Saturday, January 6, 2018

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

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.

67 comments:

  1. I was very surprised by the result from the Cangelosi and Harnad study where the researchers found that the creatures that were able to learn about a category from other creatures (ie: through instruction) as opposed to only through induction out-survived the other group after just a few generations. I initially thought that directly engaging with the category in question would give the induction-learners a more profound understanding of the category and thus grant them some sort of advantage over the other group. Upon further consideration, I see that really it doesn’t matter in the end how the category was obtained because the category (like any other) will be abstracted in both instances and be equivalent in this sense. As the article mentions, the effort needed to engage in the abstraction is very different where the instruction learner group didn’t need to put in as much effort to learn about the category and without the dangers and length of time needed to learn something by trial and error. This idea ties into the Pinker article for this week on the importance of communication and social groups for survival purposes. If people were able to impart categories onto others or to be able to observe and abstract from the behaviors of others this would dramatically increase a person’s survival rates and as the above mentioned study demonstrates, over a few generations, only the descendants from these instruction learners would remain in a population.

    ReplyDelete
    Replies
    1. But the crucial thing (besides the power of propositions) is that some categories need to be grounded directly, the old way, in order to ground all the rest.

      Delete
  2. "…it seems as if fewer than 1500 category names plus a few functors and syntactic rules is as much as you need in order to get language's full expressive power."

    This paper discusses the advantages of learning categories through instruction rather than induction. For one, it's much less time-consuming and risky compared to learning through trial-and-error. However, induction is still the fundamental first step in this process. Once we have grounded this minimum set of squiggles and squoggles, everything else isn't just squiggles and squoggles anymore - we can combine the grounded squiggles and squoggles to learn all other categories in a very efficient manner. We can even use this combinatorial method to create new categories that can't be attained through sensorimotor grounding (eg. unicorns).

    Thus, everything is possible based on sensorimotor capacity, which again drives home the point why cognition is not just computation and T3 is the minimum to explain everything that we can do. T2 may have the capacity to use grounded words and combine them to learn new categories. However, T2 lacks the capacity to learn categories through induction, and consequently lacks the grounding kernel necessary to reach all other words by definition alone.

    ReplyDelete
    Replies
    1. So T2 only has meaningless squiggles and squoggles, plus shape-based rules for manipulating them.

      Delete
  3. “Learners may have begun picking up new composite categories through passive observational learning”. The discussion of induction & passive observation as the first steps to developing language reminded me about the work of Karl Popper. From what I remember, hasn’t he extensively refuted the idea that induction exists, or at least that that is something which humans do not/cannot do? I think his argument is essentially that you cannot passively learn through observation, because observation (for the purposes of learning in the way we are interested in) is always selective and would presuppose categorization/language/interests... So the inductive process, while useful in some respects, isn’t what people do and therefore not part of the explanation we are looking for when explaining human cognition?

    ReplyDelete
    Replies
    1. You can learn some categories by observation alone (e.g., mountains vs. valleys, zebras vs. giraffes). But not most categories.

      Delete
    2. To Hugo, do you mean the instructive process? Maybe I am completely missing something, but isn’t induction learning through direct experience and instruction by word of mouth (or in the example, by passive observation)? The inductive process, as I understand it (based off the reading and those definitions), would be sensorimotor engagement with the environment and is very much what people do to learn a lot of things. Maybe I am also not wrapping my head around what Karl Popper is suggesting, as I cannot imagine how one could argue either induction or instruction could possibly not be part of the equation – as we as people seem to do both pretty frequently as we go about living and learning in our lives.

      Delete

  4. “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, with the teacher actively miming the new category descriptions to the learner. (This motivation and strategy would be a natural extension of existing strategies to execute skill so that others can imitate it.) The category “names” would also become shorter and faster—less iconic and more arbitrary—once their propositional intent (the “propositional attitude”) was being construed and adopted mutually by both the teacher and the learner.”
    I really liked this passage because it’s relatively intuitive and it demonstrates the process of induction to instruction that is inherently shaped by culture. It was hinted earlier that the benefits of instruction depend essentially on mutualism and this, I believe, shows how it’s dependent through the relationship between teacher and learner. It reminds me of words that are created by academics/intellects that are gradually disseminated through other institutions and eventually to the lay public, which all relies on an agreement of the propositional intent.

    ReplyDelete
  5. « So as the power of language in category acquisition became encephalized, the tendency to assign arbitrary names to categories—and to combine their names into propositions defining new categories—migrated to the auditory modality, which was already so admirably prepared for the task in other species as well as our own (though no doubt it too had to undergo a period of intensive further encephalization, under selective pressure from the adaptive advantages of increasingly efficient speech). »

    I am not sure I understand what the authors mean when they say the tendency migrated to the auditory modality. Are they suggesting that as the arbitration of categories became increasingly easy, we became more motivated to share those arbitrary names as opposed to the experiences behind them? Or is this suggesting that once a category has had an arbitrary name attributed to it, it becomes easier to describe thus making this an evolutionarily advantageous way to categorize?

    ReplyDelete
    Replies
    1. Once signs became arbitrary rather than iconic or instrumental, and once their combinations became propositional, the sense modality no mattered. (That's part of being arbitrary!) So the practical advantages of the verbal/acoustic modality took over.

      Delete
    2. In thinking about our day to day life and our experience in learning something, have you ever learned a new idea simply by the person just using the arbitrary name for it? For instance in a biology class when the professor introduces the mitochondria he doesn't simply move on and then use the arbitrary template mitochondria in describing other phenomena, but rather he defines and explains the meaning of this arbitrary category using other arbitrary categories that you are already familiar with. Then you study the meaning of the term mitochondria and eventually it gets filed away in your drawer as a word you know without needing the definition. Further the term mitochondria is extremely complex and takes many arbitrary words to understand and each of those arbitrary words take words to explain (symbol grounding idea) so rather than "as the abitration of categories became increasingly easy" it is the opposite that these categories became increasingly more complex and words became simpler than gestures because you could combine many many gestures into a single utterance. Also as the paper mentioned, if your hands were full it was advantageous to be able to speak.

      Delete
    3. Yes, words (and their definitions) have a powerful role in "chunking." We lexicalize words (coin a name) that are useful in what else we say and do; it's much more efficient than repeating a long-winded definition or description every time (as you note).

      On chunking, see this important paper that was discussed in Week 6 (or categorization):
      Miller, George (1956) The Magical Number Seven, Plus or Minus Two: Some Limits on Our Capacity for Processing Information. Psychological Review 63:81-97

      In maths they say the test of whether you have proved a useful theorem is if it generates a definition (i.e., some new thing to name for future use).

      Most of our categories are not lexicalized, however. Because we only use them occasionally, on the fly. The example I gave in class was "things that are bigger than a breadbox." You won't find that in a dictionary, though we could have called it "Suprapanarcachic" or "Megakivotpromic" or just "Hyperbreadlarderate"... or (these days) "BBoXX"...

      Chunking happens in piano-playing when your working unit grows from single notes, to scales, to arpeggios, to chord sequences, to melodies... And in chess, with bigger strategic chunks and anticipations. (Funes presumably couldn't do any of this because his only "chunk" was each individual instance in its infinite detail, exceeding all verbal description, and eligible only for a unique, never-to-be-forgotten proper name: that's why Funes gave a proper name to each number instead of using, say, the decimal system to name them.)

      Delete
    4. To Zachary, if I understand the section you’ve quoted correctly then I believe the authors do a good job at reiterating this in a more easy to understand way:
      “one a species picked up the linguistic ball in the gestural modality and began to run with it...the advantages of freeing its hands to carry something other than the ball, so to speak...along with the advantaged of naming when the teacher was out of the learner's line of sight....would quickly supersede the advantages of gesture as the start-up modality for language.”

      Delete
  6. I found this reading very informative and straight-forward. It outlined the different aspects of language including UG and learning categories by induction versus instruction. We discussed a few lectures ago that anything that can be said in one language can be said in another language even touching on the 'words for snow' problem. I'm wondering how this applies to meaning, for example, idioms. There are some expressions whose meaning would not translate to another language (ex: 'kick the bucket'); does this provide a limitation to the universality of language? If so, is it an important limitation?

    ReplyDelete
    Replies
    1. “Sa dai cu piciorul in galeata inseamna sa mori.” We don’t have the idiom “kick the bucket” in Romanian, but its MEANING could be easily conveyed with the previous sentence (“To kick the bucket means to die”). To describe the origin of the idiom would take a few more words, but even I didn’t know what it referred to in English (other than death) before I googled it just now. And once the reference is understood, you could easily adopt the literal translation (“sa dai cu piciorul in galeata”) in Romanian (or whatever language), even if it sounds a little awkward.
      From Symbol Grounding and the Origin of Language: “For those readers who have doubts, and think there is something one can say in another language that one cannot say in English: please tell us —in English, at any desired length—what that something is, and why one cannot say it. You will find that you have generated your own counterexample.” In terms of limitations, they did at one point cover cases that elude the grasp of language, for example you can describe the colour red as much as you want, but a person won’t truly have understood until they’ve seen it themselves (direct sensorimotor grounding/experience). Which I’d argue isn’t really a LIMITATION of language so much as a part of it.

      Delete
    2. While I think your analysis of idiomatic expressions is interesting, I don't necessarily think it's fully accurate. Just because there is not a specific phrase that acts as a metaphor for dying in Romanian doesn't mean that you can't express that idea, as pointed out by your first sentence. Furthermore, there might just be a slightly altered idiomatic expression that expresses the same ideas. For example, in Spanish we use the phrase "llover a cántaros" which directly translates to "to rain buckets", which is the Spanish version of the English phrase "raining cats and dogs". These two phrases might not be direct translations of each other but they do express the same ideas. Also, your point on the colour red (and more generally, colour vision as a concept) asks the larger question as to whether you can understand colour without actually seeing it. This is what is known as the "knowledge argument" where there is a scientist named Mary who lives in an entirely black and white room and has never seen colour, but studies it intensely and knows everything about it. So, does Mary know what colour is, or does she not? What would happen if she were released from her black and white room? Would things be exactly as she'd hypothesized after years and years of scientific inquiry on the concept of colour or is she blown away by what colour is in the environment?

      Delete
    3. I think we're arguing about the same thing here, I probably wasn't as clear as I should have been. You're exactly right that you can express the same idea in whatever language with equivalent idioms/sentences. And as for Mary's room argument, if we understand learning/language as necessarily grounded in some sensorimotor experiences, Mary's knowledge was incomplete to begin with and she'd certainly learn something new upon seeing the colour red

      Delete
    4. Tina, translatability refers to content, not form. So you can convey the content, but with different combinations of words, not word-for-word translation. That rules out verbatim idioms, but you can give their equivalent, or at worst, describe what they mean and why (as Christina and Allie point out). And, yes, the direct sensorimotor grounding of at least some words is necessary, otherwise language is meaningless and circular. (Once grounded, language remains circular, but not meaningless, because sensorimotor grounding breaks the circle.)

      Delete
  7. When we put the induction-only learners in competition with the instruction learners, within a few generations the instruction learners had out-survived and out-reproduced the induction learners, of which there were no longer any survivors left.
    Although this finding is quite simple in results what it says about evolution and the way that we evolve is incredibly interesting. One question that immediately comes to mind from this is whether the ability to verbalize (as in the instruction condition) allowed for the learners of this group to better understand this idea. For example, in many cases in our day to day life we have times where even though we can’t objectively say what the correct response is to a given situation (as in the induction learning group) we can FEEL that one response is better than another and consequently pick this. However in a case where we are given the instructions “x will result in a better outcome” it is quite obvious that this would lead to better results. In these cases the individual with the instructions is able to question the information; they can consider whether x really will lead to a better outcome. They are able to think of cases where x lead to a better outcome or even choose the alternative to test a hypothesis which will ultimately lead them to conclude that the information provided was correct (all this being much quicker and more precise than the method of induction).

    ReplyDelete
    Replies
    1. No question that verbalizing -- spelling out in longhand, making explicit -- helps us analyze, understand, remember, explain better. (It can be overdone, though. Sometimes leaving out a few details is a good idea: gives the listener more of an active-kitten role; best of all, of course, is verbal interaction, including skywriting.

      Delete
  8. Re: Section 4 Categorization

    When we encounter symbols and shapes in the environment, it is through a direct sensorimotor experience that our later ideas of them are formed. It is then through these ideas that novel categories or concepts are born. While it is obvious that sensorimotor induction is the very basic means or prerequisite for our learning to categorize, instruction facilitates our ability to recognize new aspects that we may not have encountered prior. It is crucial to understand that instruction must come after induction; a slow process followed by a quick process. Cangelosi & Harnad (2001) demonstrate how instruction may be superior to induction in its ability to form novel categories. Pantomiming required prior induction of objects and experiences in the world. They state that it is impossible to explicate a novel category by pantomime, but I’m not sure I agree given that we can do this using words.
    A good example of this is when we must combine concepts from two different categories to form a new one. If we imagine a creature that had the body of a dog and the head of a lizard (let’s call it a Dizard) we first have to imagine these two categories of animals seperately and combine them to form the Dizard in our mind. Obviously, this kind of creature could never exist due to the constraints of biology and genetics, therefore no one has ever seen a Dizard in real life (i.e. we lack sensorimotor induction of this creature), but we are still able to imagine what it could look like. Yes, induction of simple categories is a prerequisite, but instruction can inform and infinite amount of new categories.

    ReplyDelete
    Replies
    1. I really loved your example in regards to the Dizard. It is a perfect way to explain how instruction is important and necessary in terms of out-surviving the induction learners. Like you said, instruction has allowed us to go beyond reality and form concepts that may not be real yet we are able to categorize them.

      Delete
    2. To copy something -- whether by imitating it or drawing it -- is not to say anything about it. It just draws attention to it, as pointing, too, does. One could draw attention to a critical feature that way, a feature that can be used to distinguish category members from non-members. But for that you need to be thinking of categories (kinds), rather than just imitating individuals, or parts/features of individuals.

      You could also imitate a combination of two things -- but that's not a proposition either, just a hybrid of the two.

      We are probably all over-estimating the power of imitation today, because we have language now, and everything we see and do is narrated -- or readily narrated -- verbally. It is hard to imagine the imitation without the sub-titles it comes with.

      The mushroom simulation was just a toy model; it did not have propositions, just a conjunction of features ("and") and a conjunction of names. It would require a lot more detailed work to show that this could scale up to propositions describing objects in terms of their grounded features.

      In sum, an imitation is not a proposition (nor is it instruction), although it could potentially be used as a category name once the association between the gesture and the category is known and being used in intentional communication, and especially once the gesture/name has lost its iconicity and become arbitrary.

      Delete

  9. “But given how smart hominins were already, and how all the pieces were already in play, such a “mutation” to adopt the “propositional attitude”—if a downright mutation was indeed needed at all—is not far-fetched to imagine.”
    I tend to disagree with the statement that a mutation of the sort causing a propositional attitude is not far-fetched to imagine. It seems a “just-so” story. The reading goes on to say that it is possible that propositional attitude evolved due to its advantages to those who were motivated to use it. But I think it is important to identify what those motivations may have been. Rather than a genetic mutation, I think it is much more realistic that the necessary cognitive elements were in place; for example, the ability to form associations and contingencies. However, it seems likely that environmental pressures were also in place where survival would stand to gain an extreme advantage through increased cooperation via communication ( such as pantomime). Perhaps the reason apes do not seem to show the motivation to pursue such communication is that they do not face these environmental pressures, especially in a zoo setting. However, early humans, without fur, nor strength on par with monkeys, and a multitude of predators, seem to have been in an environment prone to the development of more effective communication.

    ReplyDelete
    Replies
    1. Yes, the evolutionary change could have been a Baldwinian disposition to adopt the propositional attitude, rather than a cognitive mutation.

      Delete
  10. This article was very informative on the origin of language and symbol system with regards to Universal language. It was interesting to see how symbols part of formal language such as arithmetic equations that have output symbols of True and False are interpreted as just equals and how we completely disregard that in interpreting equations. This is something that has never really occurred to me regardless of classes I've taken within the realm of logic. This makes me think how its interesting to see how we categorize language itself ; into natural language and formal language and how we use language to do so.

    ReplyDelete
    Replies
    1. Most words in natural language (nouns, verbs, adjectives, adverbs) are content words, naming categories. But there are some function words (not, if, and, equals) that are more syntactic, as in maths.

      The point in maths, though, is not that the symbols have no meaning, but that you cannot use the meaning when you do a formal proof.

      Delete
  11. The article states that “With the exception of a few function words, such as the and if and not, most of the words in a natural language dictionary are content words, and as such they are the names of categories.” This would imply that the content words of the natural language are in fact learned through experience of grouping items based on relevant invariant features, and not derived from an innate understanding. In contrast, Universal Grammar (UG) is “complex set of rules that cannot be acquired explicitly from the child’s experience because the child does not produce or receive anywhere near enough information or corrective feedback to learn the rules of UG inductively, by trial and error” suggesting these rules are inborn and therefore innate. If both these claims are accurate, is it to say that the UG rules we are born with are not categorical?

    ReplyDelete
    Replies
    1. This is an interesting question. I don’t think “innate” and “categorical” are mutually exclusive though. As with the mountains and valleys example Professor Harnad gave in class, some very basic categories (although not all) must be innate. On some level you have to be able to see that things are different if you’re to classify them at all. Further, I think that the rules that shape UG-compliance are indeed categorical. I’m assuming nobody in the class actually knows the rules, but we can usually tell when a phrase or sentence violates them. Either something is UG-compliant or it isn’t, and when it isn’t, it’s because it violates one of the rules. There’s nothing in between.

      An interesting question that arises from this is whether there are other categories, on the same order of complexity as UG, which are also innate. Aside from knowing UG, the examples generated in class were extremely simple (i.e. being able to tell when something occupies space vs. when it doesn’t). Is UG the only example of a very complex and systematic set of engrained categories? And if so, why?

      Delete
    2. UG-compliance is all-or-none, categorical, as Willem notes (though some linguists have suggested that sometimes compliance can be a matter of degree: better/worse; that's a controversial issue, so I can't say anything about it.

      Many sensory categories (e.g. colors) are innate. Figure/ground. Probably also numerosity (0, 1, 2, 3, more). The frog's bug-detector. The duckling's disposition to follow the first moving object (as mother). Many of these dispositions are Baldwinian (quick-learning) rather than rigid.

      Delete
  12. “This, then, is our hypothesis about how and why language began. As many others in this volume report, all the essential cognitive components already seem to have been in place in other hominins 250,000 years ago— Just as many of them are also present in our contemporary ape cousins as well as in other intelligent social species today: the capacity to learn the features distinguishing categories by sensorimotor induction; the capacity to learn by observation and imitation; the capacity for pointing, shared attention, and mind-reading”
    This passage discusses the idea that the cognitive components of language were present before language itself, and it is these components that led to the development of language. I have a few questions regarding this; is it the case that these cognitive components are the basis for UG, or just the precursors to the development of OG through experience and time? Also, did the capacity to use these components to form language develop through new brain structures/evolution, or did we simply learn how to use these components in a way that allowed language to emerge?

    ReplyDelete
  13. Re: Section 1. Show vs. Tell:
    “In many ways, the origin of language amounts to the transition from show to tell.”
    “PowerPoints are “show and tell,” but more show than tell.”

    Section 1 led me to think about the role technology plays in this ‘show-to-tell’ transition...

    Language evolved from gestural ('show') communication, to vocal ('tell)' communication. However, with the advances in technology, communication is becoming increasingly encoded in ‘show’ representations, whether it be in PowerPoints as mentioned here, or even in emojis, memes, and gifs as we discussed in class a couple weeks ago. We are returning to a world of language without orality, where ideas can be communicated without speech. Hence, I contend that with this technology-induced increase in ‘show’ communication, ‘tell’ communication is regressing, and the ‘show-to-tell’ transition is reversing.

    ReplyDelete
    Replies
    1. Maybe technology enhances show over tell. But it sure facilitates tell too (google plus fake-news). Week 11 will be about cognitive technology.

      Delete
  14. “Chimps have categories... They just don’t seem to be motivated to do so, even if they sometimes seem to “get it,” locally, for individual cases.”

    “We are not sure whether chimps really do get it…What seems missing is not the intelligence but the motivation, indeed the compulsion, to name and describe.

    I am struggling to understand the authors’ explanation for the human development of language to be a species-specific motivation, when other species are social creatures (i.e. apes) and thus I would assume, would too be inclined to improve communication and acquire language.
    What is this motivation that drove humans and not apes? What do the authors mean when they refer to motivation? What sort of motivation were apes lacking that humans possessed? And what primitive social differences would have generated the vast differences in motivation?

    ReplyDelete
    Replies
    1. I don't know the answers to most of these questions. But a motivation is a disposition to do something, like the child's disposition to like and eat sweet things. Without it, they would be indifferent to candy...

      Delete
  15. I am a bit confused with inductive and instructive processes, or more exactly where can we draw the line between the two. Is instructive learning exclusive to language? If induction is the learning by direct experience, couldn’t other species also learn to categorize based on instructions that are verbal but not linguistic per se. I can only imagine a situation where a baby animal would try to approach a predator and his mother would make a sound indicating “don’t do that”. The baby animal would not have had direct experience to categorize the predator as dangerous, but with this signal from his mother, he could still learn to categorize. Some people are probably going to argue that the sound made by his mother is actually a direct experience, thus this scenario would be considered as induction. But then what isn’t a direct experience…?

    ReplyDelete
    Replies
    1. Induction is unsupervised and supervised learning (trial/error/corrective feedback). Instruction is mostly verbal (though some showing is a kind of "instruction" too, but nonverbal instruction, which is hard for us compulsive verbalizers to separate from verbal instruction).

      Warnings and threats and requests can be verbal statements or nonverbal communicative acts. When animals hear human verbal warnings, etc., they can treat them as nonverbal communicative acts. But they are certainly not induction (thought they could serve as corrective feedback in learning by induction).

      Delete
  16. What I particularly liked in this paper was the mention of humans trying to teach chimpanzees to talk. While they apparently were able to categorize things and name them they never seem to evolve their capabilities to form propositions. It is especially interesting because the parts of the brain that are associated with speech and language in humans, particularly Broca’s area, are also found in chimpanzees but do not serve the same functions. Why similar brain structures came to evolve so differently is fascinating and truly makes us question why we evolved the way we did and developed language. The explanation the authors give is a valid one, but one can’t help but feel like there must be more to it. If that’s all there was to it, then why did the chimpanzees not also develop language? They might not have been motivated to do so or might not have needed so badly for survival, but then what motivated the humans?

    ReplyDelete
    Replies
    1. I dont think we have the answer on why chimpanzes havent learned to talk yet. Although, its incredibly fascinated to contemplate why we developed language yet chimpanzes havent also developed language. As someone whos works with children with autism, youd be surprised how many parents are absolutely convinced there kids havent develop language or have hearing impairments. However, it turns out that the kid could talk, had developed BOTH OG and UG but just weren't motivated and didn't feel a need to actually speak. Either because their parents would just figure out what they wanted through trial and error during tantrums or they had just learned to independently do things they wanted without asking. That being said, I dont believe the issue is motivation in chimps as you could use similar methods to coerce them into speaking using something they are highly motivated by and still come to the conclusion that they could'nt use language.

      Delete
    2. Yes, many of these questions are not yet answered. (I'd say the lack of motivation of autistic children to speak, even though they can, is not quite the same as the lack of motivation of chimps to use signs propositionally. But there might be something to be learned there...)

      Delete
  17. "Because we were more social, more cooperative and collaborative, more kin-dependent—and not necessarily because we were that much smarter—some of us discovered the power of acquiring categories by instruction instead of just induction, first passively, by chance, without the help of any genetic predisposition."

    A fascinating point, and one that fits in well with the paper's overall argument. However, is this true? Are we as a species actually more social than our living ancestors? I'd be interested in seeing data to back this up.

    I agree that motivation may well have played a part (though not the sole role) in the development of vocal language, but why, given the chance, do chimps not have this "motivation" while we do? Is it in their genetic makeup to not be motivated for the development of language, or is there another force at play?

    ReplyDelete
    Replies
    1. I don't know why chimps don't use (human) gesture language propositionally. (But it's not partivularly about the motivation for vocal language.)

      Are humans more social than other species? Certainly in terms of the quantity, breadth and quality (propositional category-sharing) of their social interactions. Some species (e.g. ants) may exceed us in some social ity, but overall it seems a reasonable (and uncontroversial) observation.

      Do you have some other forms of sociality in mind?

      Delete
  18. "Because we were more social, more cooperative and collaborative, more kin-dependent—and not necessarily because we were that much smarter—some of us discovered the power of acquiring categories by instruction instead of just induction, first passively, by chance, without the help of any genetic predisposition"

    I find this passage to be quite important because it marked some significant questions I have about social behavior and the origin of language. Say language is a social evolution that allowed us to pass on categories through instruction; I find it incredible that at some moment in time, two hominins grasped a category at the same time and were able to build on that.

    It seems that language originated from a mix of neuronal predisposition, social behavior, and the right homonim in the right time and place.

    ReplyDelete
    Replies
    1. Perhaps in the family settings, where parents are eager that their progeny (hence, distally, their genes) acquire new categories they need. Also, the parent usually has the category and the youngster needs to learn it.

      Delete
  19. "Since then, some people have tried to answer the question anyway. But explaining the evolutionary origin of UG faces some problems, because UG is not the sort of thing—like wings, limbs, hearts, or eyes—for which a straightforward evolutionary story can be told without an awful lot of ad hoc hemming, hawing, and vagueness. Other options are to deny that UG exists at all, or to argue that it can be learned after all, and hence that there is no need to worry about how it could have evolved.”

    Although I certainly do not deny the existence of a Universal Grammar, I find that much of our class discussion, and this paper, unquestionably accepts UG. This paper acknowledges the UG debate, but glosses right over it and jumps right into discussions about language and language acquisition that rely very heavily on Universal Grammar. I think stronger evidence for the existence of UG is called for. How could we test if UG is learned or engrained? What about cases of language deprivation until later childhood or adulthood? If individuals deprived of language past the critical period can acquire language (although perhaps broken/incomplete language) that is still in accordance with UG, then this is strong evidence for UG being engrained in the brain. If there are not cases of language-deprived individuals forming phrases in accordance with UG, there are two possibilities: either UG is not engrained, or UG is engrained but can only emerge during the critical period.

    ReplyDelete
    Replies
    1. The evidence that languages all conform to UG is overwhelming. So is the evidence that there is no negative evidence (for supervised learning) and of course no instruction in UG. And it is provable that the rules of UG cannot be learned without negative examples, by unsupervised learning.

      The question of whether a first language can be learned after a critical period does not bear on the innateness of UG one way or the other: why would you think it does?

      Delete
  20. ""Translatability" thesis: Anything you can say in any natural language can also be said in any other natural language- though it is important to add, not necessarily said in the same number of words. Word-for-word translation is often not possible, as you have discovered if you have used Google Translate. One language may say some things less economically than another. But it can still say them all."

    When reading about the translatability thesis, I enjoyed the picture it painted about the universality of natural languages. It’s a logical idea that any word or phrase you can say in one natural language should be translatable to any other natural language. On the other hand, I was also immediately struck by what seemed to be exceptions to this rule, which the thesis then went on to take into account. Specifically, the thesis accounted for my exceptions by dictating that when word-to-word translation is not possible, some less-economical translation will be. This description makes the thesis true as far as I can tell, but I wonder how far from point-to-point we should allow translations to go before ruling them as being too far reaching to be considered truly translatable.

    The initial exceptions that occurred to me were those words that you often hear of that only exist in one language, and tend to illustrate a very specific situation or feeling. Out of curiosity, I Googled a list of these words. Let's take one of the words I found, the Hindi word jijivisha. Jijivisha refers to the strong, eternal desire to live and continue living. You may respond to me here and say, look! You've successfully translated the word. I would argue that I have not. Though I have given the barebones meaning of the word, I fear that much of its true meaning and significance is lost in the translation.

    I'm not a student of linguistics so I'm not coming at this from a very informed place, but I'll take my best crack at describing what I mean. Firstly, I believe that a word's use and impact is different than that of a phrase in speech and in writing. A word is easier to use, more appropriate that a whole phrase in many situations, and also brings a level of succinctness to a description which a phrase often lacks. In this way, the translation of a word to a phrase causes it to lose a certain quality which is often part of its very identity.

    Secondly, the word jijivisha seems to be one that is deeply engrained within Hindi culture. Therefore, when people in the culture use the word, it brings with it all the weight of its position in the culture and its particular significance for Hindi people. In this way, I feel that it cannot be used in the same way by an English-speaking culture who knows it only by its drawn out factual definition. Therefore I would reject the thesis on the grounds that words like this cannot truly be translated into other languages because of this very lose. Language is something so intrinsically intertwined with the culture and the history of a place that I fear we do it a disservice by the translatability hypothesis.

    ReplyDelete
    Replies
    1. "I wonder how far from point-to-point we should allow translations to go before ruling them as being too far reaching to be considered truly translatable."

      When it gets so far that you can't convey the message -- cannot reduce the uncertainty. (As far as I know, this never happens.) (After all, even with just a picture, 1000 words may not be enough to describe it.)

      "Jijivisha refers to the strong, eternal desire to live and continue living. You may respond to me here and say, look! You've successfully translated the word. I would argue that I have not. Though I have given the barebones meaning of the word, I fear that much of its true meaning and significance is lost in the translation."

      If something is missing, what is it...?

      A single word is punchier, and more portable than a long description. But (as discussed in other commentaries on "chunking" earlier), that's why we lexicalize some descriptions with a single word (but not "things that are bigger than a breadbox"!) "Life-wish" would do it for "Jijivisha" (and "death-wish" already exists as its category complement!), at least to a first approximation. (We can tighten it once you state (sic) what your definition was missing...)

      You have to distinguish a word's literal meaning or denotation (which is what we are talking about here) and its cultural connotations and associations. "Crucify" meaning "to tie or nail someone on a cross till they day" does not convey all the connotations it has for Christians, but like denotations, connotations can be explained too, even if it takes 1000 words on the significance of the cross in Christian culture. (Language has nuclear power! and it's as universal as computation's power to simulate.)

      The same could be said for conveying metaphors...

      Delete
  21. (The Latent Structure of Dictionaries)

    ○ "Here we note that almost all the words in any dictionary (nouns, verbs, adjectives and adverbs) are "content" words, meaning that they are the names of categories (objects, individuals, kinds, states, actions, events, properties, relations) of various degrees of abstractness."

    Trying to navigate the concepts of categorization, language and meaning, I find this passage particularly interesting. When we were first presented with the concepts of categorical versus continuous perception, I was interested in the dichotomy as it applied the language. I understand that much, if not all of language is categorical, and the only concepts left out of these discrete categories are understood to be things like relative judgements, which are considered continuous.
    The passage brings up a question I had held about the categorical identity of language. Although it is easy to accept that nouns, verbs and adjectives are categorical, I was having trouble accepting that more specific concepts, like relations and other abstract ideas were equally categorical in nature. Explained as it is here, however, I can clearly see how most all of the elements of language can be considered categorical- with a varying degree of abstraction.

    "Clearly, every dictionary D has many grounding sets. For examples, the set of all words in D is itself a grounding set. But how small can a grounding set be? In other words, what is the smallest number of words you need to know already to order to be able to learn the meaning of all the remaining words in D through definition alone?"

    Grounding sets seem to be the lynch pin of this analysis of dictionaries. I find them super interesting, especially this approach to their membership criteria. I wonder if a minimum grounding set could be comprised of just one member, and if that is not indeed the definition of the kernel words in a dictionary. If the kernel words in a dictionary are meant to be those words already grounded in our mental lexicon, and which can be accessed directly through sensory experience, are their smallest grounding set not simply themselves?

    The concept of grounding sets also bring to mind "selective ignoring" as it applies to categorization. I've understood selective ignoring as the capacity to ignore all but one distinctive feature in an individual towards the categorization of that individual based on that feature. Can the same approach be taken in analyzing these grounding sets? Am I correct in assuming that a minimum grounding set would require all of its members to share at least one given feature, and that this feature would be one associated with the kernel member of the set? In this way, would all the members not share one definite feature- what we could call the deterministic feature of the set? If this is the case, then minimum grounding sets could be interpreted as a form of categorization, all based on one deterministic feature.

    ReplyDelete
    Replies
    1. Relations (like bigger) are relative, but it is categorical whether or not X is bigger than Y. (And besides relative judgments, non-categorical cognition includes all continuous motor skills, such as copying and imitation, dancing, swimming, etc.)

      All words (except Funes's) are abstract. Categorization is abstraction.

      The grounding set for a dictionary, or a language, cannot be a single word. It must be possible to define all the rest of the words using only propositions composed of words in the grounding set. Propositions are essentially combinatory. There are no single-wrod propositions. You need at least a subject and predicate, each a category

      The Kernel is the set of words out of which all the other words (inside and outside the Kernel) can be defined once you have removed all (defined) words from the dictionary that define no further words. If you remove any words from the Kernel, however, they may no longer be definable from the rest of the words inside the rest of the Kernel. Yet the Kernel is not the smallest number of words from which all other words can be defined. That is the minimal grounding set (MinSet). But to get the MinSet, you have to drop the requirement that it should be a dictionary: The words in the MinSet can define all the words outside the MinSet, but they cannot define the words within the MinSet.

      The MinSet is not unique. There are many of them in the Kernel. But the MinSet can only ground the rest of the dictionary if the MinSet itself is itself grounded in some other way than via definition! Our hunch is that the MinSet is grounded via sensorimotor category learning.

      There must be at least one feature that distinguishes the members of a category from the nonmembers. But the MinSet is a set of categories -- and a pretty diverse set, since the MinSet cannot define its own members. It can only define the rest of the dictionary. MinSet words name categories, and those categories have features, which may themselves have names. But chances are that the names of many of those categories will not be in the MinSet. And there is no reason to expect that the members of a MinSet will share many features.

      I think you might be making a category error here, Gwyneth, mixing up categories, their features, the names of their features, and the set of categories (and their names) needed to define the names of all other categories.

      Delete
  22. Re: 4.3 Pantomime to proposition: The transition from show to tell
     
    This passage is exceptionally helpful when it comes to breaking down the theory of how language began. We have read numerous different theories and postulates about the origin of language, but I find this one to be the most clear cut and feasible explanation. It seems to me that the development of language came from a combination of many different components that gradually worked themselves together into what we have now. It does make me wonder -- as was touched on in this article -- why other species that are similar to our own have not developed the capacity for language. While this article makes light of a lot of questions that surround the development of language, it is clear that we are a long ways from solving the mystery of how and why language began.

    ReplyDelete
  23. From Show to Tell really grasps the complexity of language, its creation, learning, expression and understanding. It is amazing to see that universal language is inborn because that means that either Cro-Magnon men had this ability and never fully used it, or that they were the ones who developed it and evolutionary psychology may retrace its origins from that moment because of the change in humanity that happened there.
    Language is not only a symbol system where we agreed on the meanings of each symbols, it also has syntax and propositions that helps us integrate and make sense of the symbols, their meaning and their function in our communication. Categorization is also interesting when taken in this context because clearly it existed before language and is inborn. Does that mean that animals and maybe insects have this capacity of categorization since they know what to eat, what not to eat, who is their predator, etc? Is this capacity related to a capacity of language as in humans?

    ReplyDelete
    Replies
    1. Since categorization is doing the right thing with the right kind of thing, plenty of animals can categorize, some innately, some though learning.

      Delete
  24. “So maybe that’s what evolution did with our species. Because we were more social, more cooperative and collaborative, more kin-dependent—and not necessarily because we were that much smarter—some of us discovered the power of acquiring categories by instruction instead of just induction, first passively, by chance, without the help of any genetic predisposition … The tendency to acquire and convey categories by instruction thus grew stronger and stronger in the genomes and brains of the offspring of those who were more motivated and disposed to learn to do so. And that became our species’ “language-biased” brain.”

    This is interesting. I don’t have any arguments to contradict it, though I don’t see why we are more motivated than apes—is it because in our nature we are physically weaker (i.e. without our tools, our organizational capacity and our ‘intelligence’ we would be defeated in combat against any other animal in a matter of seconds)? We are social, but I wouldn’t necessarily argue that we are more social than apes (from the point of view that apes live within ‘communities’ as far as I know); perhaps the fact that we define ourselves as more social comes from the very fact that we have the capacity for language, which allows for a more peaceful (diplomatic) resolution of conflicts rather than purely aggressive/confrontational as is the case within many species; we are capable of having peaceful relations with other “tribes”, whereas other animals might be inclined to attack their enemy rather than resolve the matter peacefully, i.e. survival of the fittest. Therefore, though this is a very interesting point, I think we can question whether we have developed language because we are social animals (and thus have more motivation) or whether the opposite is the case.

    ReplyDelete
    Replies
    1. I have no idea why only our species evolved language. Certainly not because they are not smart enough.

      We're more peaceful than other species? What species has destroyed so many of its own kind, not to mention destroying species other than its own kind, relative to their own numbers and the numbers of their victims and victim species? (Any other species that came close would have wiped out its prey and probably themselves as well: an evolutionarily unstable strategy.)

      Delete
  25. I consider the argument that language is the result of the motivation to social and exchange ideas dubious. Everything can be categorized as a out of coincidence because so far we do not have a better explanation for all the inventions in our societies. The paper maintains that the us and the chimps are not that different intellectually. In fact, the reason we differentiate from each other in terms of language simply due to their lack of motivation to socialize and communicate.
    However, i do not think chimps have less desire to socialize since they have a very well-structured social hierarchy, competition and cooperation between different status and means of communication to coordinate defences and fights. There is a really high level of cooperation happens within their tribes and lots of inventions such as usage of tools. Therefore I do not understand how us humans are more cooperative than chimps, or is this seemingly well adaptive cooperation a result of the advent of language ability?
    Or is it because of the fact that universal grammar is something genetic, of which the chimps do not possess, therefore lacking the ability to construct a syntactic structure that carries propositional meaning?

    ReplyDelete
    Replies
    1. Don't forget kinship and tribalism. They can be the driver for the evolution of a powerful new way to share categories. (UG remains a mystery; it's not needed for making simple propositions. And surely at the advent of language its adaptive value did not require being able to say that "this is the cat that found the rat that ate the (vegan) cheese that.... that Jack built...")

      Delete
  26. "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."

    I find the motivation behind this interesting. Later in the paper you describe a mother's intent to teach her child A B and C categories. Her reason for doing so would most likely be to help her child progress through life and survive. However does this condemn categorical combinations that serve no clear purpose (other than being an exercise in combination), to be lost as it is not adaptive. Could this be an explanation for monkeys not being forced to learn language (because it's not directly adaptive, or the way it can be is not clear to them in the immediate tense).



    ReplyDelete
  27. In Section 3, the difficulty in explaining the evolutionary origins of UG are explained by the fact that unlike other products of evolution like limbs, there is no straightforward evolutionary story. This explanation fits well with discussions of evolutionary psychology in lecture, whereby it was made clear that evolution approximates. This sense of approximation is reflected by the fact that evolutionary explanations of low level processes are possible yet fail to account for developments that arise as a consequence of evolutionarily developed capacities. An example of this is the evolutionarily based repulsion to reproduce with kin, which is based on a proxy of biological relatedness, proximity. Evolutionary explanations of this sort are possible, however processes that arise as a consequence of other developments(with UG being a potential candidate), are not possible since it is unlikely that they were anticipated by evolution.
    Additionally, in Section 4.1, the mushroom example of learning through induction is presented and would have been a perfect opportunity to introduce the critical property of categorization, the consequences associated with doing the wrong thing with the wrong kind of thing or vice-versa. This important component of categorization is not mentioned in the paper though it constitutes an important part of learning through trial and error and subsequently distinguishing members from non-members.

    ReplyDelete
  28. Section 4.3
    I found the discussion of canonical and mirror neurons really illuminated the trouble in understanding why animals with similar capabilities as humans for communication have not been recorded to have the motivation to create propositional language.

    This paper really clearly explains the benefit and adaptive advantage of categorization in terms of survival, and further the way that reducing data driven experiential categorization in favor of instruction driven categorization freed up energy, time, and bodily resources.

    In the discussion in section 5, I found myself wondering about pantomime categories and what combinational, propositional pantomiming would have looked like. Would it look like a version of modern sign language (with less arbitrary motions)? Is the suggestion that these pantomimes were translated into grunts/sounds that became the words we now use (as in kernels/grounding of referent?)?

    I am struggling to understand how this jump was made, especially if the feasibility of protolanguages is in question. Would language have started with a vocabulary for categories paired with gestures? How were function words created? How can we say that language immediately assumed the full power of proposition?

    ReplyDelete
  29. I found the section about protolanguages to be extremely interesting, because it seems impossible for me to imagine a little bit of language. After failing to come up with any probable theories, I started thinking about other species and their communication systems, because it seems like their communication is relatively limited, compared to that of humans. However, I now wonder if their system follows Katz’s definition of a natural language, which is, “a symbol system in which one can express any and every proposition”, where a proposition is “any statement, subject plus predicate, that has a truth value, true or false”. Although their categories are not learned by word of mouth, i.e. instruction, we are not able to truly understand their signals and responses to know their categorical power. We must draw conclusions about them from interpretation because they don’t use our natural language, meaning words, and we can’t possibly dare to assume we know what is going on in their heads due to the other-minds problem. Is it possible that although we have the propositions to describe that we don’t understand them, we don’t have the propositions to explain what they are detecting? For example, if we try to explain the visual performance of a mantis shrimp (that has 12 cones), we can guess by doing experiments, but in reality, we have no sensorimotor experience that gives us any clue to what it is actually like to have 12 cones. In the same way, isn’t it possible that animals can produce any possible proposition, but limited to their interpretation of the world?

    ReplyDelete
  30. “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, with the teacher actively miming the new category descriptions to the learner. The category “names” would also become shorter and faster—less iconic and more arbitrary—once their propositional intent (the “propositional attitude”) was being construed and adopted mutually by both the teacher and the learner.”

    It was very interesting to see how a gesture that described a new category description could be shortened once a shared meaning was established between the teacher and the learner. When the category description has been learned and shorter gestures are adopted to represent it, it means that the gesture has become grounded. It was later described in the paper that this phenomena of needing to describe these categories are unique to humans. We have the motivation to categorize -- which sets us apart from chimps. It’s surprising that the explanation comes from neither intelligence nor sociability, but motivation. However, it still puzzles me as to why humans have this motivation, and where it stems from.

    ReplyDelete
  31. “Chimps have categories. We keep training them to “name” their categories (whether with gestures, symbolic objects, computer keyboards, or words) – even to combine those names into quasi-propositional strings. And the chimps oblige us if they are hungry or if they feel like it. But what is striking is that they never really pick up the linguistic ball and run with it. They just don’t seem to be motivated to do so, even if they sometimes seem to “get it”, locally, for individual cases.”

    I find the argument that chimps lack the motivation to use language unsatisfying. If the reason why we were motivated to name and describe things is because it gave us an adaptive advantage (i.e. avoiding the risks involved with direct sensorimotor induction), then surely language would have given chimps an adaptive advantage too. Additionally, I am not sure that humans are “more social, more cooperative and collaborative, more kin-dependent” than chimps are since chimps live in large communities, thus I do not think that they lack the pro-social dispositions that would make language useful. In short, seeing as multiple experiments on chimps have shown that they are intelligent enough to use language, it is a mystery to me why they have not developed it, and I do not think that arguing that chimps are less social or less motivated to use language provides a satisfying explanation.

    ReplyDelete
  32. The use of “motivation” as an explanation for what chimps appear to be missing in order to have developed language seems to dodge the question. Motivation is vague - akin to an empty explanation like the homunculus. What kind of motivation is needed to predispose a species to learn, use and recombine symbols? The theory of its absence in chimps implies a sort of complacency, as if there was choice involved.

    The way I understood it was that chimps can categorize (aka do the right thing with the right kind of thing) and use induction to apply arbitrary names for those categories. However, they aren’t able to use instruction given that they don’t use proposition. Therefore, while they can combine categories into “quasi propositional strings”, they seem to be lacking an essential capacity that would enable them to get to further steps of language comprehension.

    ReplyDelete

Opening Overview Video of Categorization, Communication and Consciousness

Opening Overview Video of: This should get you to the this year's introductory video (which seems to be just audio):  https://mycourses2...