Objection 2: you compare exactly one pair of languages, Mandarin and Hawaiian; as it happens, my guess is that in broad terms the same pattern holds quite generally, but you really need more evidence. Objection 3: the difference you look at between these languages is in phonology, not in vocabulary; it’s not obvious that the same goes for both.
This example was an intuition pump, to help identify the basic principle at play, not meant as a knockdown proof. But it’s no accident that the first two languages I thought of (which I chose as extremes of the richness vs speed spectrum) illustrate the point I made. If you do the same calculation for Spanish, Japanese, German, or any other language, we should expect to find the same pattern, that the bits per second comes out to the same cadence.
Objection 4: in your comparison, the “richer” language is still faster.
No. If you look again at the calculation of my rough estimate, you will notice that I didn’t use the actual speaking speed of Hawaiian, since I couldn’t find a good number for it, and instead plugged in the cadence of Japanese, which is slightly slower than Hawaiian, so the number I provide is an underestimate of the actual information speed of Hawaiian. Furthermore, even if the number wasn’t an underestimate, it’s not clear to me that the difference between 64.2 bits / second and 59.6 bits / second is statistically significant. (I don’t know the error bars, since my source was a secondary source and didn’t identify the paper where they got their numbers from)
Objection 5: you consider only spoken language; there might be similar effects for writing and typing, but it’s not clear that they’re the same.
I don’t see any reason to assume they’d be different, unless you know of a reason to think they won’t be. My theoretical justification (which I’m aware you aren’t yet sold on, but which generated the test of Hawaiian vs Mandarin that matched the prediction it made) holds just as well for written language as speech
Objection 6: the comparison you offer as evidence is between average data rates, but the likely effect of losing bits of vocabulary is to make particular things harder to express; if communicating simple things becomes faster and communicating complex things becomes slower, this may not show up in such comparisons but could be a big deal.
This doesn’t seem correct to me. For example in Sona ‘momentum’ would be ‘ganru’: “matter” + “movement”, and ‘derivative’ would be ‘nuakiagu’: “change” + “speed” + “trend”, literally “rate of change”. In this case, ‘ganru’ is even shorter than the word it replaces, and ‘nuakiagu’ takes me about as long to say ‘derivative’ when I pronounce them in paces I find natural for each language, maybe even a little less.
In natural languages, simple words tend to be said quite quickly, while more complex words take more time to say—in general because they’re longer, or have more complex sequences of sound, while simple, common words tend to be shorter and easier to say. Similar effects should happen in a polysynthetic language (such as Sona)
If you don’t have a word for “momentum”, that doesn’t stop you talking about momentum, but it makes it clumsier, and it makes higher-level thinking about related topics much clumsier.
Dancers have their own word for momentum: body-flight. ‘body-flight’ serves just as well to enable higher-level thinking about related topics, since once you have gotten familiar with the phrase, the brain treats it as one word, not as “body” + “flight”, except in addition to being able to serve all the same purposes ‘momentum’ serves, its meaning can also be easily inferred by someone who has never heard the word before. If you want to talk about angular momentum, you could just as easily say “angular bodyflight” in a world where physicists used the word ‘bodyflight’.
If I had to say “mass times velocity” and “change per unit time at infinitely small scales” instead of “momentum” and “derivative”
Of course you don’t want to use a definition as the handle you use for a word! But the thing is, you don’t have to do that (as illustrated above by ‘bodyflight’ and ‘momentum’).
I do not agree that the “obvious naive explanation” explains nothing, and so far as I can see you’ve offered no argument to support that criticism;
What? My previous comment was exactly an argument to show that.
What extra effort do I put in when I say “momentum” instead of some circumlocution?
You first had to learn the word ‘momentum’, and second, you had to store this sound and its meaning in the brain. To our concious selves, this doesn’t feel like work, but from a biological perspective, our brains have to do a lot of work to make that happen. In contrast, with ‘bodyflight’, you still have to be exposed to that fixed term before you can use it, but you can get a gist for what it means even if you have never heard it before; and your brain has to store less information to be able to go from the concept of bodyflight to the word ‘bodyflight’, because instead of storing an entire sequence of sounds, it can just point to words that it has already stored.
I haven’t yet read Hanson&Simler, but I have read a fair bit of other Hanson and I am aware that, crudely caricatured, he claims that everything is signalling.
The first part of the book delves into why the brain does stuff (especially, but not exclusively, signalling) subconciously. I will also like to note that as far as the two authors go, I personally hold more esteem for Kevin Simler (who writes at Melting Asphalt) than Robin Hanson, I feel that Kevin’s biological explanations of many human phenomena delve into very interesting dynamics that have changed how I think about the species homo sapiens. I particularly like his blogpost Music in Human Evolution, written 5 years before The Elephant In The Brain was published, as a showcase for the kind of thinking Kevin does.
On objection 2: Your expectation of finding the same pattern whatever two languages you compare is not evidence that in fact it holds. For the avoidance of doubt, I do in fact expect a weak form of the pattern to hold near-universally: languages with, say, fewer bits needed to specify each phoneme will tend to be spoken with those phonemes occurring more rapidly. But you’re making a substantially stronger claim—that this will occur to just the extent required to hold the rate of information transfer constant, and that this will apply when there are vocabulary differences as well as when there are phonological differences—and it seems to me that when making so strong a claim you really ought to provide more evidence. (Or else present it as a conjecture rather than a factual claim.)
On objection 4: ah, so in fact I should have said not that you did the calculation only for one pair of languages, but that in fact you didn’t do the full calculation for any pair of languages!
On objection 6: I’m not sure I understand your objection to my objection :-). If “ganru” and “nuakiagu” actually express the same notions as “momentum” and “derivative” then all that means is that Sona does in fact have those words (I don’t see that the fact that they are constructed from simpler parts is significant), and then I don’t see what a comparison between Sona and English tells us about differences in vocabulary. (Of course it may only “have those words” for Sona speakers who have learned some physics and mathematics, but that’s true of English too: to people without the relevant technical knowledge, “momentum” is the name of a splinter group in a UK political party and “derivative” means “copied from other works” or “one of those weird finance things”.)
If you were only ever claiming that a language that lacks certain terms can be extended by adding new words that mean the same as those terms do … well, sure, I agree, but I thought you were saying something much stronger than that.
Much the same goes for “body-flight”. If that means (approximately?) the same thing as “momentum” then what that means is that ballet dancers have discovered some of the same ideas as physicists, and like physicists have coined a word for it. Again, the fact that it’s a word made out of smaller meaningful parts doesn’t seem relevant to me. Unless you’re suggesting, here or in the case of Sona, that just from those parts you can work out the full meaning, so that if you use a smaller language then you never need to learn about momentum because you can just slam together “body” + “flight” and get the right concept by magic. But I bet you aren’t suggesting that, because it seems to me very obviously not true.
What? My previous comment was exactly an argument to show that. [that my “obvious naive explanation” explains nothing—gjm]
Well, as I said, I don’t see anything in that comment that looks to me like an argument showing anything of the sort. Of course it’s entirely possible that you did in fact make an argument, perhaps a very strong one, with that conclusion, and I just failed to grasp it. (The way it looks to me is that you made some non sequiturs.) If you want to persuade me, then I think you will need to make your argument clearer and more explicit. (Of course you are in no way obliged to make that effort.)
You first had to learn the word ‘momentum’, and second, you had to store this sound and its meaning in the brain. [this is answering my question of what extra effort I need to expend on account of saying “momentum” rather than using a circumlocution—gjm]
OK, sure. But those are both one-off extra efforts, and (so it seems to me) this effort is amply repaid by the reduced effort every time I need to use the concept thereafter. (Compared, again, to a hypothetical situation in which I don’t have a word for “momentum”. It seems a bit as if you’re now shifting to what seems to me an entirely different claim, namely that we would do better with a different word for “momentum”, one whose origins are more transparent. That might be true but if it has anything to do with what you were originally saying, I don’t see what.)
So it looks to me as if when I learned the word “momentum” I was paying an immediate price for a larger future benefit. This absolutely is a thing human brains do all the time.
Not every instance of learning a new word will end up actually being a benefit on net. (To take another technical example, once upon a time I learned the meaning of “regular” when applied to a topological space. So far as I can recall, I have never once had a need to use that term or the concept it names, and I probably never will again.) But it seems to me that much vocabulary-learning has positive expected net benefit.
Again, it’s possible that the case of technical terminology is misleading; learning the meaning of “luxuriant” isn’t much like learning the meaning of “momentum” and its benefits are different. So, while I think it’s very obvious that “momentum” pays its way, I wouldn’t make nearly so strong a claim for “luxuriant”. But I think that, to put it mildly, it is not obvious that having more words has insufficient non-signalling value to explain the fact that lexicons grow, and I am never impressed by “I am not convinced that X is practically useful, therefore X must really be all about signalling”, which is what it seems to me your argument comes down to: there are just too many guesses and gaps in the chain from “languages with richer phonology tend to be spoken slower”, with which I do agree, to “languages with richer phonology are spoken exactly slower enough to cancel out the difference in information rate” and then to “the same applies to languages with richer vocabulary” and then to “and the same goes for thinking as for communication” and then to “and this doesn’t merely hold on average, it holds for every specific case”—which is the point at which I think you’d have a reasonable basis for claiming that there must be some “non-functional” explanation for large vocabularies, though not necessarily for identifying signalling as the specific best explanation.
I am never impressed by “I am not convinced that X is practically useful, therefore X must really be all about signalling”
I don’t mean to claim it is 100% necessarily about signaling, however I do mean to claim that A) there’s a solid argument to believe that signaling plays a role, and B) that the “naïve obvious” answer has very little to do with it. (Regardless of whether you are convinced, this is the main claim of the post, and I stand by this claim) There could very well be other reasons that I haven’t considered which make a large vocabulary useful that don’t have to do with signaling, I don’t know.
I am aware that I haven’t proved beyond a shadow of a doubt that my claims are true, but I have given both a theoretical justification and actual examples that illustrate something that is interesting (according to me) and counterintuitive, which is more than enough to justify making a post here.
Storing things in memory isn’t a one-off cost, since you need to keep it there, which takes up space, and I believe a non-zero amount of maintenance in the context of the human brain
This example was an intuition pump, to help identify the basic principle at play, not meant as a knockdown proof. But it’s no accident that the first two languages I thought of (which I chose as extremes of the richness vs speed spectrum) illustrate the point I made. If you do the same calculation for Spanish, Japanese, German, or any other language, we should expect to find the same pattern, that the bits per second comes out to the same cadence.
No. If you look again at the calculation of my rough estimate, you will notice that I didn’t use the actual speaking speed of Hawaiian, since I couldn’t find a good number for it, and instead plugged in the cadence of Japanese, which is slightly slower than Hawaiian, so the number I provide is an underestimate of the actual information speed of Hawaiian. Furthermore, even if the number wasn’t an underestimate, it’s not clear to me that the difference between 64.2 bits / second and 59.6 bits / second is statistically significant. (I don’t know the error bars, since my source was a secondary source and didn’t identify the paper where they got their numbers from)
I don’t see any reason to assume they’d be different, unless you know of a reason to think they won’t be. My theoretical justification (which I’m aware you aren’t yet sold on, but which generated the test of Hawaiian vs Mandarin that matched the prediction it made) holds just as well for written language as speech
This doesn’t seem correct to me. For example in Sona ‘momentum’ would be ‘ganru’: “matter” + “movement”, and ‘derivative’ would be ‘nuakiagu’: “change” + “speed” + “trend”, literally “rate of change”. In this case, ‘ganru’ is even shorter than the word it replaces, and ‘nuakiagu’ takes me about as long to say ‘derivative’ when I pronounce them in paces I find natural for each language, maybe even a little less.
In natural languages, simple words tend to be said quite quickly, while more complex words take more time to say—in general because they’re longer, or have more complex sequences of sound, while simple, common words tend to be shorter and easier to say. Similar effects should happen in a polysynthetic language (such as Sona)
Dancers have their own word for momentum: body-flight. ‘body-flight’ serves just as well to enable higher-level thinking about related topics, since once you have gotten familiar with the phrase, the brain treats it as one word, not as “body” + “flight”, except in addition to being able to serve all the same purposes ‘momentum’ serves, its meaning can also be easily inferred by someone who has never heard the word before. If you want to talk about angular momentum, you could just as easily say “angular bodyflight” in a world where physicists used the word ‘bodyflight’.
Of course you don’t want to use a definition as the handle you use for a word! But the thing is, you don’t have to do that (as illustrated above by ‘bodyflight’ and ‘momentum’).
What? My previous comment was exactly an argument to show that.
You first had to learn the word ‘momentum’, and second, you had to store this sound and its meaning in the brain. To our concious selves, this doesn’t feel like work, but from a biological perspective, our brains have to do a lot of work to make that happen. In contrast, with ‘bodyflight’, you still have to be exposed to that fixed term before you can use it, but you can get a gist for what it means even if you have never heard it before; and your brain has to store less information to be able to go from the concept of bodyflight to the word ‘bodyflight’, because instead of storing an entire sequence of sounds, it can just point to words that it has already stored.
The first part of the book delves into why the brain does stuff (especially, but not exclusively, signalling) subconciously. I will also like to note that as far as the two authors go, I personally hold more esteem for Kevin Simler (who writes at Melting Asphalt) than Robin Hanson, I feel that Kevin’s biological explanations of many human phenomena delve into very interesting dynamics that have changed how I think about the species homo sapiens. I particularly like his blogpost Music in Human Evolution, written 5 years before The Elephant In The Brain was published, as a showcase for the kind of thinking Kevin does.
On objection 2: Your expectation of finding the same pattern whatever two languages you compare is not evidence that in fact it holds. For the avoidance of doubt, I do in fact expect a weak form of the pattern to hold near-universally: languages with, say, fewer bits needed to specify each phoneme will tend to be spoken with those phonemes occurring more rapidly. But you’re making a substantially stronger claim—that this will occur to just the extent required to hold the rate of information transfer constant, and that this will apply when there are vocabulary differences as well as when there are phonological differences—and it seems to me that when making so strong a claim you really ought to provide more evidence. (Or else present it as a conjecture rather than a factual claim.)
On objection 4: ah, so in fact I should have said not that you did the calculation only for one pair of languages, but that in fact you didn’t do the full calculation for any pair of languages!
On objection 6: I’m not sure I understand your objection to my objection :-). If “ganru” and “nuakiagu” actually express the same notions as “momentum” and “derivative” then all that means is that Sona does in fact have those words (I don’t see that the fact that they are constructed from simpler parts is significant), and then I don’t see what a comparison between Sona and English tells us about differences in vocabulary. (Of course it may only “have those words” for Sona speakers who have learned some physics and mathematics, but that’s true of English too: to people without the relevant technical knowledge, “momentum” is the name of a splinter group in a UK political party and “derivative” means “copied from other works” or “one of those weird finance things”.)
If you were only ever claiming that a language that lacks certain terms can be extended by adding new words that mean the same as those terms do … well, sure, I agree, but I thought you were saying something much stronger than that.
Much the same goes for “body-flight”. If that means (approximately?) the same thing as “momentum” then what that means is that ballet dancers have discovered some of the same ideas as physicists, and like physicists have coined a word for it. Again, the fact that it’s a word made out of smaller meaningful parts doesn’t seem relevant to me. Unless you’re suggesting, here or in the case of Sona, that just from those parts you can work out the full meaning, so that if you use a smaller language then you never need to learn about momentum because you can just slam together “body” + “flight” and get the right concept by magic. But I bet you aren’t suggesting that, because it seems to me very obviously not true.
Well, as I said, I don’t see anything in that comment that looks to me like an argument showing anything of the sort. Of course it’s entirely possible that you did in fact make an argument, perhaps a very strong one, with that conclusion, and I just failed to grasp it. (The way it looks to me is that you made some non sequiturs.) If you want to persuade me, then I think you will need to make your argument clearer and more explicit. (Of course you are in no way obliged to make that effort.)
OK, sure. But those are both one-off extra efforts, and (so it seems to me) this effort is amply repaid by the reduced effort every time I need to use the concept thereafter. (Compared, again, to a hypothetical situation in which I don’t have a word for “momentum”. It seems a bit as if you’re now shifting to what seems to me an entirely different claim, namely that we would do better with a different word for “momentum”, one whose origins are more transparent. That might be true but if it has anything to do with what you were originally saying, I don’t see what.)
So it looks to me as if when I learned the word “momentum” I was paying an immediate price for a larger future benefit. This absolutely is a thing human brains do all the time.
Not every instance of learning a new word will end up actually being a benefit on net. (To take another technical example, once upon a time I learned the meaning of “regular” when applied to a topological space. So far as I can recall, I have never once had a need to use that term or the concept it names, and I probably never will again.) But it seems to me that much vocabulary-learning has positive expected net benefit.
Again, it’s possible that the case of technical terminology is misleading; learning the meaning of “luxuriant” isn’t much like learning the meaning of “momentum” and its benefits are different. So, while I think it’s very obvious that “momentum” pays its way, I wouldn’t make nearly so strong a claim for “luxuriant”. But I think that, to put it mildly, it is not obvious that having more words has insufficient non-signalling value to explain the fact that lexicons grow, and I am never impressed by “I am not convinced that X is practically useful, therefore X must really be all about signalling”, which is what it seems to me your argument comes down to: there are just too many guesses and gaps in the chain from “languages with richer phonology tend to be spoken slower”, with which I do agree, to “languages with richer phonology are spoken exactly slower enough to cancel out the difference in information rate” and then to “the same applies to languages with richer vocabulary” and then to “and the same goes for thinking as for communication” and then to “and this doesn’t merely hold on average, it holds for every specific case”—which is the point at which I think you’d have a reasonable basis for claiming that there must be some “non-functional” explanation for large vocabularies, though not necessarily for identifying signalling as the specific best explanation.
I don’t mean to claim it is 100% necessarily about signaling, however I do mean to claim that A) there’s a solid argument to believe that signaling plays a role, and B) that the “naïve obvious” answer has very little to do with it. (Regardless of whether you are convinced, this is the main claim of the post, and I stand by this claim) There could very well be other reasons that I haven’t considered which make a large vocabulary useful that don’t have to do with signaling, I don’t know.
I am aware that I haven’t proved beyond a shadow of a doubt that my claims are true, but I have given both a theoretical justification and actual examples that illustrate something that is interesting (according to me) and counterintuitive, which is more than enough to justify making a post here.
Storing things in memory isn’t a one-off cost, since you need to keep it there, which takes up space, and I believe a non-zero amount of maintenance in the context of the human brain