8b. Blondin Massé et al (2012) Symbol Grounding and the Origin of Language: From Show to Tell
Blondin-Massé, Alexandre; Harnad, Stevan; Picard, Olivier; and St-Louis, Bernard (2013) Symbol Grounding and the Origin of Language: From Show to Tell. In, Lefebvre, Claire; Cohen, Henri; and Comrie, Bernard (eds.) New Perspectives on the Origins of Language. Benjamin
Arbib, M. A. (2018). In support of the role of pantomime in language evolution. Journal of Language Evolution, 3(1), 41-44.
Vincent-Lamarre, Philippe., Blondin Massé, Alexandre, Lopes, Marcus, Lord, Mèlanie, Marcotte, Odile, & Harnad, Stevan (2016). The Latent Structure of Dictionaries. TopiCS in Cognitive Science 8(3) 625–659
Organisms’ adaptive success depends on being able to do the right thing with the right kind of thing. This is categorization. Most species can learn categories by direct experience (induction). Only human beings can acquire categories by word of mouth (instruction). Artificial-life simulations show the evolutionary advantage of instruction over induction, human electrophysiology experiments show that the two ways of acquiring categories still share some common features, and graph-theoretic analyses show that dictionaries consist of a core of more concrete words that are learned earlier, from direct experience, and the meanings of the rest of the dictionary can be learned from definition alone, by combining the core words into subject/predicate propositions with truth values. Language began when purposive miming became conventionalized into arbitrary sequences of shared category names describing and defining new categories via propositions.
Organisms’ adaptive success depends on being able to do the right thing with the right kind of thing. This is categorization. Most species can learn categories by direct experience (induction). Only human beings can acquire categories by word of mouth (instruction). Artificial-life simulations show the evolutionary advantage of instruction over induction, human electrophysiology experiments show that the two ways of acquiring categories still share some common features, and graph-theoretic analyses show that dictionaries consist of a core of more concrete words that are learned earlier, from direct experience, and the meanings of the rest of the dictionary can be learned from definition alone, by combining the core words into subject/predicate propositions with truth values. Language began when purposive miming became conventionalized into arbitrary sequences of shared category names describing and defining new categories via propositions.
Chimps, just like humans, can learn categories through direct sensorimotor induction. However, when defining new categories by combining symbols, they are not able to construct propositions. Creating propositions to learn new categories (learning by instruction) is unique to the human species. In this paper, it is suggested that this skill might require a cognitive capacity that chimps do not have either because Baldwinian evolution did not enhanced it or because they do not have a genetic mutation. Humans started learning by instruction passively, but individuals who were the most motivated to learn this way were advantaged. It was less time consuming and less risky. Being able to learn by instruction was favored by Baldwinian evolution because of its adaptive advantages and was even more so when vocalization took over gestures. Eventually, offsprings’ brains became wired for the propositional power of language.
ReplyDeleteThis is definitely an interesting point, but I think that it remains very relevant to highlight the suggestion made by the authors that this lack of propositional constructions in chimp communication could also be related to a lack of motivation. It does not seem impossible that chimps create more complex categories through symbol combination, nor that they start teaching their young categories rather than letting them learn them through induction.
DeleteI think that where your point about Baldwinian evolution comes in is that it could be that, as humans are extremely kin-dependent (the human offspring being particularly helpless), a group of humans discovered that categories could be learned by instruction. This providing an evolutionary advantage, it could have motivated the modificaiton of cognitive modules or the expression of genes favouring the learning of these methods (appartition of a disposition toward learning these methods).
***EVERYBODY PLEASE NOTE: I REDUCED THE MINIMUM NUMBER OF SKYWRITINGS. BUT THE READINGS ARE **ALL** RELEVANT TO AN OVERALL UNDERSTANDING OF THE COURSE. SO, EVEN IF YOU DO NOT DO A SKYWRITING ON ALL OF THEM, AT LEAST FEED EACH READING YOU DO NOT READ TO CHATGPT AND ASK IT FOR A SUMMARY, SO YOU KNOW WHAT THE READING SAID — OTHERWISE YOU WILL NOT HAVE A COMPLETE GRASP OF THE COURSE TO INTEGRATE AND INTERCONNECT FOR THE FINAL EXAM.***
DeleteSophie "learning by instruction passively"? Isn't language much more active and interactive than that? And how could it have started that way?
Sofia, Chimps are every bit as kin-dependent as humans! (But motivation rather than capacity might still have been what theaccelerated language learning once it had started: What do you think started it?)
Sorry, I meant that when mother would push the bad mushrooms away as a mean to say “don’t eat, it’s bad” and the child would learn the category by observing the behaviour multiple times (passive by induction). When sharing new (composite) categories to help kin, miming new categories to learn had the purpose of information transmission. It is very interactive and active as it helps to reduce uncertainty by informing others so they can do the right thing with the right kinda thing. The first instance of using propositions might have been accidental (”passive”) without even understanding the power it held, but individuals that were motivated to use propositions benefited from survival advantages. Learning by instructions/propositions (indirect grounding) instead of sensorimotor induction is faster and allows for the recombination of categories (composite) without the risks involved in induction. It became the main way to share categories because motivated individuals who used instructions had survival advantages and their offsprings would inherited of this motivation and predisposition to learn that way.
DeleteThis paper is about how language evolved from showing to telling. Before language, people communicated via pantomime, which are iconic non-arbitrary gestures that represented real world things directly. This is the stage of showing. Over time, pantomime turned into propositional communication, which are statements that have truth values. This is the birth of language. Evolution then favored biological structures that supports symbolic communication, such as specialized brain regions, vocalization mechanisms etc.
ReplyDeleteThe origin of language is illustrated through artificial life simulations by Cangelosi and Harnad (2001; Cangelosi et al. 2002). Creatures learned to categorize mushroom type A and B directly through sensorimotor experience (trial and error). For mushroom type C, some creatures also learned by induction, while others learned by instruction—observing that mushroom type C are ones that are both type A and B. Over generations, instruction learners out survived induction learners, showing adaptive advantage of symbolic communication.
I found it interesting that chimps can label categories but does not seem motivated to roll with it. They seem to understand them as associations, but it’s unclear whether they grasp them as propositions. Since language is evolutionarily advantageous, why are chimps not motivated to utilize the power? Are they missing some subtle cognitive capacity or motivational bias that prevents the development of language?
This paper offered an interesting dive into the origin of language and how primitive “symbolic” gestures eventually led to the emergence of verbal propositional communication. What particularly interested me was the concept of “grounding kernels”. As mentioned many times previously, for a symbolic system to have meaning, it requires being grounded in sensorimotor experience.
ReplyDeleteEssentially, using graph-theory, it was discovered that every word can be defined by other words and there are a core subset of terms called “kernel words”. These are the foundation from which all other word meanings can be reached through definition alone. However, this isn’t done hierarchically where kernel words simply branch off extending to other words but rather, are cyclical and interconnected. Kernel words are used to define one another too.
Emily, it indeed seems like no kernel acts as a true foundation, but rather that networks of kernel words constitute the basis for those propositions that cannot be grounded directly. We talked, a few weeks back, about grounding levels, and while reading this article, it became somewhat clearer how those symbols, which cannot be grounded directly through sensorimotor interactions with referents, can be indirectly grounded by way of linking them to those symbols that are directly grounded. This could have then served as a way for the emergence of language, as the first, grounded, symbols started to be combined, whereby language then would have involved 2 complementary (not successive nor merely distinct) processes: induction (the sensorimotor grounding), and instruction (combinations of the prior), the latter allowing for faster and more efficient language learning (indeed, no need for all induction processes to happen to expand one's language repertoire anymore, allowing one to talk about unexperienced events). I wonder if there's a reason for the emergence of communication about experiences not lived: if language evolved out of necessity, could there have been some necessity in communicating unexperienced events?
DeleteThis paper explains how humans moved from “show”—gesturing—to “tell”—communicating using propositions. Baldwinian evolution helped drive this shift: being able to learn through instruction gave major adaptive advantages. As vocalization took over from gesture, these benefits multiplied, and over generations, human brains became increasingly wired for the propositional power of language.
ReplyDeleteHow Natural Language Developed
1. Language (non-vocal): Language is a symbol system in which you can “say anything that can be said in any natural language.” Early humans used gestures to represent meaning before true language emerged.
2. Vocal Modality: Before humans had vocal language, they likely used non-linguistic sounds to communicate. “Once it had made its power and adaptive benefits felt, however, language simply migrated (behaviorally, neurally, and genetically) to the vocal modality, with all of its obvious advantages” (Steklis & Harnad, 1976). Speaking replaced showing — freeing the hands and expanding how, when, and where humans could communicate.
3. Vocal Language: Humans likely already had language by the time tool and weapon making began to multiply rapidly. This shift might have been sparked by a mutation, or perhaps the capacity was already there, needing only greater motivation: “The capacity or disposition to construe an A is a B as a proposition… may either have been a mutation, or primates may already have had the capacity and only needed to have it amplified motivationally and behaviorally by Baldwinian evolution.”
4. Evolutionary Hard-Coding: Over generations, the capacity for language became hard-coded into our genes and brains. It’s unclear when this genetic evolution slowed and soft-coding through learning took over.
5. Historical Soft-Coding: Today, language continues to evolve culturally rather than biologically — passed down not through genes, but through shared experience and communication.
Hey Lucy! I like how you described the shift from show to tell. It’s fascinating for me to think that once language moved from gestures to speech, it didn’t just make communication easier for us but also probably changed the way we thought too. Talking instead of showing might’ve pushed us to form clearer ideas, since words have to represent things not actions. It also makes me wonder how that shift changed social life. Once people could describe things that weren’t right in front of them (imagining stuff and sharing stories), maybe that’s when human culture really began to take off.
Delete
DeleteLucy very good synthesis! A few important details:
(1) It's true that gesturing is much better than vocalizing to show than to tell (and vice versa). But both of these production media can do some showing (vocalizing can show sounds, gesturing can't), but gesturing can do a lot more showing (why?).
(2) But it's also true that once you have evolved language, it can be in either medium -- and for non-hearing people it is entirely gestural. Gestural language is a full-blown symbolic language, and it can express any proposition (what does that mean)?
(3) But that means it is much more likely that language first began in the gestural medium. (Why?)
(4) We don't yet have an explanation of how propositionality (what is that?) evolved. But once it did, iconicity was irrelevant -- and even a handicap (why?). So then it was easy to migrate to the vocal medium, and its many other advantages. (What are they)?
(5) But pre-propositional gestural communication is still just showing, not telling (why?). So there isn't really a "primitive" language. If it is just iconic (i.e., the gesture just resembles what it is imitating or pointing to) instead of arbitrary, then it isn't language at all.
(6) Nor is it just arbitrariness that makes language language: it's propositionality. And with propositionality you can say any proposition. What would be a language in which you could just say some propositions and not others?
(7) Propositions (remind yourself what that is) do mean something. What does "the cat is on the mat" mean? The content-words in that proposition (which ones are they?) don't really "mean" anything: they refer to something (what?); if they don't, then the proposition doesn't mean anything.
(8) But once you have propositionaility, there is no reason you cannot express any proposition at all. Content-words refer mostly to categories (kinds of things) rather than to individuals (with proper names). If you learn the category, and the content-word which is used in the language to refer to it, you can say any proposition, in any language.
(9) Yes, the capacity to learn language is coded in our genes, but the language itself has to be learned. The question of how propositionality evolved is, I think, not yet an answered question.
(10) The question of when language took over from evolution is not a very clear one. Evolution never stopped. What took over was learning new categories from others, rather than directly: those who had already learned the new category's distinguishing features (like the mycology professor arriving to the mushroom island), could tell them to the castaway who was trying to learn them the hard, long, direct way (as long as the professor and the castaway spoke the same language:
(11) Indirect verbal grounding took over from direct sensorimotor grounding. Trial+error sensorimotor learning has its own evolutionary, genetic basis, but, apart from the fact that many complex categories were only made possible to learn through language: It would be hard to learn the category "catharsis" directly, but easy to learn it from a verbal definition or description,
(12) Sample a dictionary to estimate what percentage of our categories (the referents of our content words) could have been learned directly, like "cat" or "carry". Truly "hard-coded" categories, like colors, are not really learned at all; only their names are. Their feature-detectors are in our genes.
Jad, see the clarification about show/tell gesture/speech above. The major change was propositionality, though orality speeded things up too. Anything can be described and discussed in gestural language. It was language itself that accelerated and extended culture. Story-telling is an important cultural development, but I doubt it was a primary driver for evolution: Propositionality would have to have started first. The increasing arbitrariness of gestures, however, could have started much earlier.
Instructor , about your comment, here are some thoughts :
Delete(1) gesturing can do a lot more than showing by increasing the amount of things you can communicate about, i.e. whatever you want to communicate doesn’t have to be in close proximity, as long as your “audience” is able to understand what you mean by gesturing. An example that comes to mind is rocking a baby from side to side for the baby to stop crying. Gesturing allows to “mime” the event, while showing would intel pointing to a mom rocking her baby? However, I believe the gesturing is said to be a form of showing : they both can’t convey about non-existant categories.
(2) “Gestural language can express any proposition” means that it meets the criterion for being a natural language (Katz’s "glossability" thesis: A natural language is a system where any proposition can be expressed).
(6) A language in which you could just say some propositions and not others would be categroized, based on the paper, as a protolanguage. Relating to (2), such language would not be a natural language as it doesn’t pass Katz’s "glossability" thesis.
The question of when did the evolutionary hard-coding of language into our genes and brains end and the historical soft-coding through learning and experience take over really clarifies what we were talking about in class on Tuesday. I, for one, was struggling to find the words to elaborate on the line between evolution and cognition but I think the notion of hard coding (sex and spiders) and soft coding (learning) sums it up well. However, defining the border between them remains a bit fuzzy to me.
ReplyDeleteI found the article The Latent Structure of Dictionaries (Vincent-Lamarre et al.) striking because it connects the structure of dictionaries to the symbol grounding problem. What I find fascinating is how the authors use graph theory to uncover a hidden hierarchy in dictionaries, showing that only about 1% of words (the “Minimal Grounding Set”) could, in theory, define all others. This makes me think about how our mental lexicon might work in a similar way: a small core of grounded, concrete words we learn early in life forms the foundation for understanding everything else through language. It also highlights a deep connection between learning, cognition, and evolution, similar to the Baldwinian view that our capacity to learn and build symbolic systems like language could itself have evolved because it was so adaptive.
ReplyDeleteThe discussion of Universal Grammar (UG), reminded me of the core-knowledge theories of child development. The authors define UG as a “complex set of rules that cannot be acquired explicitly from the child’s experience because the child does not produce or receive […] enough […] corrective feedback to learn [its] rules by trial-and-error”. This argument, referred to as Chomsky’s poverty of the stimulus (POS), implies that there is an innate brain mechanism that has evolved specifically for language acquisition. The authors further hypothesize that this adaptation was selected for (through Baldwinian evolution) because language proved to be an incredibly powerful tool for indirect category learning in our species. This ties back to core-knowledge theories, which propose that some kinds of knowledge (ex. language, basic physics) are so important for survival that they are the first to be acquired in children, with the help of innate brain “modules” that predispose those kinds of learning.
ReplyDelete
DeleteCendrine, I’m interested to know more about core-knowledge theories (one of my favourite courses at McGill was language acquisition), but I’ve never heard of them. A notable observation in studying children acquiring language is that when they make a grammatical mistake, they are often resistant to adults or others’ corrections. For example:
Child knocks over a toy: “He falled!”
Parent: “You mean he fell.”
Child: “Yes, he falled.”
Presumably the child has never heard the word ‘falled’, but they are applying grammar rules that they have made up themselves (add ‘-ed’ to the end of a verb for past tense) in an incorrect context. I wonder if there are parallels to this in core-knowledge theories or other similar areas of development, where it’s important to acquire but the plasticity of acquisition sometimes sends children down the wrong path.
This paper looks at how language might have started with people going from “showing” things to each other to “telling” each other about them, which really changed the way we communicate.
ReplyDeleteWhat stood out to me is how language isn’t just about naming stuff. It actually gave humans the ability to describe new ideas and work together better. If I think about it, this ability probably helped us share knowledge and build on each other’s ideas way faster than just learning by ourselves. The paper made me realize that the origin of language could be more linked to the way people interact and support each other in communities, not just about talking. Maybe the real power of language is in how it lets people create and share meaning as a group and that seems like a big reason why culture keeps evolving.
I really like how you linked the shift from “showing” to “telling” with the idea of shared meaning in communities that’s what stood out to me too. The paper says that once humans started using symbols instead of pantomime, they weren’t just naming things; they were creating a way to teach and learn categories from each other. That ability to tell rather than show let knowledge spread without everyone needing the same direct experience. It connects well with Harnad’s later work (the other reading) on how meaning depends on shared perceptual grounding (we all understand words because we live in the same world and perceive it similarly). So, as you said, language’s true power isn’t just talking, I think it is also how it lets groups build and evolve culture together.
DeleteI love how both of you emphasized this shift from showing to telling. That’s exactly what Blondin-Massé et al. (2012) argues: early humans didn’t suddenly just begin speaking; they likely began by pantomiming actions in order to teach others what to do. Then, over time, those gestures got replaced by arbitrary symbols (words) and that’s when language really took off.
DeleteWhat struck me most in that paper was that this wasn't a matter of convenience, but rather how learning worked. Animals and early humans could learn through induction, trial and error with the world. But once we developed language, we would be able to learn through instruction in that we ourselves didn't have to personally experience it first; instead, someone can tell us what a dangerous mushroom is rather than finding out the hard way.
I also really liked the dictionary idea in the reading on how there’s this concrete “core” of grounded words, and then the rest of language builds out from that based on definition and proposition. So, I think I agree with both of you: language isn’t just naming things. It lets us share categories, pile knowledge together, and build culture way faster than any species that only learns by doing.
What I find fascinating about this paper is the discussion of why chimps have not reached a point where they can understand propositions. As other comments and paper have mentioned they lack the motivation that humans had to make that leap. In which the social aspect of hominins made the ability advantageous to possess. I think this necessity humans have of language is interesting. There are clearly much more social animals in the world that aren’t even close to developing the propositional based system that humans have like those that live in colonies (ants or bees) that not only communicate through different modalities but in a different way altogether. I just wonder, even with our social nature as a contributing factor how propositions and languages became what was chosen.
ReplyDeleteI really appreciated the blurb under Figure 1 in the paper. I thought it nicely captured the overall point, articulating well how categories A and B have to be learned through direct sensorimotor experience whereas C is able to be learned linguistically. It demonstrates how there is a fitness advantage for linguistic telling over showing. This said, part of the figure itself confused me. What exactly is 𝝰-45/360-.12? I’m assuming the box with the little dots represents the “learner” but I’m not very sure what the arrows represent nor the angle. Further, the paper writes that “only about 10% of the dictionary words need to be learned through direct experience, with the rest learnable though definitions.” But I’m curious how we are able to identify which words are in that “grounding kernel”. Does it vary from person to person? Is it universal?
ReplyDeleteElle, I am also confused about the formula 𝝰-45/360-.12 in Figure 1, but I will try to answer your questions about the grounding kernel. To identify the kernel words in a dictionary, we must eliminate all the words that are defined but that are not used to define other words. The word ‘banana’ can be defined as “a yellow fruit” but is not needed to define other words. We then eliminate all the words that can be defined using the other remaining words. The word ‘yellow’ can be defined using the words ‘colour’ and ‘light’, but ‘colour’ can be defined using ‘light’ and ‘dark’, so ‘yellow’ and ‘colour’ are eliminated. I think the grounding kernel varies not between people, but between languages because everyone speaking the same language use the same words for definitions, but not in a different language. Not only the symbols, but also the number of symbols may also differ. In the “translability” thesis, anything said in any language can also be said in any other language, but the number of words needed to translate the same thing in every language is not necessarily the same.
DeleteI would like to correct my answer after discussing about it in the class of Tuesday. I think that every dictionary has a different grounding kernel because the set of words defined in every dictionary is chosen by the authors and thus varies. Therefore, it is not universal and does not vary between people, but rather between dictionaries.
DeleteBlondin Massé et al. (2013) explain how humans went from showing to telling things through an evolutionary lens. Early people used miming, gestures that demonstrated what to do with what kinds of things, conveying categories. Over time, this gave way to speech using propositions —statements with truth values —which allowed people to define and transmit new categories efficiently at a distance and hands-free. This refined way of communicating had evolutionary advantages for humans, a particularly social and kin-dependent species, increasing their survival and ability to produce offspring across generations, which, in turn, favored genes and brains biased towards learning any language. That said, while the authors note that intelligent chimps seem uninterested in using language-like systems, I find it quite intriguing that, in principle, if the right ecological and motivational conditions were held for long enough, chimps could likely develop a similar linguistic system in the far future.
ReplyDeleteIn the reading “From Show to Tell”, the authors describe how humans moved from pantomime (communicating though gestures tied to direct experience), to symbolic language (where words stand for ideas and can express propositions). This transition marked a major evolutionary step, creatures who could learn through instruction (using shared symbols) gained an advantage over those who relied on only induction (learning through experience). The authors argue that while animals like chimps can learn categories and communicate through grunts or gestures, they lack the motivation to use symbols to express or teach new ideas. This suggests that what truly defines human language is not just the ability to use symbols, but the drive to communicate, share and explain to others.
ReplyDeleteThe “Latent Structure of Dictionaries” connects to this by showing what happens once language exists. It reveals that the entire structure of a dictionary depends on a small “core” of words. These core words are learned early, and come from direct experience, while more abstract words build on them through definitions. This reflects the same principle as the first paper, language grows from grounded experience, and meaning spreads through symbolic connections.
However, if human consciousness emerged through a bottom-up process, rising from perception and experience toward symbolic thought, could AI, evolve through the top-down route, starting with symbolic manipulation and connection, and eventually giving rise to a new form of perception, understanding, or awareness from within its own algorithms? Making it that instead of experience giving rise to symbols, symbols might give rise to experience?
One thing about the text that caught my eye was that it talks about chimpanzees and my friend in anthropology loves discussing this. Their motivation to “learn a language”, in the text, suggests that their lack of interest may be one of the main reasons behind their absence of complex language. However, this lack of motivation can be explained by social and biological factors unique to their species. For instance, chimpanzees have highly developed control over their gestures due to their prehensile hands and feet, which may lessen the need to invest effort in vocal communication. Most of their vocalizations serve to display strength or to issue alarm calls. Thus, sound-based communication fulfills very specific functions for them, which may contribute to their lack of interest in extending it to other contexts. According to the costly signaling theory, human language is a costly form of communication that provides long-term evolutionary advantages. In theory, chimpanzees could have also benefited from such a system, but their adaptive environment may not have exerted enough selective pressure to drive evolution in that direction.
ReplyDeleteAn interesting concept to this week's reading is the role motivation plays in the drive for language acquisition and how this relates to other primates. While other primates also have the capacity to connect symbols to experience and form categories, however they lack the social and communicative motivation to form a natural language. This might be a result of the innate cooperative tendencies of humans as we get our strength from numbers and working together.
ReplyDeleteThis comment has been removed by the author.
ReplyDelete8.b. Harnad argues that language evolved from showing to telling, from miming actions to communicating them symbolically. It’s an interesting explanation for how meaning became detached from direct perception, but the transition still feels underspecified. How did symbols first gain shared meaning without already having language? The paper’s graph-based analysis of word grounding is insightful, yet it seems more descriptive than causal. Still, Harnad raises an important question about how humans moved from sensorimotor grounding to abstract thought, which remains central to understanding language origins.
ReplyDeleteThis paper talks about how humans moved from showing with gestures to telling with words. At first, people used pantomime to show real things, but slowly it became more symbolic. Blondin-Massé et al. (2013) say this is how language started, when communication became propositional and not only visual. I think this means humans started to talk about things they imagine, not just what they see. The idea about grounding also made sense to me. Words we learn from real experience, like “rock” or “run,” help us learn harder ones like “freedom.” It shows how humans began to build new meaning just by speaking, not only by doing. I think this is why language became so important in human evolution.
ReplyDeleteHere, I found quite striking the idea that language was discovered through pantomiming and then later transmitted through Baldwinian evolution. In kidsib terms, human ancestors were first acting out whatever they wanted to communicate. However, you can say that the need for language came about because there is only so much you can represent through iconic miming. You need arbitrary representations to communicate as many categories as possible. I can mime an animal, but how do I mime "rock" or "sea" purely through iconicity? You can't. So, maybe someone associated an arbitrary sound like "rock" to an object, and then language was discovered from there. Eventually, this leads to people passing this on to their offspring. Then, you let Baldwinian evolution take place, and we end up with language acquisition capacities in humans. My main question that I failed to understand is why us and not other animals? Why did we develop language as we do? Other animals are able to categorize. They are able to communicate (like the japanese tit being able to distinctly communicate ground predator vs. sky predator). What is the missing key ingredient that humans have? The paper seems to suggest that it's lack of motivation, but I am not sure I follow why they lack this motivation.
ReplyDeleteJean-Rémy:
Delete1. Gestural Origins of natural language is just a hypothesis, a strong one, pragmatically, but like the other language-origins hypotheses, weak empirically.
2. First, there was learning capacity and nonverbal communication, as in most interactive species. Communication could be through visible movement, sounds, physical contact, olfactory signalling or even chemical signalling. This nonverbal communication could be voluntary or involuntary (and unconscious).
3. The visual/motor medium is the richest medium of voluntary, iconic nonverbal communication for many terrestrial species (including our own).
4. If you include pointing, imitation and copying, the visual medium is even richer (and it does include rocks and water that can be pointed to and mimed!).
5, But it does not include “republican” or “wastefulness” (and an infinity of other complex, composite categories that can only be expressed verbally).
6. No, language did not begin with “association” — a non-explanatory weasel-word.
7. And, according to the Gestural Origins hypothesis, language did not begin in the vocal/acoustic medium, which is far more impoverished than the gestural/visual medium. (Why? How?)
8. Your account missed all the points from category-learning capacity, to gestural communication, to arbitrariness, to propositionality, to migration to the vocal/acoustic medium: What are they?
9. No one has yet given a plausible explanation for how propositionality began, but that is a crucial question. (And neither the Japanese tit signalling nor the vervet monkey signalling is propositional.)
10. And, yes, it is not clear why language, with all its adaptive benefits, has not yet evolved in any species but our own. (And with the damage our species has done to the planet and its other species, it is not clear language can ever evolve in another species.)
Here is my attempt at fixing what I missed in my original comment now that I have the entirety of the course behind me while keeping a kidsib tone:
DeleteFirst, I had glossed over a crucial step that was necessary for language to even begin. You need to be able to form categories way before you get to language. When you are using language, you are using the right symbols and rules to express the right kind of things you want. Therefore, you need categorization capabilities first. Once you have acquired the capacity to categorize, you may need to share these categories with others in your group.
According to this theory, you share them by iconic gestures. Why? I believe gestures and pantomiming are more easily understood. We are equipped with mind-reading abilities and mirror capacities. Thus, it could be that combining these, they made an initial form of communication using these by showing what’s out there in the world, miming them, using our mind-reading abilities to deduce and infer what others may be trying to show you, and then maybe using our mirror capacities to use the same gestures in the future that you saw others do in order to communicate with them. However, this did not yet occur in the vocal medium. The reason for that is because sounds are arbitrary representations. This makes it much harder to share the categories among new people because there is no way for them to guess what you mean by “cat” if they have never heard English before. However, if you point at the cat or mime a cat, you have a greater shot at them understanding you.
Yet, the power of arbitrariness is exactly the reason why we had to eventually move to language. While iconic representation enables us to represent everything (or almost) that is out there in the world, you are unable to create or communicate things that aren’t out there. I can mime a dog, but how do I mime a corrupt politician (let’s pretend I couldn’t just point to Trump) or mime “common objects” (as pointed out above by Stevan)? Here is why we need the power of arbitrary symbols that can represent anything. When you read the categories mentioned previously, you now have something that comes to mind even though they do not actually look like what they represent.
But this is still not language. We are merely pairing categories to symbols. You need to be able to express things beyond me just saying cat and others knowing what a cat is. I need to express more complex thought. What about that cat? Where is that cat? What is it doing? I need to be able to say things that are true (or false) and provide meaning while following some rules. For example, in English I can say “The dog is sleeping on the bed.” English speakers hear this and are able to understand the thoughts I am trying to express.
Now, while these are the reasons why we may have wanted to move away from the iconic to the arbitrary, why did we end up with language as we speak it now and not stay using gestures? Well, this shift may have been caused by the advantages of the vocal medium once you have moved to arbitrary representations. You are able to communicate across long distances, have your hands free, or just rapidly say something that may have taken longer if you were still using your hands to sign all of it.
Natural language's expressive power allows us to say anything that can be said, the translatability thesis- mainly through subject/predicate propositions. This capacity relies on resolving the symbol grounding problem, where meaning cannot be infinite definitions. Dictionary analysis confirms this structure revealing a grounded vocabulary base, the Kernel. Evolutionarily, language gained its adaptiveness by shifting category acquisition from laborious sensorimotor induction ("show") to efficient verbal instruction ("tell"). The necessary precursor for this propositional leap was flexible, ad hoc pantomime, which relies on Complex Action Recognition and Imitation (CAR&IM) to conventionalize actions. This conventionalization bypasses the dyadic limitations of Ontogenetic Ritualization (OR) paving the way for universally expressive language where translatability is constant even if word count varies. Therefore, the foundational complexity of the lexicon reflects the sophisticated cognitive steps required to transition from action based communication to universally propositional language.
ReplyDeleteThe interesting part of this article, to me, was this statement: “MinSets are not unique, this dictionary has a vast number of MinSets”. Many combinations of grounded words can form a MinSet, which allows to ground all the remaining words in the dictionary through their definitions. A MinSet is required prior to opening a dictionary, because without any grounded words, then no definition would be understandable to you due to their cyclical nature. Now, this sentence made me wonder whether there would be words present in every MinSet across a single dictionary, words that are essential to a MinSet. First, every MinSet is equivalent to a feedback vertex set, meaning that any circuit is connected to U, the grounding set. A grounded MinSet renders the dictionary word set acyclical, as there is one word that can be grounded with no unknown predecessors. With these assumptions in mind, it is safe to assume that MinSets from the same dictionary share words, as they must each include enough Core and Satellite words to break all definitional cycles. As for a word essential to every MinSet in a dictionary, there is no clear answer.
ReplyDeleteHere, Pinker seems to push the idea that POS is evidence for UG. However, the examples used there are not UG. They are OG mistakes which can arise from overgeneralizing a linguistic rule and then be corrected. You will never come across a UG mistake, which leads to the question: how do you acquire UG then? Let us go back to the mushroom island as an example. You are trying to learn what an edible mushroom is, but on this particular island there are only edible mushrooms. How could you ever acquire the features of the edible mushroom category if you have never seen what isn’t an edible mushroom? Similarly, how do we acquire the rules of UG if we never encounter a rule violation? I believe that is the crux of POS and it has nothing to do with whether you are able to figure out that a sentence is grammatical or that English doesn’t allow giggled. That is still in the realm of OG.
ReplyDelete“We suggest that this is how the proposition was born. Learners may have begun picking up new composite categories through passive observational learning. But once the knower— motivated to help kin or collaborators learn by sharing categories—became actively involved in the communication of the composite categories, it became intentional rather than incidental instruction …”
ReplyDeleteBlondin Massé et al. (2013) suggest that the power of language was not invented but rather discovered through our existing ability to categorize and learn from others. Early humans could combine pre-existing categories to create new composite categories. For example, if A = fruit and B = red objects, then A+B could form a new category, C = red fruit. This ability enabled humans to acquire new “concepts” socially from those who already understood the original categories. Learning thus shifted from passive observation to intentional instruction, where category knowledge was actively taught and shared. This transition later marked the emergence of propositional language—statements consisting of a subject and predicate that can hold a truth value, such as “that apple is red and round.”
Organisms' success depends on categorisation, which is the ability to "do the right thing with the right kind of thing". Historically, category learning was primarily through induction, which is slow, risky, trial-and-error sensorimotor experience. The adaptive advantage of human language is that it allows for instruction, which allows us to rapidly acquire new categories by word of mouth. This is proven useful in the mushroom categorisation example.
ReplyDeleteThe power of human language is in propositionality, our capacity to express statements, a subject & predicate, that have a truth value (T/F). Early hominins had categorisation, intentional communication, and pantomime. However, pantomime alone is communication, not language, because it is insufficient to convey new composite categories. The evolutionary transition "from show to tell" involved moving to the "propositional attitude", combining increasingly arbitrary category names into propositions. For this to work, our words must be grounded in sensorimotor experience, preventing meaning from being "dictionary look-up all the way down".
I find the question of why non-human primates have not developed the capacity to learn natural language quite interesting. It seems that, until now, there is a missing piece in the puzzle, given that they possess the capacity to categorize and communicate but cannot speak. In this paper, it is argued that animals lack the motivation to learn how to communicate through language, unlike humans. This question may be related to the Bischof-Köhler hypothesis, which we do (sort of) have a potential explanation for; it has to do with motivation as well. In brief, the Bischof-Köhler hypothesis details the idea that animals cannot assign reward to future goals. In other words, chimps cannot aspire to study hard so that they can get into a prestigious university and climb their way up the corporate ladder to become CEO of a company. They lack the motivation to plan for a reward that isn’t immediately necessary (like eating, reproducing, etc.) Some literature suggests that this discrepancy may be because humans have a capacity to “mentally time travel”, orchestrated by the brain regions that allow for meta-cognition and awareness of the self in the world, that is unique to our species. Surely, one could posit that the progression from non-linguistic communication to natural language in humans may have come about as a result of our ability to identify a reward in the distant future and plan accordingly, thanks to brain structures present only in the human PFC. Perhaps subconsciously, the human mind recognized the evolutionary advantage it would gain from practicing categorization by instruction and “planned” accordingly. Chimps, on the other hand, due to their different brain wiring, never deemed it directly necessary (maybe).
ReplyDeleteBeran, M. J., Perdue, B. M., Bramlett, J. L., Menzel, C. R., & Evans, T. A. (2012). Prospective Memory in a Language-Trained Chimpanzee (Pan troglodytes). Learning and motivation, 43(4), 192–199. https://doi.org/10.1016/j.lmot.2012.05.002
Reading this paper made me wonder whether the incredible efficiency of instructional learning comes with a hidden cost. Language lets us acquire categories almost instantly - no trial and error, no risk, no slow buildup of experience. That’s clearly an evolutionary advantage. But it also means we now put far less effort into learning directly from the world. Inductive learning, though slower, produces fast, automatic, perceptually grounded responses. Those are the kinds of reactions that matter when something dangerous happens - the kind that kept our ancestors alive long before language existed. Instructional learning can’t replace that; in fact, it produces slower reaction times, even when the rule is understood.
ReplyDeleteI wonder whether our growing dependence on symbolic, instructed knowledge has made us less behaviorally fit in certain ways. Have we traded effort and embodied experience for a kind of abstract, rule-based understanding that isn’t always aligned with real world demands? And if so, is this a progress trap where the cultural benefits of language unintentionally weaken some of the capacities that induction once strengthened?
I really enjoy your thought process here. I too was asking myself similar questions while reading. I believe this is a question that humans have struggled with in many domains and will continue to in the future. It's somewhat of a paradox that making things easier to do will widen our capacities to do so, but the lack of effort may effect the end result.
DeleteThis same thought process can be applied to how technological advances have been extremely helpful in many ways, including learning and teaching, but have also potentially brought many problems. Inductive learning, in your arguement, can be compared to the way students used to research things. We used books and spent hours at the library searching for sources and ideas, whereas now we can find hundreds of sources with a few words in a search engine (let alone the AI that can fully write the paper, with the sources necessary). This is meant to make things easier for us so we can more time to learn more things. However, maybe the lack of effort has made us learn less?
What makes this paper especially fascinating is its concrete, non-mystical explanation of how language could have evolved from simple category learning. Instead of focusing on syntax or vocal evolution, the authors argue that the true breakthrough was the shift from showing to telling: the ability to acquire new categories through instruction rather than slow, risky trial-and-error. Their simulations show that “symbolic theft” provides a massive evolutionary advantage, and their analysis of dictionaries reveals a grounded core of concrete words from which all others are defined. This framework elegantly links symbol grounding, category learning, and the emergence of full propositional language.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteI found it interesting that both Pinker and Bloom earlier, and the paper by Blondin both saw language as an adaptation. The first argued that language originated through natural selection, whereas the latter argued that language originated from the need to transmit categories efficiently and to teach new categories through propositions. To support their argument, Blondin and colleagues focused on categorization and symbol grounding. I found the dictionary example very interesting. In fact, they argued that all the words in the dictionary are content words (names of categories) and that many of those words can be learned through 'trial-and-error sensorimotor experience, guided by feedback that indicates whether an attempted categorization was correct or incorrect'. This is induction, which is the slow biological way of learning categories that animals and humans use. However, humans also use instruction, which is learning by word of mouth. Going back to the symbol grounding problem, this paper addresses it because it shows that in any dictionary, every word is defined using other words, which creates a loop where nothing is grounded unless some words are learned from direct experience before others can be learned. For instance, one might first learn what an apple is through seeing, touching, or eating it (induction). Once that basic category is grounded, one can then understand its characteristics (round, red, fruit) through verbal explanation, or even learn new categories through propositions like “an apple is a red fruit”. Thus, the power of language doesn’t come from naming alone but from the ability to combine grounded category names into propositions that define new concepts. This makes instruction exponentially more efficient than induction, which is the reason language itself evolved, as it allowed humans to bypass slow, risky trial-and-error learning and instead share categories rapidly and safely. The shift from “show” to “tell” thus becomes the central adaptive advantage that explains the origin of human language.
ReplyDeleteWhile reading the paper, the concept of expressive power really stood out to me. This is the idea that there is nothing you can say in one language that cannot be said in another. This reminded me of moments where I know an expression in one language, but when I translate it literally, it loses its meaning. Yet when I explain the idea behind it, the person often finds an equivalent expression in their own language, even though the wording is completely different. This actually illustrates what the authors argue: expressive power depends on the ability to express the same proposition, not on using the same words. Because each language grounds basic concepts in shared human experience, speakers can reconstruct meanings even when the phrasing changes. So, expressive power ultimately rests on symbol grounding since once the core concepts are grounded, any language can combine them in different ways to express the same thoughts.
ReplyDeleteI was a little confused regarding the distinction between the Grounding Kernel and the Minimal Grounding Set described in this paper. So correct me if I’m wrong but here’s my attempt at summarizing them. The grounding kernel represents the approximately 10% of dictionary content words out of which all other words can be recursively defined. These kernel words are generally concrete, frequent, and learned at a young age. So, definitions within the kernel are circular, and the grounding kernel itself is a unique set.
ReplyDeleteThe minimal grounding set is related but it's a fundamentally different concept. It is defined specifically as the smallest number of content words required to define all the other words in the dictionary. Unlike the unique grounding kernel, the minimal grounding set is not unique, so there are many alternate sets of words that could serve as the minimum grounding set. Basically, the grounding kernel is the cluster of words that are most concretely grounded, but the minimal grounding set is the absolute smallest possible foundation needed to achieve the full expressive power of language.
Massé et al hypothesize how language evolution involved human communication occurring in a manner of showing to increasingly in a manner of telling that is more abstract through Baldwinian evolution. The fact that it may have occurred through Baldwinian evolution connects the hypothesis to Pinker's paper which argues for language being innate. The hypothesis also relates to the Arbib paper because it outlines how categories may have come to be able to be shared instructionally and social / without direct experience.
ReplyDelete“The power of language (according to our hypothesis) was, in the first instance, the power of acquiring a new composite category from other people who already knew the old categories out of which the new one was composed.” (Blondin Massé et al., 2012, 9)
ReplyDeleteThis theory of the origins of language highlights the importance of socialization and collaboration between humans. The “power” of language, and what made it possible to develop in the first place, is this capacity and motivation on one side, for the learner to learn from someone else, and on the other side, for the teacher to teach and pass down their knowledge. This way, both sides are needed— even if the learning is unconscious— for knowledge to be transmitted, for information about new categories to be shared. This emphasizes the fact that language is dependent on and necessitates socialization.
What’s especially important is that the “power” of language isn’t just internal cognition but socially mediated learning. Language only becomes useful once there is both a learner who can acquire categories indirectly and a teacher who already has grounded categories to transmit. This brings into mind grounding. Indirect grounding lets you steal knowledge from others, but only if the symbols used in the proposition are already meaningful to you through direct sensorimotor grounding. Otherwise, you’re stuck in the Chinese–Chinese Dictionary loop, where symbols never touch the world. This also shows why language couldn’t bootstrap itself from nothing. Someone, somewhere, had to ground the base categories the hard way. This makes language inherently cooperative and also helps explain why language didn’t evolve in isolation within single minds but within social groups where sharing hard-won knowledge (like dangerous or useful categories) dramatically reduced learning costs for everyone involved.
DeleteThe paper discussing the minimal grounding set in the dictionary is quite fascinating. I am now wondering which words different languages might use in their dictionaries as the minimal grounding set and if there are any patterns that are similar across all written languages. I am also wondering how a unilingual speaker of an exclusively oral language might decide to write their dictionary and if that would or would not match patterns that are seen in other written languages.
ReplyDeleteMassé et al. offer one of the most compelling accounts of how language could evolve from showing to telling. The idea that communication began in iconic gesture makes intuitive sense as gestures can directly resemble actions or objects. What makes the transition powerful is de-iconization, which si the move to arbitrary symbols that are combinable and no longer tied to perceptual resemblance. This shift enables indirect grounding which allows knowledge to be transmitted propositionally rather than discovered through trial and error. Still, even this framework leaves open the deepest puzzle, which is how arbitrary symbols first became combinable into truth-valued propositions at all.
ReplyDelete