How does language shape the way we think?

Transcript

Intro: This podcast was made on the lands of the Wurundjeri people, the Woi Wurrung and the Bunurong. We'd like to pay respects to their elders, past and present, and emerging. From the Melbourne School of Psychological Sciences at the University of Melbourne, this is PsychTalks.

Nick Haslam: Hello and welcome to another round of PsychTalks. I'm Nick Haslam,

Cassie Hayward: And I'm Cassie Hayward. We're your hosts, and we're just itching to deep dive into more fascinating research in psychology and neuroscience. This season of PsychTalks is already halfway, so if you haven't already, make sure you subscribe so you don't miss any episodes.

Nick Haslam: Today we're talking with Dr Frank Mollica. Frank studies the wonders of human language, what it tells us about how we think and how we interact with the world around us.

His research crisscrosses numerous cultural and linguistic settings, and it also has a lot to say about the horror of legal language. Let's get into it. Welcome, Frank.

Frank Mollica: Hi, thanks for having me.

Nick Haslam: So, Frank, you describe yourself as a computational cognitive scientist, and now that might be unfamiliar to some of our listeners. So what does it mean? And can you explain what cognitive science is and what's computational about your approach to it?

Frank Mollica: Yeah. So, cognitive science is an interdisciplinary field. It mixes anthropology, computer science, linguistics, neuroscience, philosophy, and of course, psychology. Uh, we study the mind like a computer. Right, uh, what are the mental representations and processes, the computer programme basically, that explains human reasoning?

So I'm a computational cognitive scientist, I spend most of my time translating between these two fields and basically taking the theories and the insights from these fields and putting them into math, and then, you know, formalising the math, making nice testable predictions that, you know, I then go out and find collaborators or go into the field and and test with large data normally.

Cassie Hayward: Um, as you've said, the focus of your research is language, but you talk about it as a cognitive technology. What do you mean by that? OK.

Frank Mollica: So, a cognitive technology, I guess at the simplest level, it's a tool. For example, uh, language and number are cognitive technologies, right, we invented that, that's our fault, right, like we have littered the world with linguistic structure and numerical structure, and then we learn from it and we use it to achieve these goals.

Um, and importantly with these cognitive technologies, we learn them, right? We learn them as kids, we use them to, you know, achieve different goals, whether it's, you know, math or whether it's communication, right, and that leaves extra structure in the world that other people learn from. And additionally, we teach these things, right, we explicitly teach people number, we explicitly teach people, right, like language.

Right, and through this repeated pattern of uh learning and teaching, uh we culturally evolve optimal solutions to our problems, right, that help us to achieve our goals, and this actually allows us to flexibly adapt to the environment in ways that biological evolution wouldn't allow us to.

Nick Haslam: This is so interesting because the idea that language is shaped by culture is sort of intuitive to most of us, um, the idea that, you know, different,

Frank Mollica: Languages split up the world in different kind of ways. But I remember back in the dark ages when I was studying language, uh, the emphasis was, was much more on universals, on how all languages are in, if you like, built from the same building blocks. So can you give us an example, um, say, colour words, how do different cultures or languages, if you like, um, break up the colour space?

Yeah, so colour is this fun case, right, we all can perceive the same colours if we have normal vision, right, we can distinguish, you know, the eggshell white from the white, um, but different languages of the world, they don't have a vocabulary that's that rich, right? Um, the basic colour terms, the ones that we use every day, right, not the eggshell, off-white, mauve, turquoise, right? There's only a finite inventory of these basic colour terms. Uh, English has about 11, right, um, blue, yellow, pink, purple, um, but other languages have fewer and some languages have more.

So for example, Agarabi, uh, a language they speak in Papua New Guinea, uh, that only has about 5 colour terms that, you know, everyone would use, right, they use like blue, green, yellow, red, and dark or black. Um, whereas other languages like uh Mexican Spanish, for example, has two blues, uh, they have a celeste kind of sky blue, and then the rest of what an English person will call blue.

Uh, similarly, Russian also has two blues, except they carve it up differently. They have like the blue that you know, generally English people would, you know, agree is blue, but then they carve off a bit that's like a navy blue, it's darker.

Nick Haslam: So interesting, so I guess part of what a computational cognitive scientist might do is see whether there are principles underlying these differences, uh, right? So, uh, one of the ones you refer to, I think when you talk about this stuff is, a principle of efficiency. So what do you mean by efficiency and how do you show that languages are efficient or not?

Frank Mollica: Yeah. So this is the day job, right? Figuring out what are these underlying principles that can explain like all of this variation. We've optimally solved these goals, right? We, we've come up with these efficient structures like language or number, right? But how do we get there? Uh, and so one thing I want to clarify is that goals don't always go in the same direction.

Right, so let's take for example, a listener, a listener's goal might be to hear a word and then be able to identify any object in the world, right? So if I give you a word that can point to exactly the object that's intended, an optimal language from a listener's perspective is a language where every possible state of the world has its own unique word, right? So that means we have a unique word for all 621 Marvel universes and every single item inside them.

Right, uh, impossible, we can't do that, especially because if you think about what a speaker's goal is, a speaker wants to be able to choose the word as quickly and accurately as possible, that's going to get their listener to understand, right? They don't want to search memory through an infinite, you know, amount of words in order to figure out what's the right word that would get my speaker to understand what I'm talking about. They have only a finite memory, they want us to be quick. A speaker's optimal language would essentially be one word, right, 'ba', and when I say 'ba', it means whatever I intended a speaker to mean.

So all of my intentions, just one word, ba, I don't have to think about it, right? Now, of course, neither of these languages actually work, right, we call them degenerate languages, but real languages have to figure out how to trade off these two goals and how they choose to budget these goals defines a sort of efficiency trade-off.

Right, it defines a trade-off between these two pressures, one for uh simplicity, we don't want a language that blows up in like a vocabulary, right, but also informativity, we want all of the new words that we add to be useful, right, to help us identify the things that we care about. Uh and so languages of the world tend to efficiently optimise this trade-off, right? Each language gets to choose how it wants to budget and spend on, you know, uh this much complexity for this much informativity.

And what's interesting is we can build these computational models off of these two principles that define all of the different ways that you could budget between informativity and uh and simplicity. And when you do that, you get a whole variety of sort of a trade-off, Pareto frontier of possible ways that languages could be, right? They could solve this problem.

And all of the existing languages for something like colour tend to fall exactly near this boundary, suggesting that every language is indeed efficiently solving this problem. They're balancing their goals, but they all pick how they want to budget, you know, uh, which, which goal do I prefer more or prioritise more differently. And that actually characterises the diversity that we see in the world.

Cassie Hayward: Is it just about efficiency, or does it shape how we see the world? I had to chuckle when you talked about the Marvel Universe because I feel like I've learned a whole new language over the past few years as my kids become obsessed with soccer. So I've learned a whole new words, new everything. Am I changing the way I see the world with that?

Frank Mollica: So I would argue that language definitely changes how you see the world, but it's not in like the Arrival, the movie kind of way where if you learn the alien language, right, suddenly you can time travel, right? And it's also not in the perceptual kind of way, where like, you know, my language has two colours for two colour words for blue, now I can, you know, see blues differently, because we can all see, you know, the shades of blue is the same, right, we can all distinguish between them.

Um, but what language does do is it points attention on all of the structure in the world and it highlights what's important, right, what's going to be useful distinctions for later things in learning, right? Language is a primary tool for social transmission, right, this core thing, uh, where if we're going to culturally evolve good optimal solutions for, you know, our different goals, then we need to be able to transmit it. Uh, so for one example, to give you, you have to make this concrete, is if you think about like kinship terms. Right, kinship terms, you know, everybody has a family tree, right, uh kinship terms are things like mother and brother and whatnot.

So everyone has a family tree, but it's invisible, you don't see it, right, like it's latent. If you're a kid and you're trying to figure out what your kinship terms are, you have to figure out basically just from the words as highlighting different people, what the underlying relationships are, and these underlying relationships can have different goals that they're made to achieve that you don't actually pay attention to or even see when you're a kid. For example, uh, we can think about the Yanomami tribe, uh, these are people that live in the Amazon rainforest, right?

Whereas English doesn't separate uh cousins, your mother's brothers, uh, your father's brothers, your mother's sisters, your father's brother's kids, they all get the same term, they're all your cousin, right? And Yanomami, they actually care about uh whether you're a parallel cousin or a cross cousin.

So a parallel cousin is your mother's sister's kid. Your father's brother's kids, right? A cross cousin is your mother's brother's kids or your father's sister's kids, right? Um, and so kids at a very young age have to learn, you know, the difference between these two, different groupings, right? Um, and this for a kid, you know, it can seem very arbitrary, it's a nice genealogical relationship, it exists, you can look at it on the tree, right? But like, why?

Um, and it turns out that these kind of relationships actually have more to do, uh, with in this case in mating, for example, where, uh, it's tough to find, uh, you know, a potential mate in these kind of tribal places, right, and so this community solves that by having preferential marriages to cross cousins when they're mate limited. So your cross cousins are potentially future mates, um, but this doesn't matter to a kid, right? Like a kid who's learning this language is not going, who am I mating, they're a kid, right?

Um, so without language, you wouldn't be able to be aware of these kind of distinctions like early on, and these kind of distinctions can highlight or build uh structures that, you know, are useful for other goals later on in life.

Nick Haslam: That's so interesting and so you've given us an explanation for why the Yanomami might have a different, way of dividing up kinship than we do here. What's the reason why, just going back to colour, why Russians would divide blue in a different way from Mexicans and why we English folk don't?

Frank Mollica: Uh, so this is a, a nice question. I don't know if I'm gonna have a satisfying answer, um. The answer that I would uh suggest or I'd go with a hypothesis at the moment is that if we want to communicate about things informatively, right, that means that we need to know how often we need to talk about or this particular tool in this case is word, is going to be useful in achieving our goal. Um, and so if you look at the, the different parts of the colour space, what we want to do is we want to normally identify things by colour.

Right, and so if there are things that are important to identify by colour, right, that's how we're going to carve up our colour space so that we can be more precise and have harder edges around those specific cases.

Nick Haslam: So sort of adapted to the local cultural or ecological or something environmental context?

Frank Mollica: Ideally it would be adapted, uh, exactly, it would be adapted to the local context. So for example, if you look at uh places that are tropical, have lots of, uh, you know, rainforests, green vibes, lots of bright, vibrant colours, uh, they tend to split the colour space so that they, they make really clear these bright, you know, uh, colourful, poisonous amphibians that you should not touch, uh, versus, you know, things that are OK to touch and interact with.

Nick Haslam: So another aspect of language you've studied, and you brought this up in relation to the uh kinship terms a little bit, is how language is acquired. And you were involved in some fascinating work again in South America on the development of mathematical, uh, knowledge and number concepts in particular indigenous people, uh, in that region. Can you tell us a bit about that work and what it showed?

Frank Mollica: Yeah. So I was lucky enough to be able to collaborate with some people who are working, uh, in Bolivia with the Tsimane people. Uh, the Tsimane are hunter gatherers, it's a hunter-gatherer society, uh, they live not far from La Paz along a river.

So what we're interested in is primarily numerical developments and what's fun is that the Tsimane language has a base 10 counting system, right, so 1-2-3-4-5-6-7-8-9-10, then it sort of resets, right? A lot of other industrialised societies also have, you know, a base 10 counting system, right, any English two-year-old can sing that song for you, they know how to count like 1-2-3-4-5-6-7-8-9-10.

The interesting thing though is that while a 2 year old can count that, 2-year-olds don't actually know what those words mean. So if you put a pile of cookies in front of the same 2 year old who just counted 10 for you and ask them, Can I have 3 cookies, right, you're just as likely to get the entire pile of cookies as you are to get however many fit in their hand, and certainly not 3.

Um, right, kids learn number words in stages. Uh, first they learn what the meaning of one is, they can hand you one cookie exactly, then they learn what the meaning of two is, right, so they can hand you 1 or 2 cookies, but you ask them for 3, you still get that handful, then they become 3-knowers and they figure out what 3 means. Sometimes you see kids that are 4-knowers, so they know what 4 means, but what's really cool is that we don't have 5-knowers.

Right, By the time that a kid would be a 5-knower, right, they just figured it out, they figured out the algorithm, they can now count. So however many numbers that they can count to, they can now accurately, well, as accurately as they can count, they can hand you that many cookies, right? What's really cool is this happens in all industrialised societies that have a base 10 counting system. It takes a couple of years, right, we see this in Japan, we see this in Hebrew, Arabic, um.

Every language that we have data for, right, kids go through the stage of development, you know, uh 1-knower, 2-knower, 3-knower, 4-knower. We want to know, hunter-gatherer society, Tsimane kids, do they show the same pattern, right? They also have a base 10 system, but you know, very different from these industrialised societies that we have data for. So, uh, you know, uh, my collaborators basically went and looked at the Tsimane children and we found out that yes, they do go through this exact same pattern. Uh, they go 1-knower, 2-knower, 3-knower, 4-knower, and then eventually they figure out the algorithm.

What's really cool is we recently did a meta-analysis, and it turns out that the cardinal principal knowers, the kids that figured out the algorithm. They all have some formal schooling, um, and formal schooling among the Tsimane is actually really recent, um, and it's not formal schooling like an industrialised society, is much more like, um, you know, when the government can provide aid, they have a teacher who comes and they teach the Tsimane kids in Spanish, uh, and so the instruction is in Spanish and it's very recent, um, but only the kids with formal schooling, uh, have, at least in our data set, uh, have acquired that full counting ability.

But what's really cool is that there are people, there are Tsimane that are adults that, you know, did not have any of this formal schooling, uh, and they can count, right? And so we wanted to figure out, you know, well, how did they figure it out, right? They didn't have formal schooling, so is formal schooling actually required, right? Um, and so the anthropologist on our team, David O'Shaughnessy, went and actually uh went to go find adults that, you know, might not have had the experience of formal schooling, but they might have had mathematical uh experience elsewhere, they might have done trade.

Right, so the Tsimane do trade with uh other people, right, and the trade language is sort of Spanish because that's a, you know, the the lingua franca of Bolivia, right? Um, and what they found, Dave, when, when he went there, he found out that basically these people maybe they, they can't exactly count, but they can do trade, and so he found this really cool uh group of people who could do mathematical computations that were necessary for trade.

So for example, uh, in the Tremane society, basically trade works in fives, you have these jatata leaves uh that they trade or sometimes they'll trade like a bunch of uh bananas or plantains if they're in the right sort of amount, like for a cluster. They basically group them in 5 and they do multiples of 5 is like the the trade. So, uh, if you ask these people, you know, 5 times 2, 5 times 3, 10 times 4, fine, 10 plus 5, fine, 3 plus 4, no idea, wrong, right, 2 plus 5, nope.

Um, right, so they figured out enough math, they figured out math that works specifically for them for the goals that they're trying to achieve, right? And I think that's why I guess numerical cognition is such a great idea, such a great example of these cognitive technologies, right, these technologies that are customarily adapted to our goals, right, if the only thing that I really need to do is trade in multiples of 5, I'm going to learn math in multiples of 5, right? Like why learn that 1 + 1 stuff?

Um, and I mean, this isn't the first time an anthropologist has found a result. Um, if you look back, uh, Jeff Sachs has worked with Brazilian candy sellers during periods of inflation, where, uh, if you're trying to sell candy in massive periods of inflation, these like 6 to 10 year old boys would basically be trying to sell candy at the public transportation, uh, stops, right, but you wouldn't know, like how much am I actually going to need to cost these things because the, you know, the, the price changes rapidly over the course of a day.

Right, so you have to sort of really quickly offhand make some heuristics that in this case, use a magnitude comparisons to figure out how you should price your candy so that you have enough to buy tomorrow's box of candy and that you still make a profit.

Right, and when you compare these, you know, like non-schooled Brazilian candy sellers to, you know, people of equivalent socioeconomic status with schooling or even the the kids that have, you know, like a good socioeconomic status and, you know, schooling, they're better at these abstract magnitude comparisons than their peers, right, they can't do the formal counting thing that they would teach you in school, but it's like they can teach you what they needed, right, and they're better at it, um, so. These kind of things really are adapted, cognitive technologies are, are really adapted to what we need to do.

Cassie Hayward: I think they're such fascinating examples of how efficient language can be, but also so adapted to the goals that you need in that, in that scenario. But you've also done research on the type of language that's spectacularly inefficient. Can you tell us about legalese?

Frank Mollica: Yes, um, legalese is notorious, right? What do I really have to say? Um, legalese is difficult in so many different ways, um, rather than just saying a clean sentence, legalese likes to take a sentence and throw it in the middle of the other sentence, right?

For example, the contractor who knowingly and in sound mind and good conscious entered this employment agreement upon termination of their employment at either the employer or their behest, will be entitled to a one-time payout of $100 million.

It's a golden parachute clause that I just made up, um, right. So when you have that clause that that's a big long sentence, but it's actually two sentences where they just threw one right in the middle, right? The outer sentence is that the contractor is entitled to a one-time payout of $100 million right, upon termination of their employment at either their employer or their own behest, right? But then we ended this other sentence, this, the contractor is in sound mind and good conscious and when they entered this agreement.

Right? Why did that need to be in the middle of this other sentence? It's memory taxing, legalese isn't fun, uh, it's a tax on memory, but also legalese uses these rare words that we never see anywhere else, um, mens rea, mens sana, um, or even simple things if you look at your, uh, real estate agreement in a lot of different languages, right? Why do we call it lessee instead of renter? Renter is much more friendly when everyone knows what a renter is. What is a lessee?

Cassie Hayward: It, is it all the complexity required because legal concepts are themselves extremely intricate, or are they just trying to confuse us?

Frank Mollica: Ah, great question. So, we've actually looked at this, uh, and worked with my grad student Eric Martinez and Ted Gibson at MIT. Uh, we basically asked a whole bunch of lawyers, right, uh, we gave them contracts that were written in sort of plain English, and we gave them contract excerpts that are written in legalese, and we asked them, are they both equally enforceable, right? Do they have the same legal content, do they have the same legal standing, right? Uh, and overwhelmingly they do, right? Plain English equivalents do exist, it's possible. So there's nothing about the concepts that make these things complex because I can easily undo them, uh, just by taking sentences outside of other sentences, using slightly more frequent words.

Nick Haslam: So if language tends to become more efficient over time, surely over time legalese has also become more efficient?

Frank Mollica: Uh, you would think that. Um, no, we've actually also looked at this. So, uh, in a study, I guess last year now, uh, we looked at the entire US legal code, uh, and I should say so far we've only ever looked at US laws, right? It might be different than other laws and we're looking at that now, but so far we've only just looked at US legal code. And if you look at the entirety of the US legal code up to about 2022.

Right, and you can find uh other texts that are comparable. We can look at like fiction from the exact same time periods, right, or even academic texts with all of their weird jargon and whatnot, um, from the exact same time period, and we can actually look at, you know, how prevalent are these really hard-to-process linguistic structures, this kind of centre embedding where you take sentences and throw them into other sentences or the frequency of words.

Uh, and if we look across all of this time, legalese is always containing more of these difficult to process structures than any of the other control tests. This is despite calls for, you know, reform in like the 1970s and even the Plain Writing Act, uh, the Plain Language Act of 2010.

Nick Haslam: Uh, I do hope, Frank, that we don't get sued for talking about all of this, and, and, uh, probably we should be evenhanded and start talking about that jargon in psychology you mentioned, but if legal language isn't getting more plain and more straightforward over time, what is the obstacle? Uh, and what does it tell us about how we should be trying to increase the use of plain language?

Frank Mollica: So, whenever you're trying to change something that exists and has this kind of structural momentum, it's really hard to just overcome momentum. It's so much easier to just copy this template and use it on the next document, right? Similarly, there's uh very few structural changes that uh really incentivise uh using simpler language, right? There's no reason to use legal language is going to change people's financial and goals basically. It's harder to change something that already has momentum when there's no clear incentive to do so.

Uh, and so we have to change the incentive structure of institutions if we want to actually see results.

Cassie Hayward: It reminds me how kids have their own language, every kind of generation have their own words for what's cool and what's not. Is it just so they have their own little kind of way of speaking that other people aren't allowed in to understand?

Frank Mollica: Uh, so when we did that study with the lawyers, we actually asked them this exact same question. We were, you know, interested to see if maybe it's just an in-group bias, so it's like you're going to signal that you're a good lawyer and so you're going to use your jargon to do that. Um, and it turns out it's not, the lawyers would equally hire, equally work with uh people who use the plain language alternatives versus uh legalese alternatives. What we actually think keeps legalese going this way, um, is actually something called performativity.

Uh, is this idea that language in legally isn't just describing something. Normally when we use language, we're trying to communicate, we're just trying to describe the state of the world. When we use legalese, we're actually changing the state of the world, right? I am now placing with this legal language an obligation over you or some kind of right or something upon you, right?

The same way where you, you know, like crack a wine bottle over a ship, and you say like, I now christen you the whatever and you've now named it, you've changed the state of the world, or when a priest says I now pronounce you man and wife, right, and you're now married, the world has changed and this kind of performativity, right, um, is the same kind of thing that happens in legalese.

Uh, and so we see this kind of performativity also in somewhere else, uh, we see this in magic spells, right, uh, magic spells are also supposed to change the world by just, you know, words themselves, and magic spells also use language to signal that they're doing it, right? So your magic spell is supposed to use some archaic language or it's supposed to rhyme, right? That's how you know that it's a magic spell. Well, our current hypothesis is that with legal language, it's basically the same thing, except the structure isn't rhyming or archaic language or maybe it is some archaic words, um, but it's throwing sentences in other sentences and being complicated, uh, and that sets it apart, that gives it sort of legal weight in people's minds, not in any sort of, uh, evaluative body of the law.

Nick Haslam: Fascinating stuff. So look, Frank, on a final note, I've got a bone to pick with you about an old paper of yours, uh, where you said that English speakers have learned only about 1.5 megabytes of linguistic information which, for listeners of my age, will remember a 5.25 inch floppy disc from the eighties, uh, would hold. I mean, surely clever people like Cassie and I know more than that.

Cassie Hayward: I mean, it's true, like, clearly people know more than just, you know, the save icon worth of information about language. Um, but when we look at language, we talk about the linguistic forms of words, right, and how things combine.

Frank Mollica: That's actually really small. One save icon, one floppy disc contains all the information that you actually need about uh language and how it combines the the actual structures and forms. The part that really, you know, takes up most of the space even on the floppy disc is the semantics, what words mean, right? Uh, it's the part that makes the large language models large, they're supposed to be world knowledge or something like that. Um, but that's, you know, separate from language, it's not much knowledge about language, um, and even in our estimate that's the vast majority of the, the information that you need to learn about a language is semantics, it's word meanings.

Cassie Hayward: I feel like I need a whole floppy disc size memory of the offside rule in soccer, which I will never understand. But, Frank, the work you do probably doesn't necessarily, isn't necessarily the first thing people think of when they think about psychology, but I think it really shines a light on how we think and learn and speak and interact. Um, and I just want to thank you so much for sharing with us today.

Frank Mollica: Thank you so much for being here. It was a pleasure.

Cassie Hayward: And that wraps up this episode of PsychTalks. A big thank you to our guest Dr Frank Mollica for sharing his insights. This episode was produced by Carly Godden with support from Mairead Murray and Gemma Papprill. Sound engineering by Jack Palmer. Thanks for tuning in. See you next time.