The problem with the human body is that is extremely hard to share information. We can absorb gigabytes of information through our eyes and other sensory organs, but we can only emit a few bits of information. When we communicate, we are essentially trying to share our thoughts through a very narrow straw that limits the flow of information. The result of this bottleneck is language. Language requires lots of context, common understanding, and being able to view the world from the other person's perspective to make sense of the tiny amount of information that one person is sharing with another
Note that the average speed of communication in different languages is the same, even though they differ in verbosity and expressiveness. This is why Spanish speakers seem to speak so fast - Spanish has more syllables on average.
So the bottleneck would appear to be, not the physical limitations of communication, but the rate at which we can structure information to be consumed by another mind.
Rather like how the limiting factor in coding productivity is speed of thought, not speed of typing.
Spanish also has heavy syllable merging, even across word boundaries. (Vowels merge together and hiatus turns into di- and triphthongs.) This reduces the number of actual syllables in speech.
When simultaneously translated English to Spanish, my translators complained because I speak fast, and they'd tell me they can't keep up. When I slow down enough, I end up losing my train of thought, so I had to practice speaking slower.
Also it might differ Mexican Spanish vs Castilian, where I didn't get much of a complaint by Castilian translators.
Your scenario introduce 'translation' as an additional task. I wouldn't be surprised if that ends up being the bottleneck, rather than anything specific to either the input or output language processing.
For me, it's not so much losing my train of thought but it drops me out of any sort of real conversational mode. At which point, I can feel like I'm giving up any sort of real engagement and just saying words.
(Though, in my experience, simultaneous translators I've worked with and listened to are pretty good. They seem to usually do a better than decent job even for speakers going along at a "normal" pace.)
> So the bottleneck would appear to be, not the physical limitations of communication, but the rate at which we can structure information to be consumed by another mind.
Yet the way in which we must structure information surely plays a role in how quickly we can do it, right? I can certainly read faster than I can speak, and I'm also confident that I can read with full comprehension faster than I can listen with full comprehension to someone speaking very fast, like at the end of a medicine commercial or the Micro Machines guy.
Take for example a physics teacher. They are fond of drawing diagrams because it's a quicker and easier way to refer to the concepts in play than relying on words alone. If we had the capability to project the imagery that accompanies our thoughts so that others could see it, I suspect we'd have to do a lot less talking. And in fact we practice this already with feelings and body language--most of us know when we've committed a faux pas before anybody says a word about it.
The same goes for programming: my thought is "i want to get all of the user's preferences", which is surely faster than finding where to put the query, when to call it, and then testing its correctness. This is one of the things that Alan Kay and Bret Victor go on about--getting the machine to be usable at a level that is closer to the problem space than to machine code. I understand what you meant, that typing at 300 WPM won't help you to design a system. But a visual development environment that actually works probably would, because it would let you work at a more conceptual level most of the time instead of typing magic words into a terminal.
"but we can only emit a few bits of information" [per second?]
I don't think that's true. Physical motion and even sound making in total involves a human being producing a lot more than a few bits of information per second - there are plenty of recordings of people singing, say, that can't compressed to that bitrate.
The only thing that's true is that we can pretty much only emit and absorb language information at a fairly low bitrate.
The simple explanation for this is that a language stream is something like a computer program - difficult to either produce or absorb.
And of course, language with complex structures is fairly unique to humans and provides some amazing things - the ability to help organize society and our own activity on a larger scale, to plan ahead, and etc, and etc, etc....
Edit: Good general discussion of these issues can be found in Tor Norretranders' book The User Illusion. Highly recommended.
There's an interesting video on the fastest meaningful musical tempo or speed, which looks at the interonset interval. It turns out that between 100ms and 50ms seems to be that limit: 10 to 20 beats per second. Or, at the low end, 20 Hz.
Which is the point at which a series of beats becomes a tone.
If you extend this out to a full day, and consider each beat a bit, via GNU Units:
You have: 20 bit/s
You want: M bit /day
* 1.728
/ 0.5787037
That is, the human auditory system is capable of processing at best 1.7 megabits of beat-based information per day. That's only 0.2 MB (bytes), or a small fraction of an old-school 3.5" floppy disk.
Mind, that's individual beats, which isn't how we convey information, but it's an interesting data rate limitation. Language encodes at different levels, and arguably people might be able to operate at higher levels.
For spoken English, most of us can accommodate roughly 100 words/min. A typical college-educated adult can read 300 words/minute, and at the high end, perhaps double that. (The less literate are closer to the 100 words/min of spoken speech.)
If a word is considered as 8 bytes (7 characters and a space), then a sustained maximum-rate reading pace would allow for just shy 7 MB per day of information consumed.
Speaking? About 1.1 MB/day.
Both of those assume a full 24 hour day without break, interruption, or sleep.
We can absorb more information in the form of complex tonal or visual signals, but I'd argue that the retention of that information is highly limited.
I'd add that the information channel is not the meaning channel.
Any information that a human processes contains a huge amount of prior knowledge and context. Just compare the size of this comment in text bits vs. the (hypothetical) size of the knowledge you need (and have) in order to understand it.
If anything, saying that we communicate only 0.2MB per day illustrates just how huge this prior knowledge must be.
From an information theory perspective human communications might better be modeled as a compressed data stream with a very very big lookup table.
That's closer to the reality, yes, though I don't think lookup-table is entirely the right concept either. Useful, yes, but doesn't quite generate the right mix of concepts of neurons, activation thresholds, fuzzy activation, multiple access paths, etc., etc.
I'm not focusing on the overall channel bandwidth because I think it tells the full story, but it does tell an important story, and in particular, it shows just how constrained that channel really is. If we're sucking (or blowing) through a straw, then there's only so much information as can be mediated.
This becomes particularly significant when you look at the amount of crap that people are flooded with, and realise how much this cuts down on the ability to process significant and meaningful information.
> It's literally true for what we are doing right now in this moment: emitting information through a keyboard.
The claim is that we _can't_ emit more than a few bits per second, and the rebuttal is that we can. Providing an example that's under the threshold is completely irrelevant.
To say this is to assume that no information is communicated through facial expressions, body language, tone, rhythm, etc. Furthermore, assuming all relevant information is contained in a person's words alone, it's also impossible to say that a particular bit rate representation of an utterance faithfully reproduces all of that information.
Non-verbal communication is definitely something for GP to consider, but doesn't change their point very much. The amount of information you can convey directly through your body is still a tiny fraction of what goes through your mind. And guess what, you still need tons of context to decode body language, rhythm, etc.
Perhaps if you measure against the text representation in ASCII, but it's not necessarily easy to quantify all of the other meaningful information that comes from tone, volume, accent, gestures, etc.
Such is to say that monkeys and birds can effectively transmit no information, which is not true.
Nit: we don't absorb gigabits of information. Perhaps there is that much bombarding our senses St any given moment but the amount that actually makes it into our brain is far less, probably on the order of a hundreds of bits/sec max
I guess I have a question of what OP meant when saying your sensory organs "absorb" gigabits per second of information. In my reading, I took that to mean sensory input that actually makes it into the brain. After all, asphalt absorbs that same information, but I don't think that is a particularly interesting use of the infirmation.
In another reply I quoted research that shows that only ~10 kbits/sec make it even into layer 4 of v1 from the optic nerve.
If you have anything talking about spinal processing I'd love to read it.
Not sure what you're looking for specifically, but a quick search for Spinal Nerve (or bundle!) Processing will give you a ton of results. Restrict yourself to pub med or google scholar, because there are a lot of mediocre results.
I think the main confusion here was over the definition of 'absorb'. Your version is much narrower than I'd use. The sensory molecular pathways and accompanying structure of nerves themselves produce significant processing right at the outset which your definition would entirely discount. There is no reason to discount what would otherwise become 'pre-processing' in your model.
"the amount that actually makes it into our brain is far less, probably on the order of a hundreds of bits/sec max"
ABCLAW:
"Even the spinal nerve bundle processes orders of magnitude more information just in keeping you balanced on your feet or chair."
Does that information actually make it to the brain? I thought sending a signal to the brain to process it there and then send back balancing signals would take too much time to keep you on your feet (at around 100m/s, a nerve signal would take about 1/50 of a second to move from legs to brain and back)
There also is evidence that your spine can do quite a bit of the muscle control needed for balancing on its own. See knee reflex and spinal cat (completely cut the spinal cord of a cat in the neck; notice that cat is paralyzed. Hang cat in harness, with feet on a treadmill. Turn on treadmill; cat walks (possibly only after some training; I don't know the details). Known since the 1950's. See https://www.scientificamerican.com/article/spinal-cats-walk/ (paywalled)).
Not all of the information processed in the main spinal nerve bundles make it to the brain. Note that I'm not replying to the 'reaches our brain' portion of his statement with my counter-example. I think that's a very arbitrary place to delimit the human information processing system.
Yes, it is in your brain. But if you think that your visual cortex receives anything akin to a high fidelity "picture", then you are mistaken. Regarding lossy image compression, you're free to look at something briefly and try to redraw it from scratch. Even if your drawing skills are perfect, you will find that your picture will be filled with confabulations and stereotypes.
> Because of a limited number of axons in the optic nerves (approximately 1 million axons in each) only ~6x10^6 bits/sec leave the retina and only 10^4 make it to layer IV of V1 [22,23]. These data clearly leave the impression that visual cortex receives an impoverished representation of the world, a subject of more than passing interest to those interested in the processing of visual information [24]. Parenthetically, it should be noted that estimates of the bandwidth of conscious awareness itself (i.e. what we ‘see’) are in the range of 100 bits/sec or less [22,23]
Considering the optic nerve can only tramsit ~6 Mbits/sec, and layer 4 of v1 in your visual cortext only gets ~10 kbits/sec, I would say you are wrong.
To test this, try seeing if you can distinguish HD from SD without the use of your fovea. You have one highly detailed section of your eye, and the rest is low quality input which is "upgraded" through previous sightings with the fovea and your brains knowledge of how it thinks the world works.
I would argue that you're confusing "brain receiving complete information" with "brain doing inference on what information it does receive to get you close enough". You don't need the full screen to tell that the image is SD vs HD, you just need to glance at a small segment. That's not 100s of mb/sec. You know that the rest of the video is happening in your peripheral, and maybe you can tell what's going on. But in actuality you're receiving a very low bit rate version of that, and your brain is simply telling you "X & Y are happening in HD in your peripheral" and it's convincing enough.
I've been tackling this from a different PoV (see recent comments) of audio bitrate sensing (20 Hz is the beat/tone boundary, where beats become tones, corresponding to about 0.2 MB/day of data acquisition), or of speaking and reading rates.
At 100 words/min spoken speech, humans can transmit about 1.1 MB/day sustained (24 hours without interruption).
At 600 words/min reading rate (a very high rate, roughly double the college-educated adult average), a human can absorb a maximum of less than 7 MB/day.
If you think that the information transferred by human speech consists only of the words spoken, you have a rather shallow view of language. You're also processing information taken from pace, pitch, timbre, intonation, volume, timing and direction (among other things). The written word is to speech what choreographic notation is to dance - it's enough to give you the gist, but you wouldn't mistake it for the real thing.
Each beat is probably better considered a symbol, so by your analysis the human audio channel is around 20 baud not 20 bits/s. The symbol size obviously has a large impact on the information bit rate.
We can estimate the size of a symbol though:
13b for primary tones(20khz range at about 3Hz resolution) + 4b for rising/falling/etc tonal patterns + 7b for volume (1dB detection over 120db range) + (42 phonemes in english so round up to 256 for convenience) 8b for phoneme patterns + 2 ears preprocessed for 2d directions is maybe 4b direction and 4b distance gives a total of around 40b per symbol.
So with that wild approximation, it's at 800 bps, or 8.64 MB / 24h
A fair argument, and yes, that expands the channel somewhat.
(My first posts on seeing this concept, elsewhere, had confused bits and bytes, so I ended up with 1.7 MB rather than 0.2 MB.)
But the point remains that that what we're discussing is almost certainly a maximum upper bound on human conscious information processing capacity. And that the values are ... pretty limited, relative to the amounts of information our machines can fling at us.
8.6 MB/day means that on average we're looking at 2/3 of that for waking hours or 5.7 MB. If you spend half your waking hours at work, that's about 2.9 MB of information acquired per day at work, max, over audio track.
And that's the tail-end of a 3.5 billion year R&D effort.
Oh_sigh says:"...we don't absorb gigabits of information."
We're talking bits/second (bandwidth) here, not simply bits. So given that...
The human visual pathway has a bandwidth of about 10 megabits per second, so we build the current "view" somewhere in our brain and constantly update it with the stream of data from the visual pathways (there are several). So we're always scanning visually, always updating our view with new info.
Well, OP never specified, so neither did I, but I did suspect that they were talking about per second. To say we absorb gigabits of information over 90 years is a different story, and one which I assumed OP wasn't talking about.
Nothing on the size of ABSL's vocabulary. Is there a linguistics notion of turing completeness? Ie. How do I know if a language is capable of expressing any possible concept ("program")? Is this even a meaningful question?
You can't because we don't know any possible concept, nor do we know if we can even conceive them, so we can't prove a language can express them.
Plus we already have concepts we can't express. We can't share what looks like the resulting feeling in your mind of something the other never encounter, like a color to a blind man or a sound to a deaf girl. There are many feeling and sensations that you encounter in your life you can't define.
We can't define what we don't know either. E.G: we can't define whatever happen after death because we don't know about it.
At last, some practices, like meditation, make you (quite surprisely) discovers some very simple things, yet it seems quite impossible to share it using words with another person. Here knowledge and complexity are not the issue. It's just that labelling is not a silver bullet for communication.
Under the assumption that the whole "you need to be exposed to something X times before you really remember it" thing is true, I love when I get reintroduced to stuff like this that I've completely forgotten about since it was last mentioned.
A Turing machine is a very simple device. Evaluating if something is "Turing complete" is simply determining if it can function like a Turing machine.
Compare this with trying to define the "set of all possible concepts". What does that even look like? What is a concept? How can it be expressed? Concepts are a lot more nuanced than bare 1s and 0s because there's subjective meaning applied to them.
The complexity here is that concepts are generally recursive, they have intricate links to other concepts, which in turn link to others.
In the same way, "Turing complete" might not mean "capable of doing any possible computation". It could be that there are some computations that can be performed but not by a Turing machine. This is uncertain enough that we call it the Church-Turing Thesis. But it's hard to describe the notion of "every possible computation", so instead we compare things relative to Turning machines.
We could do the same thing with languages. Say a language is "English complete" if it can express at least everything expressible in English. I suspect all commonly used modern languages are English complete. The most obvious way I can see that a language would fail to be complete would be if it lacked nonce words or any equivalent device.
> I suspect all commonly used modern languages are English complete...
English is just one language, and while perhaps unusually expressive because of the amount of abuse it's received over the centuries, plus its ability to wholesale absorb new words on a whim, it is incapable of expressing some things with the same degree of succinctness and precision other languages have.
There are words in other languages that take a paragraph to explain in English because of all the cultural context they contain, or because they refer to things that most people who speak English have never seen, nor felt a need to describe with that degree of precision.
The problem with using English as a basis is it's very hard to explain in English what the language is incapable of expressing. This is like a fish trying to explain what they can do on land when the very concept of land is hazy at best.
This approach is about investigating what you would need in order to start defining other vocabulary that would let you eventually define all of the vocabulary that any language uses. However, this could be seen as an artificial process that doesn't imply that the NSM itself can be used directly for complex communication (since perhaps the definitions for many common everyday concepts might end up running to book-length). I'm also not sure how they've thought about ostensive definitions, to help distinguish things like particular colors or particular animal and plant species.
I've used toki pona, which is a deliberately small constructed language that aims to make it inconvenient to say certain kinds of things. And indeed, it is very hard to unambiguously refer to particular species, other than by ostension and subsequent convention in a particular speaker pair or community.
It's very tempting to think that ad-hoc linguistic experiments like this have appeared and disappeared all the time through history, with only the "fittest" surviving today.
There is no such thing as a kind of turing completeness in linguistics. natural languages are not even finite and are rarely used to express "computations."
Infinite tape is a requirement for Turing completeness. A whole lot of language is used for making sets of things, and inclusion or exclusion from those sets. Take, for example, your comment. And mine.
Maybe set definitions and memberships aren't computation though. I think they are, but I'm a fuzzy thinker.
The core claim of the title, that somehow non-decomposable sign language goes counter to something doesn't match anything I've ever heard in linguistics. Many signs in ASL are fairly representational as well.
While not a natural language, aUI [1] is a conlang (constructed language) that sort of tries to emulate the idea described in the article. Each sound represents a very basic concept (they map to semantic primes [2]), and they're combined to form more complex ideas.
Unfortunately, this leads to nonsensical literal translations like "ONE-QUANTITY-ABOVE-LIFE-TOWARDS" for "banana," but it's still a neat/interesting concept.
They are saying each word in ABSL is a unique sign without being composed from other sign elements.
In BSL, for example, [as I learnt it some time ago, at a basic level] the signs for class and group share the hand movement but use respectively the hand-shapes for C and G; they're compound gestures. Or any easier one, vehicles like car, bus, etc. are signed by miming use of a steering wheel and saying/mouthing the word (eg http://www.signbsl.com/sign/truck - see the "SignStation" version of this and http://www.signbsl.com/sign/car); again compound.
ABSL presumably has words for car and truck but they don't share visual elements?
To attempt to answer your question I think the idea is that it's surprising that a language is not compounded from smaller elements, like verbs using "ing" to denote action? Or like how Greek prefixes/suffixes are used in English, xeno- say for foreign (eg in xenophobic: fear/hate for foreigners; and xenolith: a rock type foreign to the surrounding rock) or -lith for rocks (as before and megalith: a big rock). One expects to quickly run out of new sounds (or signs) if you don't compound/combine them atomically.
> In BSL, for example, [as I learnt it some time ago, at a basic level] the signs for class and group share the hand movement but use respectively the hand-shapes for C and G; they're compound gestures.
Same in ASL. But I wouldn't call it a compound sign, since it's not combining the COLLECTIVE sign with TOPIC CLASSIFIER:INITIAL C/G IN ENGLISH sign.
Would you consider ASL HURRY as a compound sign, since it uses the H handshape as a mnemonic in it? How about ELEVATOR and its C?
I would consider compound signs to be things like THROAT DOCTOR, PET STORE, SKIRT LONG, where we use two signs with distinct meanings which are easily comprehended in isolation, but when used together communicate a different concept.
> One expects to quickly run out of new sounds (or signs) if you don't compound/combine them atomically.
I can see this with the limited options available in spoken languages: there only a few hundred ways to combining vowel-consonant pairs, but a few thousand ways to combine fingers+limb+facial expression+short movement (although the challenge becomes how to reduce interpretation errors i.e. was that index finger flicking down 3 times or thumb flicking up twice?).
As a linguist, I have to refute some of the claims made in this article. The understanding for the past few decades is more in line with ABSL than represented. The only hold-outs are hard and fast supporters of Universal Grammar in the face of contradictory evidence. The only genetic factor commonly recognized as important to language development is merely the ability to learn language. Even recursion within language is in doubt due to the language Pirahã. For the most part, UG is held up by die-hard defenders of Chomsky, or is given as introductory education in Linguistics despite not being the widely held view.
This seems like a dubious claim to me. If you want other examples of languages where words aren't built from meaningless atoms, look at early writing - the word for 'tree' is a crummy picture of a tree, etc.
Later on, writing developed into either using meaningless signs to represent sounds, and stringing those together (as in the case of what you're reading now), or using signs to represent fragments of meaning which are combined to form complete concepts (at least, that's how i understand Han ideograms). Rather like the business with the slide whistle.
I was just thinking that "...a new sign language emerging..." would more likely consist of whole concepts embodied in independent expressions, which later get broken down into some parts that are more easily recombined into new words/concepts. Thanks for casting sharing this perspective!
Note that even genome hes this hierarchical structure. It's composed of genes, which are composed of codons, which are composed of bases. It may be that the property is not just a way human brain works. It may be that it just happens to be a good technical solution of transferring information with high fidelity.
The problem with the human body is that is extremely hard to share information. We can absorb gigabytes of information through our eyes and other sensory organs, but we can only emit a few bits of information. When we communicate, we are essentially trying to share our thoughts through a very narrow straw that limits the flow of information. The result of this bottleneck is language. Language requires lots of context, common understanding, and being able to view the world from the other person's perspective to make sense of the tiny amount of information that one person is sharing with another