203 | N.J. Enfield on Why Language is Good for Lawyers and Not Scientists

0:00:00.3 Sean Carroll: Hello everyone, welcome to the Mindscape Podcast. I'm your host, Sean Carroll. We talk about a lot of different topics here on Mindscape, from wine to the origin of the universe. But there is one thing that they all have in common, that when we talk about them, we are using language to do the talking, you are hearing me speak words, usually with a guest also speaking words, forming sentences and so forth. The structure of language is very basic to how we communicate about the world and is also basic to how we think about the world and how we perceive it.

0:00:33.4 SC: There's a famous hypothesis, the Sapir-Whorf Hypothesis in linguistics that says that the way that we shape... The way that we think about the world is only through language, you sort of cannot perceive things that you don't have the language for. Now admittedly, this is not exactly true. We've actually talked about this on the podcast before and we know that there's more to it than that, but there's something true about it. The existence of concepts in our vocabulary helps us access the world in a very direct way. In fact, it's a kind of coarse graining, which we've talked about in the context of statistical mechanics, and also a kind of emergence in a sense, because the world is really really complicated. There's a lot of things going on out there.

0:01:17.0 SC: And yet, when we talk about it, when we both think about it internally and also express it to other people, we have a finite number of vocabulary words with which we can do that. Sometimes we're straining to find the right word, but sometimes, certain words just don't exist. The number of colors, for example, that a human being can perceive, as we talked about with Ed Yang very recently, is very large, but the number of words for colors is relatively small. So what is it that lets us choose exactly which words should exist? What are the processes by which we evolve into a language structure that lets us describe the world in a way that is both efficient but also good enough for the purposes that we have?

0:02:02.5 SC: Today's guest, Nick Enfield, is a linguistic anthropologist who's thought a lot about the origin of languages but also about the purpose. It's not just, "Well, this language grew up in this region," but why did that language grow up, what were the reasons that were in that moment really, really useful to the people who are inventing these ideas of languages. And as you might expect, upon a moment's reflection, a perfectly scientifically rigorous way of thinking about the world is not necessarily the only motivation we have. So the natural tools we're given to describe the world in terms of language are not necessarily optimized for scientific rigor and accuracy. What are they optimized for? Well, let me just say that Nick's new book is called Language vs. Reality: Why Language Is Good for Lawyers and Bad for Scientists.

0:02:55.5 SC: The idea being that a major reason why language was first invented is not for description but for persuasion, for trying to get some social cooperation, either an individual to do something with you or a group of people to go along with something. That's not necessarily bad, that's an important feature of social life as a communicating animal, that's what language is really for ultimately. But then we re-purpose it for scientific purposes, and this is science in the broadest possible sense of the word, trying to understand the world in an accurate way.

0:03:29.0 SC: So what that means is that this tool that we have for talking to each other about the world isn't really meant for talking about the world in an unbiased way that has really interesting implications for how to think about language, how to be careful not to be falling into traps of bias and over-simplification and so forth and also maybe how to use it better as we learn more about it. So this is both interesting intrinsically as a study of linguistics and how language is used and also as a practical matter, how to think about the world in a more precise way if our goal is more scientific, not to mention if our goal is to persuade people to go along with us. So let's go.

[music]

0:04:25.2 SC: Nick Enfield, welcome to the Mindscape Podcast.

0:04:27.3 Nick Enfield: Thanks very much for having me.

0:04:28.8 SC: We're gonna talk a lot about how language is used, you have this fantastic subtitle on your book about how language is great for lawyers and terrible for scientists, so that's just candy for someone like me, we have to talk about that, but let's start with something that is not directly what you talk about, but important, which is the origin of language. This is something people have talked about a lot, maybe we don't know that much, what do we know about how language got started?

0:04:56.2 NE: Well, there's been a great deal of activity in the last few decades, famously there was a long time ban on talking about the evolution of language, within linguistics anyway, because it was regarded as being overly speculative. But in recent decades, I'd say since the '90s especially, there's been an enormous amount of attention being paid to the question and amazing advances in new data to do with the evolution of humans and the vocal apparatus, a lot of advances in thinking and also computational techniques for modeling hypothesis about language evolution and a lot that's going on. And there's a whole society for studying language evolution and they now have a massive conference every year, and so it's a very lively field. There is, of course as you would expect in a field like that, not a huge amount of consensus.

0:06:12.7 SC: Consensus. [chuckle]

0:06:14.6 NE: There are some really quite different ideas about what are the important puzzles to solve, but language is just this incredibly multifaceted phenomenon, so there's a whole lot to talk about. But from my point of view, what I think is most important about that question is the kind of cognition that is a prerequisite for the evolution of language. And so I'm in the camp of researchers like Mike Tomasello and Steve Levinson in the cognitive sciences who have emphasized, well, Robin Dunbar too for that matter, who have emphasized that social cognition is really crucial. And prior to that, the emphasis was more on how we process information and logical operations like recursion. And these things are important, but I think what we've learned in recent years is that social cognition, social coordination, so-called theory of mind, these are really crucial prerequisites for language.

0:07:27.1 SC: Well, you mentioned something when you talked about the evolution, that I've always been impressed with when I hear people talk about it, which is the sort of give and take between our physical biological evolution and the evolution of language. I think that there's a metaphor that a lot of modern people will have whereby language is more like software and our bodies are more like hardware and those are just different things. But clearly... Well, clearly the biological evolution has influenced the evolution of language and maybe even the other way around.

0:07:58.7 NE: Yeah, I think the first thing to really emphasize is that when we talk about the evolution of language, there are two really very different evolutionary tracks and one of them is the biological development of the human species, the fact that we're different from other species and the fact that you can take a newborn from anywhere in the world and put them anywhere else in the world and they'll acquire the language that's being spoken around them. So there's something in our biology that makes it possible for us to acquire language, there's a lot of discussion around what that is. But that is entirely different from another evolutionary track of language, which is the historical evolution of individual languages. And there's 6,000 or 7,000 languages spoken in the world today and it's hard to estimate very accurately, but most people would say well over 100,000 languages ever spoken. And those languages are all products of evolution on the cultural historical track.

0:09:12.6 NE: And I think the interesting questions now are about the, as you suggested, the kind of interplay between those, where human languages, the ones that we speak, such as English we're speaking right now or any of the other ones, have evolved historically and adapted to the biological properties of humans. So things that are relatively easy to learn, things that are relatively easy to produce, things that give the best trade-off between effort and reaching your goal and so on, those are the things that will be more likely to circulate. So there's a really strong interplay between that kind of historical evolution or cultural evolution track and the biological wherewithal for understanding language. And this is an open question, but the question that you really started with, are we... Have we been changed because of language?

0:10:17.6 NE: I'm not sure if we can say just yet if we've been biologically changed because of language but certainly people would say yes, in terms of the vocal tract and so on. But I think that the obvious answer to the question has to do with what society is like and the fact that we've just had this unbelievable revolution in social organization and culture around the world. Humans are off the charts in terms of what any species can do. We have science, we have cities, we have these amazing technologies. And arguably, none of those things would be possible without language. So in a sense the language has changed us radically in so far as it's created a world for us that we need language to continue, so it's a kind of a niche construction idea.

0:11:15.1 SC: Do we know if language developed independently for different groups, or was there one less universal common ancestor of language like there are for biological organisms?

0:11:26.3 NE: We don't know that, there's certainly a lot of discussion about that. There's controversy around the question of whether our relatives, Neanderthals and other species, had language or not and it's certainly not settled at all. I think received wisdom is that they didn't, but it's certainly being challenged these days and there's a lot of extrapolation that has to go on. So obviously, language itself doesn't fossilize, so you have to extrapolate from questions around what are the... What's the evidence of cultural organisation around archeological findings? Also, what's the evidence from the anatomy? You can sort of tell evidence about the brain structure very... It's very sort of inferential, but just brain size, cranial capacity and a whole lot of elements that come into these discussions but in answer to your question, certainly there is no received answer. And the question of whether all existing languages are somehow descended from a single proto language is again in a certain sense unknowable because we can't go back that far in time, but there's some really interesting work that's been done in recent decades on language birth in a sense.

0:13:09.2 NE: So in some situations, in society where... So there's a few different context in which you can get something that we refer to as language birth and one of them is quite political but certainly linguistically very interesting and that is the emergence of languages out of... In colonial times with slave populations, you had people being brought together who came from many different language communities and these are the types of context in which you get so-called contact languages and what we now might call Creoles, and those are languages that are sort of, the words are often from a donor language, a European language such as French or English, but the structure of the language kind of grows over time with the first few generations. Particularly when children learn the language, they bring structure to it, new structure and within a couple of generations you get what used to be a kind of a simple contact language, quite quickly becoming what we might call a fully-fledged language.

0:14:30.8 NE: Now that is developed from a vocabulary that's been borrowed. There's another type of language birth that is fascinating, which is the birth of sign languages in communities of people who are deaf and hard of hearing. In a family setting, you get situations where a child is deaf and so they won't acquire spoken language in the same way, and you get what are sometimes called home sign languages or there are other terms for them, these are systems of signs that take on some firmly linguistic structures and they grow for practical purposes within homes.

0:15:26.1 NE: And oftentimes they run a shortish course and then they don't need it anymore, people move and all of that. But when you get a concentration of deaf people within a community, and that's happened historically many times where, for example, you have a high genetic incidence of deafness within a small community or you have a lot of deaf people coming together in a city, deaf schools pop up around the world and then suddenly you have a critical mass of people who are using a language and in the case of deaf sign you have quite a strong tendency to draw on more iconic representations, things that roughly look like what they're representing, and within a few generations, again, you have a fully fledged language coming sort of out of nothing in a sense. Most languages aren't like that, most languages draw on a historically developed language that you can trace back in time, maybe a few thousand years, probably not beyond, but there do exist languages that are born anew in the ways I've just described.

0:16:43.7 SC: And this gets us on to what is more about the subject of your book which is, not the give and take between our biology and our languages, but the give and take between our way of perceiving the world and our language. Maybe it's useful to go all the way back to the old fashioned Sapir-Whorf hypothesis that said that the way that we conceptualize the world is entirely in terms of our language and I get the impression that this is not completely true but also not completely discarded.

0:17:16.0 NE: That's right. The Sapir-Whorf hypothesis is controversial and it's extremely catchy. People love it and the idea being that the language you speak changes your perception of the world or alters your reality and it's been, because it is quite catchy, it's been understood in ways that are convenient to non-experts or sort of fun for non-experts and that's what has kind of circulated around the world. For example, you'll hear it said that a language is a prison for the mind and you are locked in to your interpretation of the world by the language you speak. And this is a very strong hypothesis and it's not something that Sapir or Whorf ever said in so many words. Edward Sapir and Benjamin Lee Whorf, they were linguists and anthropologists of the first half of the 20th century and they were following in the footsteps of Franz Boas, another anthropologist. They were fascinated by encountering languages that had very surprising structures to them. There was something that Whorf referred to as standard average European, the kind of language that standard average Europeans speak.

0:18:44.2 SC: Yep. [chuckle]

0:18:45.4 NE: Of course, European languages are different in many important ways, but there are many similarities and their idea was, oh, people in Europe who have this sort of tradition of thinking about language in recent terms and they think about the diversity of language as being the difference between Russian and French and English and Greek or something, and they pointed out, you haven't seen the half of it, this is a small amount of diversity. Go to Native America and learn languages like Hopi or Nootka, the languages that they were studying, and they find that the way in which reality is described is very, very different. So there were hypothesis about how time was understood, the way in which the language described time not as an arrow but as a kind of cycle. And in fact, Whorf found it difficult to articulate the difference, and this, I think, led to the idea that you can't understand the Hopi world view until you speak their language, I can't simply explain it to you in English it's part of the idea.

0:20:02.2 SC: I see.

0:20:03.3 NE: So there were interesting things that... Or fascinating things that Whorf and Sapir both laid out in terms of just how languages are structurally different, but what Whorf, I think is perhaps a bit more famous for, is his writing, which is actually quite limited, but it's probably his most famous writing that comes out of his work as an insurance inspector. He worked for an insurance company and was called out to fires to check them out and find out what was the cause of the fire and did the company, were they entitled to an insurance payout? He looked at all these fire events and he describes how he believed that the way in which people describe the situations in their surroundings led to certain behavior that would not have arisen if they had thought about or described it in other ways. And it's pretty simple what he's arguing, so he's basically saying... The most famous example is the fuel drums that were empty, according to the people who he spoke to in his investigations. And he points out, well, yeah, they didn't have fuel in them, but they were full of fumes and this was explosive and people were careless with cigarettes and a fire was started.

0:21:39.8 NE: And he gives other examples like a substance called spun limestone, which he says is also flammable and people wouldn't think that because the name of it has the word stone in it and stone doesn't burn. So it's a pretty straightforward kind of claim that he's making, and I think that the claims about Whorf being all about how your reality is fundamentally different because of the language you speak, they're really rather exaggerated. In a few places he may have spoken in those ways, he did, but also a lot of the time he was being much more concrete and for me, I think the most interesting part of the claim is that the language you speak affects your reasoning which in turn affects your behavior. But one thing I've tried to emphasize in the book is something that he didn't really pull out, but it's in the work and that is that we also use language to explain what has happened, to explain why we've acted in the way we have and to justify or to give an account for what we've done.

0:22:56.2 NE: And that's exactly the context in which Whorf was extracting these descriptions of how fires started, is that people were giving an account, "Well, the fire started, there was this spun limestone and, gosh, who knows it was gonna go up in flames?" So I think that points to my mind as something quite overlooked in the literature on all of this, which we refer to as linguistic relativity and that is that it's not only about cognition, it's not only about thought, it's just as much about social coordination and about presenting states of affairs in a certain light and particularly where that presentation of states of affairs is in your interest or it aligns with what you're trying to do, and one of those kinds of things would be to defend your actions. And that is where the lawyer comes in, and it's the metaphorical lawyer, I have nothing against lawyers per se in that.

0:24:04.7 SC: And this is exactly a crucial feature that shows up over and over again, not only in science, but in philosophy, probably in other areas of academia, where a lot of the work you have to do is seeing through the ambiguities that are handed to you by your language. The natural languages that we invent to get through the day is not up to the task of discussing quantum mechanics or metaphysics or what have you. And what you're saying in part is, it was never meant to be, that's not why we invented and came up with language in the first place.

0:24:40.2 NE: That's right, it is... How we came up with language in the first place is an interesting way of putting it, so going back to your question about evolution, the cognitive wherewithal for language, of course, we didn't invent that. Modern languages are in a sense created by people, and they are a kind of technology in that sense, but most of the time, they're not invented by people and intentionally, so it's quite interesting when you raise the question about science, because in science we do coin words and we do say, "Oh, here's a concept, we don't have a word for it. So we better create this word and here it is." And then, of course, we use other words to define it.

0:25:25.6 NE: So anytime we're creating a new bit of language, whether it's on purpose like that or it's much more sort of organic, new words pop up all the time in the culture and no one really knows who invented them, but they are being built up through language that already exists. So for me to explain to you what this new word means, I'm gonna oftentimes be using words that already exists in my language. And I think that, of course, that puts a kind of bias on how our language can develop and there's something that people like Whorf and Sapir might have referred to as worldview, they're themes that permeate throughout a language, get reproduced in some sense, partly because they align with our cultural norms and all of that. I think that one of the crucial problems or downsides of all of this is that once you have terminology, even if you've invented something that captures your brand new specific concept that no one knew about before, you are creating a sense in which people become satisfied with that label and satisfied with that description.

0:26:46.0 NE: So in a book I say that words are off switches for the mind and that's the sense in which I mean that. So they... You do a kind of satisficing, you do a kind of shutting down of your reasoning processes when you have language that you're happy with simply because everything's going by so fast we're articulating language and processing language at this great rate. And you don't wanna be inspecting the meanings of words for very long, you don't wanna be negotiating them. Of course, you can, and we do, but you certainly can't negotiate every word as it flies past. So under all of those pressures, terminology, whether it's everyday words for objects or scientific terms, they will tend to bring about this kind of, I don't know, something like a kind of complacency in a way that we all know what this means. And we stopped talking about what it means.

0:27:50.9 NE: But under the surface, do we really all have the exact same understanding of it? Well, we don't detect that a lot of the time and the answer to that question really depends on the function of language. And I think that's one of the key pieces in all of this, is that if I say to you something like, "Here's the keys to my car, it's in the parking lot, can you bring it around, it's the red one," so you don't have to know what exact shade of red it is as long as I'm confident that it is the one that is red, as opposed to all the other ones. So there's a contrasted meaning and that indicates that often, that's a piece of perceptual language, it describes a color. But it's just good enough for current purposes. So there's a real sense in which language is optimized and that, of course, means that it only gets good enough, it doesn't get better than that. And so the real answer to the question is, what do we actually use language for?

0:29:00.9 SC: Well, yeah. You mentioned in the book that there are something like 2 million distinct shades of color that we can in principle identify as human beings, but no language has 2 million words for colors. And so what you're doing is you're offering an explanation for why that's true, it's because language has no purpose, no reason to be that precise, it would take effort, it would take brain power and cognitive effort to be that precise without any gain in functionality. So it kind of makes sense that language is doing well enough without doing better.

0:29:33.2 NE: That's right. It's a classic trade-off, as you've just described it. It's gonna be diminishing returns very quickly, the more that we distinguish terms for colors. You can go into a paint store and they might have 5,000 words for colors. They have all of these labels, technical terms. The ones I cite in the book I think include Aztec tan and Wing Commander, you can guess what colors they are. You can only guess, they're highly specialized and in that context, the context of ordering paint, you do need to have quite a high level of granularity, but still, and I think this is crucial, you are orders away from the 2 million distinctions that you just mentioned. Even the most fine-grained taxonomy of color that language can handle is miles and miles from what we can actually perceive. And I think it's also worth emphasizing that the color spectrum that we can perceive is also another massive number of orders away from what's actually out there that we could conceivably perceive if we had different kinds of sensory apparatus. And that's... In the book, I describe this as a two-step reduction in our experience of reality.

0:31:08.3 NE: So the first is going from the full catastrophe, all of reality it's out there, we have this radical reduction to just, for example, the visible spectrum of light, just that section which is small. And then you have another radical reduction down to... Well, I mentioned a few thousand and the paint store example but in reality, no language in terms of the literature on color in languages, no language is gonna have more than a dozen terms, what are called basic color terms. There are oftentimes more specific terms like Scarlet or alizarin, but basic terms like red, yellow, blue, green, those will be fewer than a dozen and oftentimes just a handful, three or four or even fewer and it turns out well those are good enough for their purpose in the cultural setting in which they have historically evolved.

0:32:15.4 SC: I like that two-stage reduction that goes on and it makes me bring up one of my favorite topics here, which is emergence, in the sense that this picture that we construct of the world through our language is much more coarse-grained than the world itself is. We're not giving you all the details when we describe the world in language. But it does well enough, so we not only are labeling the world in our language, but we're constructing a model of it, because with the example you already gave of going to get the red car, the reason why you can group together relatively close colors of red under one word is because it functionally describes a world that we can then use to manipulate and predict and so forth. So this is kind of an amazing feature about the world that language can describe it so compactly and efficiently.

0:33:09.1 NE: That's right. And I think one thing I wanna add to that is that often when we read, particularly in cognitive science literature about language, the way of talking about it is pretty much like the way you've just spoken about it, which is to do with reasoning and reducing your amount of processing and prediction, things like that. What is equally important, or I think actually more important, is the way in which these categories are not just good enough for thought, they have to be good enough... What they have to be good enough for is social coordination. And so by that I mean, I might have an incredible sensitivity to different colors in my own personal vision, as we all do, we walk around and we can detect these differences, but they don't... So they are real for us and they're part of our cognition and they're part of what helps us navigate the world. They've been given to us by our biological evolution. And this is where we come back to our evolutionary question, is that the categories in the language are not given to us by the biological evolution, they're given to us by the cultural evolution and what's crucial about that is that words don't just spring directly out of patterns in our minds, they don't spring directly out of concepts that we have in our head or percepts.

0:34:37.7 NE: They come from instances of social coordination, like the example of me telling you which car is mine, that's not an abstract relation between me and a car, it's actually what you might call a triadic relation. So there's me and there's you and there's the car, and the function of language is to get us to align, to coordinate around some bit of reality and language is our little map for that, it's our little ticket for that. And so its job, the categories of language are being focused on very heavily as being all about conceptualization and the mind, which clearly they are, but I think very under-discussed, the social payoffs and the collaborative payoffs of those categories. And again, it's not just in, for example, directing a person to a piece of reality, that's one type of action we do, but also very important is the kinds of actions we do to, as I mentioned earlier, in relation to Whorf, we justify our actions, we try to convince people of what's the right thing to do, we try to persuade people of a certain analysis, and the words that we choose to do that are words that have succeeded in the past, in our kind of cultural milieu. So that I think is one of the crucial things never to forget, is that the evolution of the words themselves is partly due to our cognition, it has to be, but it's equally to do with our social interactions.

0:36:27.3 SC: Yeah, and this might be a tiny bit of a detour, but I was hoping that you could explain, without using pictures, since this is an audio-only podcast, but you give the example of Schelling's map in the book, which I'd never heard of before, but as an illustration of how when we think as human beings, not only... And you're completely right and you're fair, I was talking like a physicist before, describing the world, making a predictive theory of it. But in the social world, we need to not only understand the world and what it's gonna do, but we need to have an image of what's in other people's minds, and this Schelling's map example was a wonderful illustration of how we just use that automatically in maybe a way that is even different between human beings and other species.

0:37:14.6 NE: So Schelling was an economist and I think was most famous for his book, The Strategy of Conflict, also later, Micromotives and Macrobehavior, or is it the other way around? But in the early work on conflict, he emphasized the idea that you are constantly trying to figure out what's the other person gonna do? I'm going to attack your village and so you think I'll come along the road. So I won't go along the road, I'll come over the hill, but you will have thought through that, so you'll expect me to come over the hill, so I'll come on the road and surprise you. So he's thinking through the psychology of conflict and that led him to realize that actually the same kinds of operations are crucial for cooperation as well. You are constantly trying to figure out, "Well, how does the other person understand what's going on here?" So he had a lot of discussion of little tests that he did with people where you ask them... You put two people into two separate rooms and you give them some puzzle. And the puzzle is you tell them that there is someone in the other room who's got the same puzzle and if the two of you can solve it in the same way, then you've won.

0:38:51.0 NE: So one of the examples was the map that you just mentioned, which was in Schelling's book. It was a map of a bit of terrain and you were to imagine that you were a parachutist who was coming down into enemy territory and another person was another parachutist also coming into enemy territory. You both have the same map, you don't know where you've landed, or I guess you know where you've landed, you don't know where the other person's landed, and where should you go to meet each other? So you don't get to communicate beforehand, all you can do is think, "Where will that person go?" But not only that, "Where would they go thinking where will I go?"

0:39:34.5 SC: Just from the map.

0:39:34.9 NE: Just from the map. So you're looking at the map and you're thinking, "Okay, that person knows I'm looking for them, I know they're looking for me, we wanna converge, where would we go?" And the map happens to have a bridge in the center of the image, and that of course kind of skews the solution a little bit, the fact that there's kind of a bit of semiotics in the map that points you to that spot, but you get the solution. There are many others, he gives a long list of these other ones. Another one that I mention in the book is the, what's related to some of these kind of table top economic games; You put two people in two separate rooms, give them $100 and you say, "You split it in some way, and the other person in the other room is gonna split it in some way," you both know that you're splitting it in some way. If you split it the same way, you get to keep the money according to that split. So if I go, "Oh, I'm gonna split 60-40," and you say, "I'm gonna split 50-50," we get nothing. And what happens is that people converge much more often than you would expect by chance on a 50-50 solution.

0:40:48.4 NE: So the upshot of the work on Schelling's maps, or Schelling's coordination problems, is really how people refer to them, is that people are well practiced and really quite good at thinking through, "How would I best coordinate with this person on this problem I currently have without being able to talk about it with them beforehand?" And this is the thing with language that every time we speak to someone, we're in a conversation, and every couple of seconds, there's a new utterance being produced. You don't get to go back and say, well, how are we gonna understand the meaning of this utterance? You just have to imagine, how do people understand this? And that's the source of convention, it's a solution to these coordination problems. So the problems that Schelling was talking about are ones where you don't have a convention and you don't get to communicate in advance. But the way that languages evolve over time is that conventions emerge precisely from repeated solutions to these coordination problems. So if people have succeeded in coordinating in a certain way, they will stick with that solution.

0:42:06.2 NE: In the book I talk about experiments, coordination experiments where, for example, you give a set of tangram figures to two people, and I have these 10 tangram figures in front of me, and all of them roughly have a kind of human form. And you're on the other side of a barrier and I'm told that you have these figures in front of you and I have to tell you what order to lay them out in, so I have to use language to get you to identify which figure I'm talking about. So there's one figure that people, when they look at it, they often say, "It looks kind of like an ice skater and it's got its feet sticking up and leaning over," and people give quite sort of long descriptions the first time around. And when you get people to repeat the task, the same pair of people just do it over and over again, within five or six times, they just have created a vocabulary to refer to these figures so that they've quickly just, instead of having this long description, they just say, "The ice skater." So the creation of convention is really very quick and the reason is, of course, that it has this great pay-off that you just suddenly no longer have to spend all this effort talking and processing language, you just go straight to the label and people can fulfill the task.

0:43:33.8 SC: Is the way in which people have an image of someone else's way of thinking that happens in these coordination problems, how culturally different is it? We had Herbert Gintis on the podcast and he talked about doing game theory experiments or economics experiments in different cultures and getting different kinds of answers. Do you think that the Schelling map problem would be equally solved by anyone from two different cultures?

0:44:00.9 NE: No, I don't. So it really depends on... I mentioned the term common ground, and there are different types of common grounds. So the psychologist, Herb Clark, has written a lot about this and indeed also about the Schelling games. The tangram task was by him and a student. He points out that common ground can be personal, so you have common experiences with people who you know, with family members and friends, and you can invoke that common experience through elliptical references and nods and winks and that kind of thing. You draw on your common ground, your common knowledge with people to sort of minimize the effort you have to go to to communicate. And that's... At a personal level, we recognize that.

0:44:54.7 NE: But then there's this higher level which is cultural common ground and that is... Of course, it includes the, all the words in the language, it also includes this whole set of kind of cultural knowledge, whether it has to do with sports or food or the way people behave in shops and on buses, this is what anthropologists study, it goes on and on and on. Just to get through your day, you're drawing on this incredible amount of cultural common ground. So the coordination problems that are being solved through things like getting on buses and ordering food and these kinds of things are all drawing on what's called cultural common ground and we don't really appreciate... Humans have to learn and learn and learn for years as children and as young people what all of that common ground is in the society that they're living in. So the answer to your question is that, because cultures are very different in terms of their contents, it means that, no, you can't just assume that other people in another cultures share all that common ground and this is the essence of cross-cultural miscommunications that we've all experienced in...

0:46:15.6 SC: Yeah.

0:46:17.7 NE: Different countries. Now that's not to say that there aren't some core commonalities in human society and human thought, and that is part of the goal of cognitive science and of the anthropological component of that. We do expect because we're all the same species that there are going to be certain commonalities, but the question is whether those commonalities come from, directly from our biology or secondarily from our biology.

0:46:48.2 SC: Right.

0:46:48.3 NE: What I mean by that is in a sense of all being subject to the force of gravity and all having human bodies that sneeze or whatever, the certain commonalities that we have and the question then is, "Well, is it because of that commonality that certain concepts inevitably arise in every single culture?" I think that is the likely answer, because it's only through being used in social interaction can these concepts find their way into the language. So you're never gonna get a direct link from the biology to the language that you actually speak, that always has to be delivered through a history of successful solutions to coordination problems using the resources of the language.

0:47:41.4 SC: Yeah, and it seems from examples you quote in the book that there is a lot of commonality, it seems to me, as an outsider, there's a lot of commonality in language, but there will be cultural differences in how we carve up the world into different pieces. One of the examples you talk about is just parts of the body, like a Japanese and Dutch will label the different parts of the body slightly differently and as long as they agree within their language community, that makes perfect sense.

0:48:09.8 NE: That's right. And the question is, is that variation unconstrained? What we find is, what I refer to in the book, is a kind of constrained diversity, and a lot of linguistics has been about trying to determine just what that is, or what the nature of that is. A field of linguistics called linguistic typology is all about trying to find out what is universal to language, what is likely to happen in language, if a language has a certain property, what other properties will it also have? So the kinds of things I talk about in the book are in the kind of semantic domain and they have to do with, if a language has, let's say three color terms, what are they gonna be? And it turns out we can predict fairly well what kind of order a language will begin to sort of make distinctions. And so you can't say beforehand whether a language will have three color terms or 12 color terms, but if it has X number, then there's something about the way in which those systems evolve that allows us to see some kind of regularity. So a lot of sets of semantic distinctions differ just in terms of the number of distinctions that they make. And the question then becomes, "Well, what drives that distinction? What drives the fact that language X has more distinctions and language Y who has fewer, but also what kind of regularities do we see in the way that the space gets carved up?"

0:49:48.8 NE: So I think the crucial thing to take away from this is that nobody would say nowadays, "Anything goes" in a language. So you wouldn't have a color term, a color system in a language that has 12 words for shades of red and one word for every other color. It's just not gonna... It just never works like that. So you get the space being carved up in certain sensible ways, and again, it comes back to functionality and what is useful in terms of coordination. So you might predict, for example, that you live in a place that has a certain kinds of ochre and you use those kinds of ochre for coloration of the body or of clothing or painting on stone, and you would predict that, yes, at a certain level you'd get a special vocabulary for shades of red. You'd also predict that would stand alongside a color system that would flesh out the rest of the color spaces just as well.

0:50:49.0 SC: And so this is where it gets a little bit heavy, I think, heavy in the sense of difficult and interesting but also significant and important. If we do carve up the world, both in how we perceive it and then how we codify it into language, does that have a sort of back reaction on what we see when we look at the world and how we think about it? Does it introduce biases or flaws that are inevitably there because of the way that we think about things linguistically?

0:51:22.8 NE: Yes, I think it does, and I talk about some of the ways in which that happens. So one is in the cognitive realm. For the reasons that we've been discussing, having certain structures in your language determines what you will kind of be less likely to question, what are the descriptions that you'll be satisfied with, what are the ways of talking that will cause you not to wanna inspect reality any more closely, what are the resources that will make you shut off your thinking in some sense. So you definitely see that experimentally there's been interesting work, which I talk about in the book, some of these studies are by Lera Boroditsky and colleagues that look at... Kind of tap into that, but they do so by looking at people's judgments. An example of this is you show people a scenario of somebody at a restaurant, there's a candle on the table, they get up from their seat and the table cloth is tangled up in their clothes and then you tell one group in experiment, she knocked over the candle and she started a fire, and you tell another group in experiment, the candle toppled and a fire started.

0:52:48.7 NE: One of those descriptions is kind of agentive and another description is not agentive. And it turns out that people, depending on how you describe it, even though what they see is the same, they will make different kinds of judgments. They will... In this case, they will literally suggest different amounts of money that the person should be fined for the damage that they caused in the restaurant. What's interesting about those examples, is that they mix something to do with reasoning and decision-making with something that is much more social, which has to do with the giving of reasons for certain judgments that you make. So for every judgment that there is, you're not just making the judgment in your head, you're using the judgment for some social purpose. And in that case, you're using the judgment as a reason for why you, for example, decided to fine this person that much money. So in that case, the reason that you're giving is actually provided by the experimenter, so the experimenter controlled that. The experimenter in one case used the active way of talking about it, in the other case the more passive way.

0:54:12.5 NE: Now, there are linguistic differences here because some languages... So in that case, Japanese is a contrast language because it allows you to be relatively agentive or relatively kind of passive in how you describe scenes, but it has much more of an emphasis towards a kind of a non-agentive depiction of events. And so that actually correlates with different people's judgments about those kinds of scenarios, that people will understand the outcomes to have been less agentive and less the responsibility or less caused by the person who is there as an agent. So those things kind of bleed into social justifications and that's a kind of a lawyer-ish element of language. I think there are some really interesting other cases that I talk about in the book, one I might just mention is the study of emergency calls in a cardiac emergency ambulance unit, so this is at a hospital in Perth, Western Australia. And the call takers were given an instruction, at the beginning of the call, they get the person's name and location, and then they ask, "Tell me exactly what happened." And then the person whose loved one is lying on the floor goes into a kind of a little narrative about what led to the situation. But some of the call takers, without being prompted to do so, they gave a very slightly different wording.

0:55:58.3 NE: Instead of saying, "Tell me exactly what happened," they said, "Tell me exactly what's happened." So they add that little S to what. And in English that's a difference between what you would call a simple past, what happened versus what's happened, is what's called a present perfect. And that second way of putting it focuses on, what is the situation now, whereas the first way of putting it focuses on what led to it. So this actually has a direct impact on how people then respond to the question. S if you ask what happened, they will give a narrative, if you ask what's happened, they will describe what's currently the case. And this study found that that response, which is completely linguistically conditioned, led to differences in the average differences in the time it took to get an ambulance dispatched to go and deal with the emergency. Now, it's not a lot of time difference, but when you're lying on the floor in cardiac arrest, you kind of want things to move along pretty quickly. In answer to your question, I think that kind of example shows that the choice of words has this obviously great consequence for how people reason and respond, and we assume that their behavioral response is kind of a result of their reasoning. But one of the key things to remember here is that not all languages have those same distinctions.

0:57:29.0 NE: That little tense distinction that I pointed to with English is not present in all languages and other languages have different ways of carving up the space of how we talk about time, just as we have if we're talking about color. So over and over again through the thousands of words that you have in the vocabulary of your language, you're setting up this kind of space for nudging people this way and that in terms of, not just how they reason, but I think more importantly how they act and how they act upon their reasoning in the real world.

0:58:05.5 SC: It's fascinating because it makes me think of a recent podcast I did with Judea Pearl who studies causality from sort of network of random variables kind of perspective, and a big part of his research is helping artificial intelligence understand causal relationships between things. But I have the impression that teaching AI [0:58:29.6] ____ causally is entirely in terms of physically what happens, and what you're mentioning is that when we human beings do it, we have this other layer of social meaning that assigns blame or responsibility based on an entirely different set of factors than the computer might even be aware of.

0:58:52.4 NE: Yeah, I agree. And I think causality is really an interesting case. Oftentimes, people will acknowledge that causality is really an interpretation of a relationship between two things, and you could also say it's a description. And why would you describe causes? Well, very often, causes are about justifications or giving reasons. Now, why do humans give reasons? They give reasons because they're trying to justify or defend or rally others into some kind of action, so there's a... I think it's really crucial not to just focus on an abstract relation between two events or two states of affairs, but linguistically, why would anyone ever talk about a cause? Well, it's typically for these social reasons. So I think it's really interesting what you raise about "explainable AI." That term is often used in the press and in the literature these days, there's quite a concern with explainable AI. And I find it fascinating because you're told that AI can't come up with explanations for why it's made certain decisions, so let's say its denied your parole application. A machine-learning algorithm has denied your parole application. We can't understand why this is, and in fact, the algorithm cannot tell us why and no human can understand why. They just know what they put in and, "Look, this is what came out."

1:00:35.6 NE: Now, what's interesting there is that when you... The implication is that humans giving explanations is adequate. The humans giving explanations is acceptable. And I find that wild because humans give all sorts of crazy explanations for what they do, which seldom, I don't know about seldom, but often don't have much relation to what's really going on. Do they really understand why they came to the decision they came to, are they letting on... So to my mind, you might wanna explain a human decision in terms of the neurons that fired, but that's equally impossible to understand as the performance of a deep learning algorithm.

1:01:25.9 NE: So I always worry about this kind of demand that we want explainable AI, insofar as it implies that human explanations are perfectly fine, whereas they're not a lot of the time and the real measure is not what's the quality of the explanation, but do people accept it and move on? And I think that's... The language people use plays a very important role in whether others will accept their explanations and move on.

1:01:58.5 SC: Yeah. And this is... We're getting into the constraints and the foibles or fallibilities that are engendered by our use of language. And another one that you mention in the book, which is maybe related, is how it over-writes our memories. We'll see something, we'll have a picture of what happened, then we'll talk about it, and we remember, if I'm getting this right, we remember our verbalization of it more than we remember what we saw the first time around.

1:02:29.6 NE: That's right. It's a little bit different to what you describe, but basically I think that's right. So there are several different experiments that I talk about that kind of touch on this. Going earlier into the 20th century or first half of the 20th century, there were some really nice studies of getting people to reproduce little abstract line drawings. So you would show people little line drawing, you'd ask them to remember exactly what they see, later on you'd have them draw what they saw with a pencil and paper. And early on, this was all the task involved but people... Experimenters noticed that people were often giving some kind of commentary. So they would say... This is a bit like with the tangram experiment.

1:03:22.1 NE: People would be drawing this little drawing and they'd say, "Oh yeah, it's like a star" or "It's like a bird" or "Oh, it's like eye glasses." And they would use this verbalization to describe what they've seen and then experimenters thought, well, hang on. Maybe people are using those verbalizations as aids to help them to remember. And so various experiments were then done where people were primed with different descriptions of the exact same drawing. So you can take a drawing that has two circles next to each other and a little line in between them with a slight bend and you give one group the instruction, or you just put a label on the drawing, "These are eye glasses." And with the other group, it's exact same drawing, you give a label that says, "Dumbbell."

1:04:15.9 NE: And later when you get people to re-draw these, the ones who were primed with the word "eye glasses" will make that bend in the line a bit more pronounced, that's where the eye glasses hang on your nose, and the other group who were told, "Yeah, this is a dumbbell," would make the line straighter than it was in the original drawing. So they're clearly being affected by the way in which language had kind of got them to construe what they were seeing, even though they were told not to think about the language but to just focus directly on the drawing. So the way in which people describe things will affect their memory in those ways. The overshadowing effect, I'm not sure if you mentioned that term, but from the section where I talk about this in the book there's more recent work about verbal overshadowing, which is similar, but it emphasizes the idea of kind of discarding information because of language.

1:05:15.0 NE: So experiments would include seeing a robbery, a bank robbery. You watch a video of a bank robbery and you see the perpetrator's face, and with one group of people in the experiment you're asked to describe the perpetrator's face as accurately as you can, another group of people in the experiment are given some other tasks, they're not given an opportunity to talk about the perpetrator's face at all, so then they don't use language to talk about it and then later when you give people the chance to identify the face in a line-up of faces, the people who described the face with words are worse at identifying the face that they saw.

1:06:01.9 SC: I see, yeah.

1:06:02.5 NE: And there's been really interesting follow-up cases of that. One I particularly like is looking at a catalog from a furniture store and you have to... In the task, you just see lamps and chairs and you just have to press a button. If it's a lamp, you press a button on the left where it's got lamp written on it and if it's a chair, you press a button on the right that has chair written on it. And that's one group of people in the task. Another group of people don't have to sort them as lamps and chairs, they just say something like, do they like them or do they not like them, or would they want them in their home or not? Some other kind of question. And it turns out that the ones who sort them in terms of the linguistic categories that are available, then later if you show them more pictures and you ask them in the earlier phase of the experiment, "Did you see this exact lamp?" Or "Did you see this exact chair?" The ones who were sorting are worse at doing this task. And the argument there is essentially grounded in the concept of categorization, which is grouping things together and throwing away all the differences between them.

1:07:17.9 NE: So we haven't really talked about this, but language is this massive collection of categories, words for objects, words for actions, all of these things just throw together really very varied bits of reality and group them together for the purpose of coordination precisely because the differences don't really make a difference for the tasks that we use these words for in everyday life. So we don't usually have to solve those experiments in normal life, and of course, psychology experiments are always pushing the boundaries of what we can do and these show that, yeah, there are these foibles and these downsides, but clearly from an evolutionary perspective, they were never enough to get weeded out.

1:08:07.6 SC: Yeah. So if I have thought about these images as lamps or tables, that's what I remember. I do not remember the details as well as if I were just looking them as images?

1:08:19.7 NE: That's right.

1:08:21.4 SC: And this seems like, not to get too dark, but maybe this is one of the punch lines of your book, that these are all handles with which other people can be manipulated or manipulate us, or we can manipulate others in some sense, right? The fact that language serves this social function gives us ways of not merely describing what is going on, but layering the descriptions with slightly normative or judgmental feelings about it, and it seems like this is, again, a universal feature of different kinds of languages.

1:08:57.0 NE: Absolutely. And it's not slightly normative, it's entirely normative, I would say. So I subscribe to the Krebs and Dawkins theory of communication. Krebs and Dawkins did a paper in the 1980s on the nature of animal communication and the whole idea of that was that it's not about conveying information, it's about acting upon others. It's about influencing others. And so instead of walking up to someone and dragging them into a room, you say, "Come inside." [chuckle] And so what you're doing is acting upon them through using sound, or whatever the medium your language uses, and they then, as Krebs and Dawkins put it, what you're doing is exploiting their muscle power for you to somehow benefit and with, of course, a very important caveat, that that benefit is typically a mutual benefit. That's what we wanna mean by coordination.

1:10:08.1 NE: So that's why I'd say, it's not too dark. We shouldn't be thinking of this as a kind of a dark view. You easily can, if you focus on language as manipulative and all about influence and so on. But it's just the nature of communication in a world of mobile agents. You are trying to align, you're trying to coordinate, whether this is self-interested or kind of group-interested or even altruistic, you're still getting things done by acting upon the other in such a way that you then get them to initiate some, either some action or some understanding or what have you. And I think that this is yet another way in which I'm trying to move the focus from thinking about meaning in languages being this thing in the head and this thing to do with imagery and reasoning and so on but more about action in the social world and kind of public events of acting upon others, and then repeatedly getting some roughly predictable kind of responses from them, which then help to condition the stabilization of words within a population and the conventionalisation and so on.

1:11:33.4 NE: So again, the whole thing kind of circles back to what have we used words for in the past, and they should be for mutual benefit. So I'm quite happy to be influenced by other people if this is part of what we do and the life that we lead together, I'm happy to be influenced by others a lot of the time, and I think part of... I should also say that I do think the book gets pretty dark in some places, because a lot of the time we are highly vulnerable to being exploited in ways that actually we don't want to be, so a lot of the kind of nudging that language does is outside of our awareness, we're typically not questioning the way that our language is put to get. We're typically not questioning the ways in which people frame the things that they say, and so this demands a certainly degree of mindfulness and it's tough, 'cause you've got to get the balance between thinking carefully about why did this person say this in that way on the one hand, but on the other hand you just wanna get through a conversation, you just wanna get the thing done that you're trying to get done. And so yeah, it's really about being mindful and thinking critically about what's happening in language in some moderation.

1:12:57.3 SC: One of my previous guests was Julia Galef whose specialty is just trying to think rationally, and one of the things she says is that if she reads a newspaper report or something like that quote someone and the article says, this person admitted that the following thing happened, she just translates that into her brain into, this person said that the following thing happened, because the word admitted adds on this extra layer of, and it was against their interest or something like that, which maybe is cheating or maybe it's just true, but anyway, thinking about it in a more value-neutral way might give us a better, more accurate perspective on it.

1:13:38.5 NE: Yeah, I'm a big fan of Julia Galef and I think that the contrast in the book we're talking about between lawyers and scientists is very much akin to the one that she's set up between scouts and soldiers, or the other way around, that the soldiers are the lawyers who have something they wanna defend and the scouts are the scientists who are trying to rediscover the truth, and I think that her approach, that particular example I think is exactly the kind of thing that we ought to be doing, particularly when we're reading news reports and we're looking at social media and so on, precisely for the reasons that you just suggested. So those are good cases where a little bit of selective language use is introducing these worlds of understanding that you just kinda have to accept as they fly past. And semanticists, if you look at a verb like admit, so one of my old professors, Anna Wierzbicka at the Australia National University, she wrote a book called English Speech Act Verbs. So this refers to verbs in English that refer to things that we do by speaking, so it would be things like admit and say and tell and instruct and all of the words, and infact it's an entire book just about fleshing out the semantics of these speech Act Verbs.

1:15:10.6 NE: There's a lot of them in English. And what's amazing is that for every one, there's a couple of pages of deep complex explication, and so just by saying a word like admit as opposed to say, you really are bringing in this whole kind of scheme, this whole kind of world of understanding, and you touched on it when you said saying something that's perhaps against their will or incriminating towards them and this kind of thing. So I completely agree with that strategy and I think that it is an essential part of the kind of cognitive literacy that we need in this day and age, if only people were more critical and precise in that way, I think we'd all be better off. But what I really value about that type of approach is it's very concrete and it says, okay, when that happens here's a step that you could take. And what I find interesting about that particular example is that when you translate admit into say, what you're doing is you're stepping to a different level of granularity. The language is very hierarchical, so say is a very general term that's gonna be semantically quite under-specified for all of those bits of manning that you get at that more granular level.

1:16:39.6 NE: So the step from admit to say is not just a random step, it's specifically a step from one level in the taxonomy of semantic specificity to another level and then that makes it, I think, certainly a less pre-judging level at which you can try to evaluate what's being said based on it's content rather than the framing that in this case the journalist gives it.

1:17:05.4 SC: And maybe to wind things up, this might be, may or may not be a good thing to do, so you can tell me. But I kinda feel like we should give some credit to the lawyers, not just to the scientists. Like the fact that we use language, not just to accurately describe the world, but to persuade people or to advocate for certain positions is not entirely a flaw or not even at all flaw, it is part of what we need to do and want to do. We want to persuade people of our political side, we wanna persuade people to take collective action for the betterment of the world and so forth. We want to let people know how we feel about them and all these other non-descriptive things. What is the kinds of insights in the language and its origins as social communication? Tell us about those aspects of language. What is the good side of being a lawyer and advocate?

1:18:06.8 NE: I think the things you've just mentioned indicate that, or the way in which we wanna understand the things you've just mentioned is to say that these lawyerly aspects of language features, not bugs. I would wanna say that this is in line with work on reasoning more generally by Dance Berber and Hugo Mercier recently, a couple of books where they've said, look, reasoning is supposedly flawed, we have all these cognitive biases, but if you think about them not as mechanisms for arriving at the truth but rather things like ways to construct convincing justifications, then the whole thing makes sense, and I think with language it's the same thing.

1:18:58.2 NE: So I definitely don't wanna be saying that the lawyer approach is bad in some intrinsic sense, quite the opposite. This is precisely what language is for, and in my view, the fact that you can kind of speak scientifically about language, that's more of the epi phenomenon, that's more the sort of excerptation that we've done and we've gone, oh my God, we can use language to do this other thing, we can use it to coordinate around new truths and we can use it to kind of get together to find out new facts and that's a great payoff, but I think fundamentally, certainly most of what we use language for is precisely for all of those functions of persuasion and so on.

1:19:46.4 NE: I think one thing that's really important to say here is that there is inevitably a conflict, I mean not inevitably, not in every case. Oftentimes there's no conflict between the truth and what we're trying to convince people of, but I don't think there's a necessary alignment, obviously there isn't a necessary alignment. So if you happen to be a wonderful story teller, if you happen to be a wonderful wordsmith, that's great, you'll convince people of stuff, but how do I know that you have the truth? How do I know that the thing you're trying to convince me of is really true, is really correct?

1:20:26.8 NE: Well, oftentimes it doesn't maybe matter that much, it all really depends on what this is about, but as you know, certainly better than I, you want to get at the truth, because knowing the truth has these untold consequences. If you're going to design technology that will send us to the second LaGrange point or what have you, you're not gonna do that just by being the better story teller, we don't want to just have the better story teller, maybe we want a great storyteller in certain context, say for example if we're trying to get the government to give us billions of dollars to make that happen, then maybe in that context, yeah.

1:21:13.5 NE: You want a great lawyer, a great soldier, a great storyteller, but you wanna be sure that the truth is getting a look in and that's the key. So I do think there's a really interesting ethical question there and that is that, any lawyer in this metaphorical sense that we're exploring here really has an ethical duty not to be trying to convince people of falsehoods. And so the lawyer and the scientists need to be working together and to my mind, ultimately the lawyer is accountable to the scientist, because if we get reality wrong, we die.

1:21:56.9 SC: [laughter] I'm not gonna disagree with anything you just said. It does make me think a little bit about the history of Lysenkoism in the Soviet Union, where they came up with a version of biology that was more flattering to the government and it can't last because you can't fool your way, you can't tell a better story about biology that is in conflict with the truth.

1:22:21.3 NE: That's right. So it also reminds me of Solzhenitsyn's book, The First Circle, where you have, in fact, there's a linguist in the first circle of hell, namely the least worst re-education camps in the Gulags, and they were supposed to come up with speech technologies that could do what apps does these days, but code your speech and not be vulnerable to spies and all of that kind of thing. And in the book the scientist are saying, oh yes, yes, we're nearly ready with the device, we're just doing a few last kind of tweaks, but of course, they had no idea how to do this thing that Stalin demanded to be done, and they were just stalling for time the entire time. So literally whether they lived another day all came down to sort of how convincing they were as to the progress of this technology but in the end, of course, it's not gonna work out.

1:23:33.8 SC: The real world does provide even language with constraints that it cannot completely work around. That's a good lesson. So Nick Enfield, thanks so much for a very informative podcast here.

1:23:44.3 NE: Thanks very much for having me.

[music]