemic
• 23:30
  Speaker 4: lockdown because you got to work, and then you report
• 23:33
  Speaker 4: they have less language, you both know they were on
• 23:35
  Speaker 4: screens and you were also not hanging out with them
• 23:37
  Speaker 4: and seeing what they knew about language. So it's very,
• 23:40
  Speaker 4: very tough to figure this out, and we don't have
• 23:42
  Speaker 4: a lot of direct assessment, like we weren't it was
• 23:44
  Speaker 4: the pandemic. We weren't going into people's homes and you know,
• 23:46
  Speaker 4: showing the ball in the box and the you know,
• 23:49
  Speaker 4: so it's really tricky. I would say from first principles,
• 23:55
  Speaker 4: I am not worried about masking as a key source
• 23:58
  Speaker 4: of issues and language comprehension and language learning. I think
• 24:04
  Speaker 4: there are momentary issues. It's hard to understand somebody when
• 24:07
  Speaker 4: a mask is on, and that's that's gonna be true,
• 24:09
  Speaker 4: you know, multi modal language input, meaning like when you
• 24:12
  Speaker 4: can see the face and the lips moving that actually
• 24:15
  Speaker 4: you know, that's obviously that really does help. And lots
• 24:17
  Speaker 4: of experiments, so understanding what you're saying in the moment
• 24:20
  Speaker 4: with a mask on is harder than with the mask off.
• 24:23
  Speaker 4: But nobody or almost nobody was growing up for more
• 24:27
  Speaker 4: than you know, if you know, for for a long
• 24:31
  Speaker 4: time with everyone around them with a mask on. They
• 24:33
  Speaker 4: were just growing up with sometimes some people with masks on.
• 24:37
  Speaker 4: My kid was probably you know, the most masked of
• 24:40
  Speaker 4: anybody in the Bay Area. Was super intense about this,
• 24:42
  Speaker 4: and they had masks on outdoors for a very long time.
• 24:46
  Speaker 4: You could shake your head at that. You know, they'll
• 24:48
  Speaker 4: still have masks on if there's a you know, a
• 24:50
  Speaker 4: whiff of COVID in a the next county. So there's sorry,
• 24:53
  Speaker 4: but there's still very very intense about these things. And
• 24:57
  Speaker 4: even so, what was he getting, Okay, six seven hours
• 25:00
  Speaker 4: a day of massed input and then he comes home
• 25:02
  Speaker 4: and we're talking to him on the weekend. You know,
• 25:04
  Speaker 4: in terms of hours a day, it's not all the time,
• 25:08
  Speaker 4: it's not so.
• 25:08
  Speaker 3: Much, all right, all right, So I want to talk
• 25:11
  Speaker 3: now about the big data, about the value of big
• 25:14
  Speaker 3: data in this space, And I actually want to tell
• 25:16
  Speaker 3: you that I am personally small data. So when I
• 25:19
  Speaker 3: was a baby, my mom put a tape recorder under
• 25:23
  Speaker 3: my crib on the hest of a language development specialist
• 25:27
  Speaker 3: named Katherine Nelson, and she recorded me for many years
• 25:29
  Speaker 3: talking to myself, and these a bunch of people wrote
• 25:32
  Speaker 3: this entire book about like my individual personal language development
• 25:37
  Speaker 3: as a child. Obviously, your approach is very different because
• 25:41
  Speaker 3: you have more than one child in many of your studies,
• 25:45
  Speaker 3: And so I'm curious how you think about the value.
• 25:48
  Speaker 3: I mean, first, it might be useful to hear a
• 25:50
  Speaker 3: little bit about kind of what one means here by
• 25:52
  Speaker 3: big data, But also then how do you think about
• 25:55
  Speaker 3: the value of that versus the kind of intensive study
• 25:57
  Speaker 3: of say one individual kid.
• 26:01
  Speaker 4: So I think I actually have that book Narratives from.
• 26:04
  Speaker 3: The Crib Yep, yeah, there it is.
• 26:07
  Speaker 4: I remember looking at it, and then somebody at at
• 26:08
  Speaker 4: one point told me that that was you in the crib.
• 26:12
  Speaker 4: I'm also, you know, good friends with my colleague Eve Clark,
• 26:15
  Speaker 4: and I also am friends with her son, who's the
• 26:18
  Speaker 4: subject of her.
• 26:18
  Speaker 1: Mind another end of one.
• 26:20
  Speaker 4: Yeah, I only once brought that up in a social
• 26:23
  Speaker 4: situation where he was placed, So so yeah, I think
• 26:28
  Speaker 4: the first thing I'd say is, there's this been this
• 26:30
  Speaker 4: custom and psychology of doing experimental studies where you get
• 26:33
  Speaker 4: really kind of tight experimental control over something in the lab.
• 26:37
  Speaker 4: And a lot of those studies historically have had like
• 26:39
  Speaker 4: sixteen kids, and each kid has contributed just a handful
• 26:42
  Speaker 4: of data points, and I think of that as really
• 26:45
  Speaker 4: small data. The kinds of things you learn from that
• 26:48
  Speaker 4: it can be very interesting, but typically the estimates you get,
• 26:51
  Speaker 4: the measurements you get are very noisy. They don't tell
• 26:53
  Speaker 4: you much about any one kid, and they don't tell
• 26:56
  Speaker 4: you that much about exactly what this experiment does. So
• 26:59
  Speaker 4: I think you could either go deeper on one kid
• 27:02
  Speaker 4: or broader with many, many kids. Both of those are
• 27:05
  Speaker 4: actually very useful and appealing from my perspective. So when
• 27:09
  Speaker 4: I got started, one of the things that I was
• 27:11
  Speaker 4: doing was collaborating with Deb Roy at the MIT Media Lab,
• 27:14
  Speaker 4: who studied his kid by recording about seventy percent of
• 27:18
  Speaker 4: his child's waking hours. And then his collaborator named Brandon Roy. Unfortunately,
• 27:24
  Speaker 4: because they weren't related at all, it wasn't his son
• 27:26
  Speaker 4: that was the collaborator, It was just another guy that
• 27:28
  Speaker 4: happened to have the name Roy. But anyway, Brandon led
• 27:31
  Speaker 4: like sixty transcribers to transcribe millions of words of this
• 27:34
  Speaker 4: one child's daily life, everything you said, everything you heard.
• 27:38
  Speaker 4: And that's an incredibly fascinating data source that we've worked
• 27:40
  Speaker 4: on a lot. So you learn a lot from that
• 27:43
  Speaker 4: kind of big data, not big n as like the number.
• 27:48
  Speaker 3: We would say like big T for like many time
• 27:50
  Speaker 3: periods in economics.
• 27:52
  Speaker 4: Yeah, yeah, exactly. So I think that that can be fascinating.
• 27:55
  Speaker 4: But like I said, language is so variable that learning
• 27:59
  Speaker 4: about one kid's development really tells you a lot about
• 28:02
  Speaker 4: that kid's development a little bit about other kids development.
• 28:05
  Speaker 4: So the opposite approach has been to look at very
• 28:10
  Speaker 4: very large samples, and so that's what we've done in
• 28:13
  Speaker 4: the Word Bank project, where we gather together surveys of
• 28:16
  Speaker 4: parents about their kids language. Those are quite detailed surveys
• 28:19
  Speaker 4: where we get them to check off what words their
• 28:20
  Speaker 4: kid says or what words their kid they think understands,
• 28:24
  Speaker 4: and then we compile those in a big data database
• 28:28
  Speaker 4: that represents, you know, not all of the world's languages,
• 28:31
  Speaker 4: of course, nor even all of the world's regions, but
• 28:34
  Speaker 4: maybe a broader and more diverse set than what typically
• 28:37
  Speaker 4: gets studied in lab studies. And so that gives us
• 28:40
  Speaker 4: a window into that big and variability, how different kids
• 28:44
  Speaker 4: are from one another, and also gives us a really
• 28:46
  Speaker 4: important window into the acquisition of different languages in different
• 28:49
  Speaker 4: cultural contexts, because there's so much about language and culture
• 28:53
  Speaker 4: that we've talked about here implicitly right, like how we
• 28:55
  Speaker 4: think we're supposed to talk to kids, what we think
• 28:57
  Speaker 4: the goal is that we want to get our kids
• 28:59
  Speaker 4: to succeed at school, and we want to talk to
• 29:01
  Speaker 4: them a lot grit. That's all very culturally conditioned. So
• 29:06
  Speaker 4: that approach has really been to look at variability and
• 29:09
  Speaker 4: to try to understand the sources of variability in kids
• 29:11
  Speaker 4: language and how those relate both within individual kids and
• 29:16
  Speaker 4: across cultures.
• 29:17
  Speaker 3: Are there some languages that are more difficult to learn
• 29:21
  Speaker 3: or take longer? I mean, I know there's a lot
• 29:24
  Speaker 3: of variability across kids, but is there something systematic across languages?
• 29:27
  Speaker 4: You know, this is one of these questions that people
• 29:29
  Speaker 4: always ask linguists, and the lagualis are always like, no,
• 29:31
  Speaker 4: it's universal, you know, the language acquisition device. Could it
• 29:34
  Speaker 4: turns out there it's not completely. There is one language
• 29:38
  Speaker 4: that we know empirically is really horrible for kids to learn,
• 29:40
  Speaker 4: and it's Danish.
• 29:42
  Speaker 1: Huh.
• 29:42
  Speaker 4: Danish has a lot of sounds in it, and the
• 29:45
  Speaker 4: sounds are all kind of slurred together and not well articulated,
• 29:47
  Speaker 4: and there's all these ways that they drop the sounds.
• 29:50
  Speaker 3: So Danish listeners, We're sorry, he's not being critical.
• 29:53
  Speaker 4: I you know. So I know this because of a
• 29:56
  Speaker 4: bunch of really fabulous Danish researchers who have this long
• 29:59
  Speaker 4: running study called the Puzzle of Danish. And the Puzzle
• 30:02
  Speaker 4: of Danish is like, why is it so hard to
• 30:04
  Speaker 4: learn our poor old language? We speak it just fine,
• 30:07
  Speaker 4: but like, man, these kids are struggling.
• 30:09
  Speaker 1: That's so interesting.
• 30:11
  Speaker 4: A lot of the things that can make a language
• 30:13
  Speaker 4: hard for second language learners, well, some of the things
• 30:16
  Speaker 4: that can make the sounds, especially of a language, hard
• 30:19
  Speaker 4: for second language learners, like when you reduce them or
• 30:21
  Speaker 4: you know, kind of they're not as well articulated, or
• 30:24
  Speaker 4: they're slurred together or changed when they're next to each other.
• 30:26
  Speaker 4: That really can make a language harder to learn. Other things,
• 30:30
  Speaker 4: it turns out, you know, kids are remarkably good at
• 30:32
  Speaker 4: learning all those word endings, and you know, in the
• 30:35
  Speaker 4: what we call the morphology, the little bits and pieces
• 30:38
  Speaker 4: that aren't themselves words, all that stuff. Especially English speakers
• 30:42
  Speaker 4: we struggle with. Kids don't have as hard a time with.
• 30:45
  Speaker 3: Yeah, I mean, that's such an interesting aspect of learning
• 30:48
  Speaker 3: to talk and learning to read in English, which both
• 30:51
  Speaker 3: of which seemed a little magical to me. Is this
• 30:53
  Speaker 3: idea that in English there are so many weird endings.
• 30:56
  Speaker 3: There are so many you know, why is the k silent?
• 30:59
  Speaker 3: You know, like when you learn to read, like it's ridiculous,
• 31:03
  Speaker 3: and yet somehow people sort of pick it up, don't.
• 31:08
  Speaker 4: Well, here, English is actually the hard one, right or
• 31:11
  Speaker 4: a hard one. So if you're in an English school,
• 31:14
  Speaker 4: they're all, you know, kids learning to read, they're going
• 31:17
  Speaker 4: to be a lot of kids that are having trouble
• 31:19
  Speaker 4: with all of those complexities of English orthography. That's not
• 31:23
  Speaker 4: English language per se, it's how we write the darn thing. Down.
• 31:27
  Speaker 3: In contrast, Spanish, every it's completely phonetic, like what you
• 31:31
  Speaker 3: see is what you get. You read the sounds and
• 31:33
  Speaker 3: then the sound. Everything is totally finetic.
• 31:36
  Speaker 4: Yeah, So you can go into an English speaking elementary
• 31:39
  Speaker 4: school and find out how good a reader is, like
• 31:41
  Speaker 4: how well kid is doing it at reading by having
• 31:44
  Speaker 4: them just read out words and if they can actually
• 31:46
  Speaker 4: pronounce a bunch of the words, that's a really good
• 31:48
  Speaker 4: measure of their overall reading success. In a Spanish speaking school,
• 31:52
  Speaker 4: that measure is going to top out at like six
• 31:54
  Speaker 4: or seven or eight because kids are just like, yeah,
• 31:56
  Speaker 4: you say the sounds, like what's the big deal? Because
• 31:58
  Speaker 4: my language makes sense? And English speaking kids are like,
• 32:01
  Speaker 4: my language doesn't make sense. I have to read hundreds
• 32:03
  Speaker 4: of books and get explicit phonics training to third or
• 32:06
  Speaker 4: fourth grade in order to just figure this thing out.
• 32:08
  Speaker 5: Now.
• 32:09
  Speaker 3: I think the low when I was teaching my daughter
• 32:10
  Speaker 3: to read it was the word no, because it was
• 32:13
  Speaker 3: just like, why are there these extra letters?
• 32:15
  Speaker 1: Like there's two extra letters.
• 32:17
  Speaker 3: It's just like not not no, like yes no, but no,
• 32:20
  Speaker 3: like I know it, It's just like why are there
• 32:21
  Speaker 3: these extra letters?
• 32:22
  Speaker 1: Who knows?
• 32:23
  Speaker 4: My son Jonah loves to get everything explained. He's like
• 32:26
  Speaker 4: been in the y phase for two and a half
• 32:28
  Speaker 4: years now, and we're trying He's right at the point
• 32:31
  Speaker 4: where he's interested in letters and we're playing you know,
• 32:33
  Speaker 4: letter games in the car, like you know an animal
• 32:35
  Speaker 4: that starts with and we're just like, let's just skip
• 32:38
  Speaker 4: see because we're gonna get stuck on scene. He's just
• 32:40
  Speaker 4: gonna be like, why c make K Like what's the
• 32:44
  Speaker 4: deal here? Like you got to explain this something like okay,
• 32:46
  Speaker 4: it starts with the Norman invasion.
• 32:48
  Speaker 3: It's like how far back do we need to go?
• 32:53
  Speaker 2: All right?
• 32:54
  Speaker 3: So I want to end by asking you about this
• 32:55
  Speaker 3: extremely interesting new paper you wrote about AI. So, so
• 33:00
  Speaker 3: your abstract starts by saying the following. Large language models
• 33:05
  Speaker 3: show intriguing, emergent behaviors, yet they receive about four to
• 33:09
  Speaker 3: five orders of bagneto more language data than human children.
• 33:12
  Speaker 3: What accounts for this vast difference in sample efficiency? My
• 33:16
  Speaker 3: read of this in lay terms is that in these
• 33:19
  Speaker 3: large language AI models, like chatch ept is built on
• 33:22
  Speaker 3: top of they're able to do these incredibly impressive things,
• 33:25
  Speaker 3: like you ask them to write a poem in the
• 33:27
  Speaker 3: style of William Shakespeare about you know farts and they
• 33:31
  Speaker 3: do a great job.
• 33:32
  Speaker 1: That's my kids like to ask that question.
• 33:34
  Speaker 3: But in order to get to that, they have to
• 33:37
  Speaker 3: be fed in an unbelievable amount of data, say like
• 33:40
  Speaker 3: ten thousand times more than a kid will here to
• 33:43
  Speaker 3: get to the same ability to write not a very
• 33:47
  Speaker 3: good poem about farts, but some original generated poem about farts.
• 33:51
  Speaker 1: So what is why is that?
• 33:55
  Speaker 4: Yeah, that's the big question.
• 33:57
  Speaker 3: Unfortunately the paper doesn't answer that. It just says this
• 33:59
  Speaker 3: is very hard, but I'm curious. You had some ideas,
• 34:02
  Speaker 3: so I was curious.
• 34:03
  Speaker 4: Yeah, yeah, there are a couple ideas there. And let
• 34:05
  Speaker 4: me clarify what I mean by kind of intriguing emergent property,
• 34:08
  Speaker 4: like the fart Shakespeare is something that you might take
• 34:14
  Speaker 4: somebody a lot of reading and writing and practice to
• 34:16
  Speaker 4: get good at. But my example that I think is like,
• 34:19
  Speaker 4: hopefully it easier and maybe more appropriate for the This is.
• 34:25
  Speaker 3: An adult podcast, so it's fine to say farts.
• 34:29
  Speaker 4: Well. So the thought experiment I do a lot is
• 34:32
  Speaker 4: like when a kid comes home from kindergarten and you
• 34:35
  Speaker 4: know it's a rainy day on their board and can
• 34:37
  Speaker 4: be like, hey, let's play UNO. Here's the rules. Of Uno.
• 34:39
  Speaker 4: I'm gonna explain them and now we're going to play Uno.
• 34:42
  Speaker 4: And for a lot of kids that works totally fine.
• 34:44
  Speaker 4: You can just explain the rules of a game, give
• 34:46
  Speaker 4: them like two examples, and then you're off to the
• 34:48
  Speaker 4: races and they're immediately starting to argue with you about
• 34:50
  Speaker 4: wild cards or whatever. For AI, that's kind of what
• 34:54
  Speaker 4: I meant by intriguing emergent capacities is that you could
• 34:57
  Speaker 4: give instructions and provide examples and they could learn from that.
• 35:00
  Speaker 4: And that's the thing that popped out after gazillions of examples.
• 35:03
  Speaker 4: You know, they have to read kind of a third
• 35:05
  Speaker 4: of the entire Internet in order to be able to
• 35:07
  Speaker 4: figure that out. So one big difference is the kind
• 35:13
  Speaker 4: of data that the AI is getting. It's getting decontextualized language.
• 35:16
  Speaker 4: It's basically like imagine just printing out all of Wikipedia
• 35:19
  Speaker 4: in a single stream with no formatting and shoving that
• 35:22
  Speaker 4: into a learning machine. It's very hard to figure out
• 35:26
  Speaker 4: what's going on there. In contrast, what kids get is
• 35:29
  Speaker 4: this very social language that's built up in conversation, in
• 35:33
  Speaker 4: context they understand. It's accompanied often by other cues, like
• 35:38
  Speaker 4: you know, you might be looking at the UNO cards,
• 35:41
  Speaker 4: it's you know, within you know, a set of conventions
• 35:45
  Speaker 4: and practices that they understand, like now we're going to play.
• 35:49
  Speaker 4: So there's all these other structures that allow the child
• 35:52
  Speaker 4: to make much more of the data they get. And
• 35:55
  Speaker 4: I think it's an open question exactly which of those
• 35:58
  Speaker 4: matters is that the social stuff? Is the that there's
• 36:00
  Speaker 4: grounded visual input or other kinds of input? Is it
• 36:03
  Speaker 4: the contexts? Like that's the kind of thing that we're
• 36:06
  Speaker 4: as a scientific community working on figuring out. People are
• 36:09
  Speaker 4: trying to train models on videos instead of written text,
• 36:13
  Speaker 4: they're trying to train models in more social ways and
• 36:15
  Speaker 4: so forth. But that's the general puzzle is like, Okay,
• 36:19
  Speaker 4: what would you do if you had much more enriched
• 36:21
  Speaker 4: data like a kid? How much does that make a difference?
• 36:23
  Speaker 4: And why?
• 36:29
  Speaker 1: All right, this is amazing. Thank you so much. I
• 36:32
  Speaker 1: really appreciate your time.
• 36:33
  Speaker 4: Yeah, thank you for having me a big fan of
• 36:35
  Speaker 4: your work. Thank you.
• 36:45
  Speaker 6: After the break more parent data.
• 37:15
  Speaker 7: The first many weeks of nursing my son were extremely painful.
• 37:19
  Speaker 7: My husband joked that his first word was going to
• 37:22
  Speaker 7: be because I said it so often when he was
• 37:24
  Speaker 7: an infant.
• 37:29
  Speaker 5: My daughter is currently fifteen months.
• 37:32
  Speaker 1: My daughter is fourteen months.
• 37:33
  Speaker 8: Old, and his first word was when he was around
• 37:37
  Speaker 8: twenty months.
• 37:39
  Speaker 4: And she said her first word eventually when she was
• 37:42
  Speaker 4: twenty months old.
• 37:44
  Speaker 5: And the first word that she said and has been
• 37:46
  Speaker 5: saying repeatedly has been doggy.
• 37:49
  Speaker 4: Her first word wasn't mama or dada. That word was
• 37:52
  Speaker 4: not mama or dada, but it was our cat's name,
• 37:55
  Speaker 4: which is olive.
• 37:56
  Speaker 5: About is Stan, and Stan is the dog, even her dog,
• 38:01
  Speaker 5: my mother in law's dog.
• 38:05
  Speaker 8: I could not find one of his shoes for the
• 38:07
  Speaker 8: wife of me, and then he brings me the shoe
• 38:09
  Speaker 8: and he goes, shoe, OHK to him, and I was
• 38:12
  Speaker 8: just like you could speak.
• 38:15
  Speaker 5: My husband grew up in New York City.
• 38:17
  Speaker 8: His first word was TAXI loved shouting it as the
• 38:21
  Speaker 8: taxis went by.
• 38:22
  Speaker 5: And so now every dog she sees, she, you know,
• 38:25
  Speaker 5: yells out doggie. But that also means that every other
• 38:29
  Speaker 5: animal she sees she thinks is a dog, also including
• 38:32
  Speaker 5: seals at the aquarium.
• 38:34
  Speaker 1: When she sees a clock, she goes.
• 38:38
  Speaker 5: But every time we try to get her to say mama,
• 38:41
  Speaker 5: she looks at me straight in the face, pauses, and
• 38:44
  Speaker 5: then says dead.
• 38:45
  Speaker 6: Dead.
• 38:57
  Speaker 3: Parent data is produced by Tamar Avishai, support from the
• 39:00
  Speaker 3: parent Data team and PI Rex. Also special thanks to
• 39:04
  Speaker 3: our house violinist, my daughter Penelope.
• 39:06
  Speaker 6: No problem Mom.
• 39:08
  Speaker 3: If you have thoughts on this episode, please join the
• 39:10
  Speaker 3: conversation on my Instagram at prof Emily Oster. And if
• 39:14
  Speaker 3: you want to support the show, become a subscriber to
• 39:17
  Speaker 3: the parent Data newsletter at parentdata dot org, where I
• 39:21
  Speaker 3: write weekly posts on everything to do with parents and
• 39:23
  Speaker 3: data to help you make better, more informed parenting decisions.
• 39:27
  Speaker 3: You can subscribe for free or sign up for a
• 39:30
  Speaker 3: paid subscription, which comes with great benefits, including an ad
• 39:33
  Speaker 3: free version of this podcast and full access to literally
• 39:37
  Speaker 3: hundreds of my posts at parentdata dot org. If you
• 39:41
  Speaker 3: like what you hear, please leave the show or a
• 39:42
  Speaker 3: positive review on Apple Podcasts. It really helps people find
• 39:45
  Speaker 3: out about us. Write Penelope, write Mom.
• 39:49
  Speaker 1: We'll see you next time.