5 ms·
"Furthermore, much of what we consider intelligence is inherently dialogic, hence social; it requires a theory of mind." That's an assumption. The contrary vie
by Animats 5y ago
"Furthermore, much of what we consider intelligence is inherently dialogic, hence social; it requires a theory of mind."
That's an assumption. The contrary view is that language is just part of the I/O system, not where the hard work gets done.
Animals with very limited language capabilities have the ability to run their own lives, and to some extent manipulate their environment. Systems built from large language databases don't have that.
After 50 years, robotic manipulation in unstructured situations is still terrible. Rod Brooks tried to crack that and his startup, Rethink Robotics, failed. DARPA has had decades of manipulation projects. Fifty years ago, they could put a bolt in a hole. Five years ago, they got a robot hand to put a key in a lock. Amazon is still trying to get robots to pick things from bins fast and reliably.
We lack some fundamental understanding about how to interact with the real world. The problem is not on the language side.
- fake-name 5y ago> That's an assumption. The contrary view is that language is just part of the I/O system, not where the hard work gets done. FYI, this is the [Sapir-Whorf hypothesis](1) > We lack some fundamental understanding about how to interact with the real world. The problem is not on the language side. This is to a significant extent a sensing issue, not a understanding issue. 1: https://en.wikipedia.org/wiki/Linguistic_relativity https://en.wikipedia.org/wiki/Linguistic_relativity
- tsimionescu 5y agoThe Sappir-Whorf hypothesis refers to specific languages, and is all but proven wrong in its strong form (a person speaking English and a person speaking Japanese seem to perform exactly the same at all cognitive tasks). But there is a different theory, due to Noam Chomsky I believe, that language in the general sense is in fact our mechanism for higher thought, with communication being just a secondary function[0]. I believe that this is the theory that the GP was combatting with the animal example. [0] https://www.youtube.com/watch?v=KEmpRtj34xg https://www.youtube.com/watch?v=KEmpRtj34xg
- thaumasiotes 5y ago> The Sapir-Whorf hypothesis refers to specific languages, and is all but proven wrong in its strong form (a person speaking English and a person speaking Japanese seem to perform exactly the same at all cognitive tasks). Japanese people show extremely high performance at "spatial awareness" tasks compared to whites.[1] So on average this isn't true. But it's true that there does not appear to be any benefit from either language -- Japanese-speaking Japanese show no advantage over English-speaking Japanese. [1] On white-normed IQ tests, East Asians generally show a bit of underperformance (relative to European whites) on verbal tasks and gigantic overperformance on spatial tasks, which is traditionally averaged out to an average IQ score moderately above that of whites. Interestingly, Native Americans show the same slant -- much much stronger spatial than verbal performance -- but at a lower base level.
- xwolfi 5y agoAnd that is due to the language and not, say, the writing tradition ? If we kept the same languages and made japanese people write it massively in 26 half sound symbols and made all of what you call a white european population use 3k+ ideogram drawings for word starting at 3 years old, do you think we d see exactly the same performance ? I bet you the writing tradition is very influencial when it varies so much and has no relationship with language, especially in Japanese where they frankensteined Chinese symbols over centuries which must require an enormous mental gymnastics and redirect large processing power vs someone who can read and write everything he hears relatively simply?
- thaumasiotes 5y ago> And that is due to the language and not, say, the writing tradition ? No, neither. > If we kept the same languages and made japanese people write it massively in 26 half sound symbols and made all of what you call a white european population use 3k+ ideogram drawings for word starting at 3 years old, do you think we d see exactly the same performance ? Yes, the influence of spoken and written language appears to be zero. I mentioned this.
- 5y ago
- slibhb 5y agoI interpret what you quoted as the opposite of Sapir-Whorf. Could you explain why they're the same?
- Mezzie 5y agoI am also quite confused. My understanding (it's been a while since I studied Linguistics) is that strong Sapir-Whorf refers to the argument that what actions your brain can take are constrained by your native language(s). So by strong Sapir-Whorf, the distinction between the French savoir and connaitre should make zero sense to me since that distinction isn't found in English. Weak Sapir-Whorf says the distinction wouldn't make sense until I learned words for it, but that the brain can learn new concepts if it acquires language for them. So SW argues that language is inherently intertwined with human cognition, rather like how the parts of the brain are all interconnected and working together as we think. Strong SWism is pretty thoroughly debunked, weak is tentatively accepted on a spectrum. I think that they might be trying to refer to generative theory versus cognitive theory. Again, it's been a while, so I might be wrong, but: - Generative theory is Chomsky et al. who believe that there is a single inherent underlying pattern to language that restricts it, and that one could, in theory, figure out all possible syntaxes, for example. Similar to how phonology and phonetics are obviously constrained by our biological makeup (we can only make certain sounds), the argument is that other structures are as well. Basically that although they all seem different to us, languages are more alike than they are different because they come from a single source. Also Chomsky and co. consider this approach cognitive; the linguists are just infighting. - Cognitive Linguists believe that language, and the systems that produce it, are under evolutionary pressure and that they change and shift in response to the environment. Basically, if you took 10k people and dumped 5k of them on one planet and 5k on a planet that was the physical opposite, forbid them to contact anyone, and visited 2000 years later, Generativists say you could find a system in common between the two and cognitive linguists say you probably couldn't, because the systems themselves would have evolved to have different starting points. Chomsky + GG: Person is born, person hears speech, person reaches for inherent language patterns in brain and uses them to understand/learn. Cognitive: Person is born, person hears speech, person picks up on the system from the speech and is encouraged to do so/pattern match by social reinforcement. No linguist that I'm aware of would claim that language is just an I/O system.
- crooked-v 5y ago> Animals with very limited language capabilities have the ability to run their own lives, and to some extent manipulate their environment. This makes me think of the ongoing effort by some people to train their pets to use talking buttons to communicate. These include the animals occasionally trying to convey relatively complicated ideas with simple sets of words, like "play help" ("I'm bored, come entertain me") [1] and "dog settle" ("make that dog outside stop being annoying") [2]. [1]: https://www.youtube.com/watch?v=zAQKA9WFlq4 https://www.youtube.com/watch?v=zAQKA9WFlq4 [2]: https://www.youtube.com/watch?v=5sY3K83Zao8 https://www.youtube.com/watch?v=5sY3K83Zao8
- thaumasiotes 5y agoManipulation involves a few things: 1. Recognize objects in the world, like a keyhole or whatever it is you want to take out of a bin. 2. Recognize your own position -- how close is your hand to the keyhole? 3. Adjust your position until it matches your goal. My impression of things is that most of our difficulty comes from task 3. It's not so much that we have trouble recognizing where we are, where we want to be, or how to move from where we are to where we want to be. But we have difficulty defining movements that don't cause damage to nearby (or target) objects. Your human hand has many features that allow it to grip an object securely without damaging the object. And you also have a decent sense for how to move your hand to an object without damaging the object -- or your hand -- though this problem is far from being solved in humans too. Ever stub your toe?
- jacquesm 5y ago> Animals with very limited language capabilities have the ability to run their own lives, and to some extent manipulate their environment. Systems built from large language databases don't have that. This may well point to the possibility that animals have a theory of mind.
- retrac 5y agoI don't know if they have a theory of mind, but some animals absolutely have very complex behavioural prediction models for humans and other animals. I swear my last dog knew whether I intended to go outside based solely on how I stood up from the couch.
- Vetch 5y agoThis can be explained as associative learning, without needing to reach for ToM, which is still no mean feat.
- Vetch 5y agoI believe it's been shown as a general capability in [corvids](https://www.semanticscholar.org/paper/Ravens-attribute-visual-access-to-unseen-Bugnyar-Reber/228a4cd7ac54275cbd3f4fa1db73f85d742462e7 https://www.semanticscholar.org/paper/Ravens-attribute-visua...), great apes and maybe possibly in pigs (and goats). It's sort of game theoretic, as in "I know that you know that I stash food here so I'm going to pretend to hide the food here". I've never read anything that goes beyond two levels of "I knows" but it can be sophisticated, such as in the linked paper on Ravens where there is line of sight through a peephole that can be open or closed. Sophisticated ToM is likely rare in the animal kingdom, not usually going beyond programmed responses to being observed and possibly empathizing with conspecific suffering. In humans, it's a much more sophisticated process of inferring motivations, anticipating moves and a number of similar inferential capabilities plus abductive reasoning.
- throwaway29103 5y agoWhen talking about language (I see in the comments below), I think it is useful to differentiate between the potential capabilities of human languages (i.e., all baby humans seem to be able to learn to adult native level of proficiency any human language, regardless of were where their parents from) and the possible effect that using a particular language might have in the speaker or in the people receiving speech in that particular language, i.e., the expressive capabilities of that individual in that language or of the vocabulary of that language specifically (e.g., crafting a joke: how do you make an egg laugh? Tell him a yoke). What I mean is that language ability of humans and language ability of animals are not comparable; they are so different that in fact, language ability (regardless of the externalization of that language, by speech, text, signs, touch), looks very much like intelligence (meaning the intrinsic ability for language production and not the specific proficiency at some sort of externalization). If you happen to spend enough time around animals and child humans you will quickly see that after a couple of years, kids can understand and use recursive / referential structures in a way that pets cannot. Pets seem to understand individual words, tones, and maybe "not". Think of "bring the comfy ones that your grandparents bought you to keep your feet warm" - a sentence a 3 year old can perfectly understand. No matter how much you spend teaching/training a dog that has lived in the same environment since its birth, the dog will not react to it bringing you the yellow slippers (unless you specifically trained the dog to associate the word "comfy" or "feet" to some specific slippers). The current state of parametric language models resemble much more that of a very memorious dog than that of a human.
- KronisLV 5y ago> That's an assumption. The contrary view is that language is just part of the I/O system, not where the hard work gets done. > Animals with very limited language capabilities have the ability to run their own lives, and to some extent manipulate their environment. Systems built from large language databases don't have that. I'm tempted to agree with this in some capacity, language is probably just one expression of thought, much like how some people have an internal narration going on during their lives, while others think in abstract concepts that aren't necessarily expressed in language. Here's the first article DuckDuckGo turned up on the subject, i bet someone can share a better one: https://www.iflscience.com/brain/people-with-no-internal-monologue-explain-what-its-like-in-their-head/ https://www.iflscience.com/brain/people-with-no-internal-mon... Furthermore, there are people who don't have the capacity to visualize things in their "mind's eye" either (aphantasia), which is another curious thing: https://www.sciencefocus.com/the-human-body/aphantasia-life-with-no-minds-eye/ https://www.sciencefocus.com/the-human-body/aphantasia-life-... Of course, when you get into it, there are also some niche theories about how language could not only help us express ourselves, but shape our thought (not that dissimilar from the "newspeak" concept in Orwell's 1984): https://en.wikipedia.org/wiki/Linguistic_relativity https://en.wikipedia.org/wiki/Linguistic_relativity
- celrod 5y agoI've had aphantasia all my life, although my dreams are still visual, as though they're real. The article claims it is more common among academics and computer scientists, so perhaps it's fairly common here. Perhaps it's helpful for dealing with abstraction, as abstractions don't always map well to visual representations.
- joe_the_user 5y agoThe contrary view is that language is just part of the I/O system, not where the hard work gets done. The meaning of "where" in your claim is very slippery. AR Luria in The Working Brain points out that damage to virtually every part of the brain can result in language dysfunction. To say that a whole lot of things happen apparently without language doesn't demonstrate that language is an extra. The brain generally isn't very modular and my alternative take would view language as accompaniment that sometimes is "just words" and sometimes touches very specific details (edit: plus allow recursive access these other brain-functionalities, as others point out). We can see how GPT-3's output is a stream of plausible things punctuated by some statement that shows "it doesn't anything" (IE, gets the implicit context wildly wrong).
- foobiekr 5y agoIf you've ever spoken to someone with Alzheimer's, or certain kinds of very severe cognitive disorders, it becomes clear that language fluency tells you little to nothing about effective cognition. There are people who can't add 1+1 who can babble about random stuff convincingly, for a while, anyway, unless you're probing it.