Blondin Massé et al (2012) Symbol Grounding and the Origin of Language: From Show to Tell. In: Origins of Language. Cognitive Sciences Institute. Université du Québec à Montréal, June 2010.
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.
“So “tell” has superseded “show” in human evolution, but, as with occasional PowerPoints, it still helps to support, illustrate, supplement, or refresh the telling with some showing. After all, direct sensorimotor experience -- rather than just indirect verbal hearsay -- is, at bottom, still what living is all about, even for Homo loquens.”
ReplyDeleteI still don’t get the distinction between “show” vs. “tell”. It was brought up in the first section of the paper, but rereading that didn’t clarify the meaning for me. Professor Harnad talks about the categorization of mushrooms example, both in class and in this paper, where humans are able to learn either through induction or instruction, with the latter being less risky and more efficient of the two. However, instruction would not be effective if the symbols weren’t grounded, meaning that the words in instruction must be otherwise defined by other words, which would, at some point, have to be defined by something other than words, otherwise the definitions repeat in an endless circle of squiggles and squaggles. It is mentioned that one way to do that is to involve our direct sensorimotor experience – would that not be “showing” rather than “telling”? If that were the case, then I don’t think “tell” can ever replace “show” in human evolution, since if we were all about ‘telling’, we’d return to the symbol grounding problem.
On another note, the paper discusses “protolanguages” but I couldn’t seem to find any sort of definition given and I am unfamiliar with this term. Is this a sort of rudimentary language, a language in the first stages (much like the first generation of NSL speakers that came together and among which a sign language arose)? Section 3.2 discusses how it is possible that protolanguages cannot express all propositions, but I can’t seem to find the relevance in observing these protolanguages, probably because I don’t actually know what they are.
Hi Vivian,
DeleteShowing = induction through sensorimotor experience, or through imitation (with the presence of the object being there)
Telling = instruction through a symbol system
In the “simulating the origin of language” toy study by Cangelosi and Harnad they make up three mushroom conditions and two worlds:
1- A, learned directly through trial and error (induction)
Humans water this mushroom
2- B, learned directly through trial and error (induction)
Humans mark the location of this mushroom
3- C, learned through trial and error (induction) OR (instruction) a conjunction by a rule of membership of A and B [Therefore condition 3 has two conditions imbedded within it- which distinguishes the two kinds of creatures: (C1) one’s that learn only through induction, and the other (C2) creatures who learn through induction and instruction]
**Because both creatures are able to learn certain categories (A & B) through induction they both acquire symbol grounding.
“Tell” superseded “show” because: a) tell became more practical and was later adopted as a genetic adaptation and b) mimes and later words in the absence of their object in reference lead to the creation of expressing all possible propositions (subject + predicate + truth value). Furthermore, in “tell”, once someone has acquired the ‘grounding kernel’: the most fundamental set of words (which tend to be circular definitions of themselves-ex: word: dark, definition: not light) in which can produce all the rest of the definitions of other words, they are then capable understanding all the other words through hear-say alone. In this way “tell” has replaced the need to “show”. The quote therefore reminds us that Powerpoints (which show pictures, gifs, shapes, and/or auditory captions), can aid the process of telling because after all, “showing” is how we make sense of and live through the world.
On protolanguage:
At one point the iconic/gestural forms of symbolic instruction are gradually replaced by words because they are less time consuming and more advantageous to use. Blondin Masse et. al. mention that this stage is what people are perhaps attributing as a protolanguage: “a language [that] lexicalizes a new category with a word, or its syntax is modified” (pg.20) (there is no definition as others would explain it, in the paper). However, I believe Blondin Masse et. al. argue that there is no protolanguage. Instead, they believe that the first symbol system was able to express all propositions, and therefore there is no ‘first’ language- it is instead just any natural language [a.k.a. can come in all forms] (pg.6-7).
Protolanguages relate to historical linguistics-- how we can actually trace back the "family trees" of languages and see where they branched off and when and what protolanguage they originated from. There's some really interesting and controversial work to find a single proto-language for all human language, like Greenberg's super-family language classifications: http://www.artsci.wustl.edu/~anthro/articles/archaeo-language.html
DeleteProtolanguage (according to historical linguists) isn't so much a rudimentary language as it is the ancestor language, before groups of people migrate away from the speaking community and develop branches off the protolanguage.
I think in the context of this article, the reason why a protolanguage can't express everything is because it's a symbol system that doesn't require symbols beyond basic categories. The expressive capacity of propositions seems to come in time with subsequent generations of speakers.
The notion that before there was language there was something not-quite-a-language -- a "protolanguage" in which you could say some things but not others -- is Bickerton's (though others have since taken it up.)
DeleteBickerton, D. (2002). From protolanguage to language. The speciation of modern Homo sapiens, 103-120.
It's controversial...
I'm still confused here. Chimps ground symbols, categorize, and sort objects into learned categories. What about that doesn't sound like a proposition?
Delete"The chimpanzees were also taught superordinate symbols for the general categories of food and tool. Once they had learned the task, both chimpanzees were able to sort new tools and foods correctly and to respond to newobjects with the appropriate symbol. The fact that they could sort symbols as well as objects into the appropriate categories suggests that the symbols were associated with representations of the objects themselves."
Gentner, D. (2003). Does Language Help Animals Think? In Language in mind advances in the study of language and thought (p. 250). Cambridge, Mass.: MIT Press.
Is there any difference between "sorting" and "proposing?" If not, is this another instance of old world apes lacking the motivation to do what they're more than able to do?
This comment has been removed by the author.
DeleteThe disposition to purpose: Intelligence or motivation?
ReplyDeleteChimps are able to name categories. “They get the associations and the contingencies; but do they get them as propositions, with truth values? It’s hard to say. What’s most puzzling is why they don’t seem to pick up on the power that is within their hands when they start systematically combining symbols to define new categories. What seems missing is not the intelligence but the motivation, indeed the compulsion, to name and describe”
I find this to be such an interesting idea. It is mentioned that the construction of propositions required a new cognitive capacity; that being motivation. There are two views that are given, either this occurred through a genetic mutation or through Baldwinian evolution (where the text weighs in more heavily on this theoretical explanation). At the beginning of the text it is mentioned that “we won’t be able to tell you when and where [natural language] began happening”, however although the text produces a very satisfactory answer to how and why natural language began amongst humans, I feel as though the why question in terms of motivation is still unanswered. What exactly is the cognitive difference between the motivation of a human and a chimp, and is it something that had to do with the species interaction, or the environment more, or were both of these playing an equal part?
Furthermore, what about in the extreme sci-fi case of T3s actually existing, will motivation have to play a role for their functioning? And is motivation even something we can program? Or would it have to be something robots learn?? Or is the robot at a complete advantage because it is already starting with parts of natural language (alongside with its sensorimotor grounding capacities)?
DeleteI found this very interesting as well. I still feel a bit unsure of whether I understand the role of motivation in the development of language abilities. According to the paper, the chimps have the ability to understand and learn category names. Chimps differ from humans in that they cannot attach truth-values to these categories. It seems to be these truth-values that give categories their power, however, I do not understand this concretely. Being able to point to a chair and identify it as “chair” seems to imply a truth-value. It is true that this object is a chair. If chimps are able to do that, what are they missing? The drive to communicate that to others? Couldn’t it be the case that human categories are irrelevant to the survival of chimps, and therefore they are not motivated to name and describe? I would be curious to know more about the communication of chimps and how it is connected to their survival. After all, isn’t survival the ultimate motivation in question?
The idea of motivation as the drive to develop a more and more complex language also interests me. I feel like what you might be touching on, Lila, is that the purpose that language serves might differ among species? Perhaps what we as humans choose to communicate is different from what great apes communicate. An example I could think of is how we have developed language that communicates the meaning of symbols beyond what is necessary for our survival. In this sense, maybe we are motivated by more than survival - or in another way, a more diverse and broad range of factors motivate our survival and thus require language.
DeleteI was intrigued by the question of motivation as well. There seems to be two ways of looking at this. The first way is using the mushroom example, if the mushrooms were poisonous, then they would be linked to our survival and then in this case, humans would be greatly motivated to learn the different categories. And, like the artificial life stimulation of creatures shows, it’s easier, more efficient and much faster to learn the categories by instruction instead of induction. In situations of survival/life and death, instruction becomes extremely useful. Language could have started in these types of situations and then expanded to more generalized and less dangerous situations. For monkeys, there is no motivation to learn that a chair is a chair, like in your example Lila. But if monkeys were challenged by situations of life and death, motivation would not be the main problem. Would monkeys then be able to learn a language?
DeleteThis brings me to the second way of looking at this that is not linked to motivation Harnad quotes: “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.” According to this quote, it seems to me that languages in humans didn’t develop because of motivation but because we are more social animals. To me, this might also be why monkeys didn’t develop language. Although monkeys are very social animals, they might not be AS social, cooperative, collaborative and kin-dependent as humans were and are. So it might not be a question of motivation after all.
Demi: Without getting into the "hard problem" (of how and why it feels like something to be motivated), motivation need only be a disposition to do something -- like the child's disposition to eat sweets. If the "propositional attitude" was really a big adaptive advantage, then selection pressure would make it stronger, by favoring those who have the genes that dispose them to make (and seek) propositions.
DeleteLila: It is propositions (subject-predicate descriptions) that have truth values, not names. "Chair" does not have a truth value (but "This is a chair" does). (Lots of human categories are not only irrelevant but inaccessible to chimps. because they are based on recombining category names to describe or define new categories, and chimps only have categories, not propositions.)
Stephanie: Communication differs between species, but not language, because ours is the only species that has language.
Catherine: It's still not clear what there was in our original environment that made propositions so useful for us, but not for our other ape cousins. I don't quite see why the difference would be greater sociality: didn't language make us more social?
I was surprised that Harnard avoided defining language in itself; cognition seem to struggle more with operational definition issues than other fields of psychology. On the other hand, given how complex language is (including all the disagreements within theorists and scientists), it is understandable that reaching a universally-agreed definition is a tedious task. I think that the Prof avoided the definition issue in a clever way, by keeping a broad scope including all language-dependent structures.
ReplyDeleteThe Prof expand on something that he alluded in the previous class which surprised me; the evidence that actual language did not emerged with vocal communication, but rather with bodily gestures. It is true that we do share some similar body postures with non-human primates. For instance, we tend to expand ourselves to display dominance, and to curl back and lower the head to display submissiveness (or shyness). Interestingly enough, athletes with congenital blindness do the exact same gesture when they win (throw both arms in the air), despite the fact that they've never visualize someone else doing it! Thus, although I was surprised at first by the Prof's proposition, it does make sense that we have innate body gestures that convey meaning (and that isn't culturally-dependent).
Moreover, I liked the Prof's suggestion that vocal language became increasingly an adaptive advantage from freeing the hands. After all, our hands are much more proficient in terms of grabbing and doing an infinite number of actions than non-human primates. And thus language would partly explain why human's hands have changes so much in comparison to our closest primate cousin
"Prof's suggestion that vocal language became increasingly an adaptive advantage from freeing the hands"
DeleteIsn't that the yo-he-ho theory? Not of language development per se, but of spoken language?
Hi Alex,
DeleteYes, I meant for spoken language. In the sense that transition from gesture to vocal acted as a catalyst for evolution towards spoken language (implying that natural language was already present; it simply shifted to a more adaptive modality).
Language is the capacity to express every possible proposition. (And I think gestures led to language via mime, not via innate gestures.)
Delete“We conjecture that language began when attempts to communicate through miming became conventionalized into arbitrary sequences of shared, increasingly arbitrary category names that made it possible for members of our species to transmit new categories to one another by defining and describing them via propositions (instructions).” (1)
ReplyDelete“Our version of the very same criterion as Katz’s Glossability Thesis was the “translatability” thesis: Anything you can say in any natural language can also be said in any other natural language--though it’s important to add, not necessarily said in the same number of words. [...] For those readers who have doubts, and think there is something one can say...” (6)
Whether you know the meaning of a word has nothing to do with your ability to define it; you just need to know when to use it. If you mess up, you’ll raise eyebrows, make people laugh or run straight into trouble. As such, knowing a word is very much an act of categorization (actually, more like judgment): is this time for this word to shine? What this means is that the meaning of a word, phrase or expression is very much tied to its history of usage in a community.
The only way for these “glossability” and “translatability” theses to be plausible at all is if we, indeed, gloss over all the nuances of and everything that is interesting about, not only language, but human experience at large. The best example I have is swear words. There is no word in French that is as offensive as the English C-word. The word itself has a translation, even a vulgar one, but it would not elicit the same kind of reaction, and therefore it would not be used in the same contexts… and therefore it doesn’t mean the same thing.
Categorization is indeed one of the things we do with language, but it is not the only thing, nor is it necessarily the most important one. So what else? Coordinating action, determining who’s in-group and who is not, group cohesion, etc. The idea that language is a kind of grooming has been scoffed at in class, but it’s not crazy. Being part of a social group, feeling part of a group, knowing you can trust on others, these were arguably more important to our ancestors than knowing which mushroom to pick. Actually, in this paradigm of adaptive success over others, why on Earth should I believe so and so that these mushrooms are edible?
Children play games before they are let loose in the real world. This allows them a low-stake space to explore the possibilities of their bodies and the possibilities that emerge spontaneously from the interaction with other bodies (e.g. climbing up higher trees, having something remote thrown in your direction). Sometimes they play games that adults taught them, or they imitate adults. But more often then not (well, maybe less so now...) they come up with their own games. These games need not have explicit rules. The rules are enacted even if no one knows them. It’s only if we stop and look at what’s happening that we pick out some patterns, some local rules with varying levels of flexibility.
This picture has the advantage of accounting for the trust needed to start this process: if I have a history of playing some language game with you, then I know how you play, so I know what to expect. If I don’t have such a history, then I’ll have a chat with you, and see how you roll, see if you’re cool, like me. Then, you can hunt in my team.
It is also much more compatible with our daily use of language which has little to do with categorization and everything to do with maintaining social ties and/or status.
Swearing, Euphemisms, and Linguistic Relativity. (Bowers, 2011)
Deletehttp://journals.plos.org/plosone/article?id=10.1371/journal.pone.0022341
Poulin,
DeleteI dug this post!
"if I have a history of playing some language game with you, then I know how you play, so I know what to expect."
What comes first, the trust? The game? Or the language?
You're talking about mind-reading. This, Pinker and Bloom suggested, is what language is for. But the ability to propose differs from mind-reading. I often find that the propositions I make get in the way of my mind-reading.
Nevertheless, language in its fullest form, I think, allows us to make propositions. We can play, we can hunt, we can trust, all without making propositions. Once we have a need for dependenable propositions, the rest flows.
I like your post. But I think that the whole social aspect of language is in fact encompassed by a definition of "categorization"; namely, words have their direct definition (e.g. the literal referent of the C-word) as well as extended uses (e.g. calling someone the C-word) and emotional connotations. All of these can vary from person to person or from locality to locality, but the process of learning each of these "slots" of a word's definition is a process of categorization in itself. Now, what you do WITH the categorization can be social - and, given that we are social creatures, will almost certainly be social - but the ability to "play the game" has everything to do with one's ability to pick up on the subtle distinctions between, say, the use of the C-word and the word "vagina". Glossability/Transatability can't translate emotions, but neither can the exact same sentence, uttered by different people in different contexts or communities or to different people - or spoken in languages - elicit the same emotion. So that isn't the main goal of the glossability/translatability theses, which just leaves us with categorization.
DeleteRosenthal, thanks for the article. I don't know if we have the same interpretation of mind-reading, but just to be sure, I was not really talking about it in the sense of trying to access the interlocutor's "real" mental state. In the premise I was setting, there is no reading per se, but merely a reaction. In other words, I don't get information, but I get a feeling. It is true that language can be used to make propositions but like you said, it's likely to be a later development. The ability to make propositions with truth values requires a shared worldview, a shared belief system. If you treat the forest as a collection of trees, animals and bodies of water, you will make different propositions than if you treat the forest as a spirit that provides food and water but that can get angry and kill you.
DeleteSztein, my contention is that what you call “direct definition” or literal definition is a later development of language. In order to be able to talk about the so-called “real world”, the “objective world”, or the “world out there”, we need an enormous amount of coordination which is what games, some of them linguistic, have allowed us to do. So it may be true that we can now break down meaning into figurative and literal, but prior to a shared conception of the world (the cutting up of experience along roughly the same joints, into comparable chunks) it does not make sense to talk about the literal nor about categorization. If what we want to do is tell a story about the evolution of language, we cannot do that by assuming as a given what it allowed us to achieve, namely, a shared world.
I find Kevan's idea of language as a mechanism to build trust very interesting, though I can't help but to agree with Dia in that the social aspect of language could be partly derivative of categorization. Additionally, it's unclear whether language is necessary for the building of a common world view. Maybe language is used just because it is convenient to express thoughts with words, but the shared worldview that lies underneath these words may be caused by the shared environment, as well as non-linguistic communications. You could say that within a language, the language-wide denotation of a word is one form of categorizing arbitrary symbols, while the local connotation of a word is a way of categorizing people's reactions to those symbols. In the later case, you are just correlating a stimulus to a response, so it could conceivably be a non-linguistic process that does it
DeleteThe distinction between the denotation and connotation of a word is very important, and it seems to me that the translatability thesis only requires the connotation of a word to remain intact between languages. Curse words were a good example for you to choose because there is a huge disconnect between the denotation and connotations of them, seeing how most of them don't *really* mean anything, they just express an emotion that is relatively context dependent.
So yes, you could say that due to the stigma against saying the C-word (in north america, australians have no problem using it all over the place), there is no word with the equivalent connotation in french. However, the translatability thesis states that languages need not have word-to-word translations, so it's completely plausible that you could achieve the same result as calling someone a C___ by stringing together some amount of expletives that are equally offensive together.
Keven, do you really believe that the nuclear power to say any possible literal proposition came from games? What game? How? Why? (And I'd say the figurative use of language came after the literal use, not before. And swear-words are not even coded in the same part of the brain as language. Aphasics can still swear (and sing)...)
Delete“We conjecture that language began when attempts to communicate through miming became conventionalized into arbitrary sequences of shared, increasingly arbitrary category names that made it possible for members of our species to transmit new categories to one another by defining and describing them via propositions (instruction).”
ReplyDeleteThe authors of this piece tackle the following question: what is the origin of language?
To get to their answer, we first need to talk about categories. Categorization is doing the right thing with the right type of thing. For example, not eating a poisonous mushroom is categorization. Humans can either learn categories through direct experience (ex: tasting a bit of a poisonous mushroom, getting sick, and learning not to eat blue mushrooms again because they are poisonous) or by instruction (ex: your mum tells you not to eat blue mushrooms because they are poisonous).
The authors posit that being able to learn categories via instruction gives an individual an evolutionary advantage (higher chance of survival + thus reproduction!). They show this using an artificial life simulation, but without getting into the details of the experiment, it just kind of makes sense: if you don’t have to taste mushrooms to learn which ones are poisonous, you are at a lower risk of getting sick and dying! Also, being told the rule for how to categorize something could be much faster than figuring it out via trial and error.
Harnad et al posit that, initially, kin taught learners new categories via overt actions. For example, a mother would teach a daughter which mushrooms were safe to eat by eating safe mushrooms when the daughter was around – the daughter would learn by example.
(I get a bit confused here because the authors say that “you can’t convey new categories via pantomime alone.” Presumably, this means you can’t convey new categories via overt actions alone either. But I think the authors place this sort of “passive learning” in the instruction category, not the direct experience category. Maybe they mean that you can’t directly convey new categories via pantomime and overt actions, but a learner can passively acquire them through these methods).
However, mushrooms aren’t always easy to find. So humans started to draw categories to the attention of learning using pantomime instead of overt actions (this is kind of tough to show with my mushroom example, but it works for other categories, such as showing that teeth go in the category of “must be cleaned” – we can imagine tooth brushing being pantomimed!).
So gestures have an advantage over overt actions for learning new categories. But natural language has an advantage over gestures. As Harnad et al write, “But once a species picked up the linguistic ball in the gestural modality and began to run with it (linguistically, propositionally), the advantages of freeing its hands to carry something other than the ball, so to speak (rather than having to gesticulate while trying to do everything else)—along with the advantages of naming when the teacher was out of the learner’s line of sight, or at a distance, or in the dark—would quickly supersede the advantages of gesture as the start-up modality for language.” Basically: it is easier to convey categories using language than gestures because you can do it in the dark, or if the other person is too far away to see your gestures, or when you are carrying things (or fighting a predator!).
“We suggest that this is how the proposition was born,” write the authors. And since they define language as a “a symbol system in which one can express any and every proposition,” this is their story about the origin of language.
Two questions / reflections I am left with
Delete1. According to this account, “was language invented”? Is language a cultural creation that we have passed down through generations? Or is it tied to our genes and passed down by those who had the genetic precursors for language (and thus had a better chance of survival?) I am not sure that what the authors of this piece think – they seem to kind of skirt around the question. On the one hand, their hypothesis is based on an experiment showing that individuals who can acquire categories via instruction have a better chance of survival. On the other hand, they don’t get anywhere near the genetic basis of language. Could Harnad et al’s account still hold if language was a cultural creation? I think so.
2. Remember that sentence that I struggled with earlier on? “[Y]ou can’t convey new categories via pantomime alone”? Well this suggests that, while we may have moved from pantomime to natural language so our hands could be free, so we could teach categories in the dark etc, the really significant advantage of natural language over gesture has nothing to do with those things but rather with the fact that we can convey new categories using propositions. Propositions, per the authors, are “category inclusion statement[s].” (the cat is on the mat means the cat is in the category of things that are one the mat).
First, regarding your main comment, I think the difference between pantomime and overt action, is that with pantomime, you don’t have the object on which the overt action would be performed, while with an overt action the object on which the action is performed is there, for example: (imagine there is no vocal language yet) is easier for a mother to show their kid that they must not touch the power outlet if the power outlet is there, but it would be much harder for a mother to pantomime a power outlet and transmit to the kid that he must not touch it in the case he sees one. I think they mean that you can directly convey new categories via overt actions, both when there is the intention of transmitting the category and when an individual is passively learning, but with pantomime this task becomes much harder. As you pointed out, pantomiming a mushroom (just as pantomiming a power outlet) would be really hard, which is why “you can’t convey new categories via pantomime alone” but this is possible in the case of overt actions. I agree with you that the significant advantage of natural language is that we can convey new categories using propositions.
DeleteHarnad et al. imply that propositions are much more efficient and less time consuming (“…began to profit from the considerable advantages it conferred—the benefits of acquiring new categories without the time, energy, uncertainty, and risk of acquiring them through direct sensorimotor induction…. to predicate and propositionalize about further categories, because of the adaptive benefits that category description and sharing conferred. 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.” (18) and “Besides, once the name has become arbitrary, its “shape”—and hence its sensory modality—no longer matters. 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).” (20)) It is more efficient and less time consuming for a mother to tell their kid: “do not touch the object with three holes that is on the wall, it is a power outlet and you can get electrocuted if you touch it” than pantomiming this whole sentence.
Regarding your first question, it seems to me that the authors agree that language is as you said: “tied to our genes and passed down.” They point out that “language itself began earlier than vocal language (though not earlier than vocal communication obviously). Once it had made its power and adaptive benefits felt, however language simply migrated (behaviorally, neutrally, and genetically) to the vocal modality, with all of its obvious advantages.” (3, emphasis added). Here, their interest is on the origins of language and how “it came about,” which is why the genetic basis is not discussed (Harnad mentions in the video that the “cultural part” is not controversial and that these aspects are trivial). I think they simply assume language became engrained in our genes (some of this is alluded to in the passages I quoted) and that is why we are so good at acquiring language, which is why I don’t believe language could have become what it is, had it been a cultural creation, Universal Grammar would not hold and it would take much more time to acquire language if we were not predisposed to it genetically.
As someone who's taken my fair share of classes in linguistics and psychology, this article was refreshing because we rarely study the origins of language. I was most interested by the topic of the kernel of language, and its relationship to a test for language ("to say anything and everything precisely"). Just as a thought experiment, I thought about what it would be like if someone were to wipe all but 500 words from my vocabulary. While the article claims it would be sufficient, I am certain that in context I would be at a huge disadvantage. I could with difficulty communicate everything I needed to say, but would become extremely frustrated, and the process of communication would be significantly hindered. You could certainly claim that I was using a language, but I definitely wouldn't be taking advantage of its full expressive power as we know it to be. I'm of the opinion that, in order to get closer to a more comprehensive definition and test for language, we must include some requirements about ease of communication with others who are using the current version of that language with a richer lexicon.
ReplyDeleteIn addition, even if we can supposedly say everything we need to say in 500 words, where does that leave the tens of thousands of other words in the English language? Are they mere peripheral linguistic accessories? Don't they deserve some credit for the richness and nuance they provide our language? I'm willing to admit that certain words - or even, the majority of words - are not absolutely essential to language, but I would argue that they are essential for modern communication, and that they are adaptive too – having an expanded vocabulary simply makes you sound smarter and may confer respect and opportunities that you may not receive with only a 500-word vocabulary.
I’ve thought about this for a few days, and I realized the response to this line of thinking is that there is a difference between the evolution of a particular language, like English, and the evolution of our language ability. Evidently, the former progresses much faster than the latter, which made up most of the content of the actual article, so my mistake was in confusing the two. So expanding your vocabulary is adaptive but can be achieved within one lifetime; it is much more difficult to conceive of how our ability to use language will continue to evolve on a geological scale.
The other tens of thousands of words in the English language are not peripheral linguistic accessories. They are essential to language to save time as with a limited amount of words you would need to use the combinations of at least two words to refer to a thing whose name is not present in your 500-words vocabulary. Having an expanded vocabulary can have been an advantage for survival.
Delete"I'm of the opinion that, in order to get closer to a more comprehensive definition and test for language, we must include some requirements about ease of communication with others who are using the current version of that language with a richer lexicon."
DeleteWhat interests me is how you said "ease of language". I agree that a richer vocabulary makes it easier to communicate my emotions and thoughts. But 'ease' is maybe the wrong word. I think that each word, though it can be explained or defined with other words, is nuanced and carries meaning that is unique. In this way, a richer vocabulary allows us to be more precise in expressing ourselves.
I think that you touch on a very interesting point, and it was something that resounded with me too. The additional words in the English language must serve some purpose, even if they are not the core component. I think that Marion’s idea that the evolution of extra words could have an evolutionary advantage is very insightful. To elaborate on this, I think about the warnings we received from our English teachers when using a thesaurus. Even though two words are listed as synonyms, they are not identical—they are used in different contexts, have different connotations, and thus convey completely different information when they are used.
DeleteI’ve already commented on the limitation of language when discussing the Whorf Hypothesis. For all of it’s power, certain ideas are difficult to teach using words alone (love, or at least romantic love, comes to mind). In these cases, people proclaim that you need to experience it to know it. The additional words in our vocabulary could be an attempt to reduce uncertainty between these types of ideas and the words we use, with the overall goal of reducing the number of these vague ideas in total. If enough people can understand the “new” word as symbolizing the idea that is not quite defined by core words, then there is less uncertainty. This clarify offered by extra words becomes especially important with the advent of writing and literacy, since there is no opportunity to re-explain oneself when someone is reading a writing text far away from you.
I agree with Marion on this. When I first read your comment, Esther, I also thought that the immense number of words in the English language would help with efficiently referring to things. In addition, this also made me think of the mutual exclusivity principle in language acquisition. This principle states that when seeing a new object and hearing a new word, the immediate response is to place that label on the whole object in a 1 to 1 mapping scheme with the object and the label. In other words, it seems to me that it is natural for humans to make new labels for new objects in their environment instead of compounding words to describe the object. That being said, the labels of the language would be reflective of the surrounding environment as the environment provides the objects to be labelled.
DeleteI find the concept of the kernel and the minimal grounding set (MGS) fascinating as well, and I am inclined to believe that having only 500 content words is indeed sufficient to define anything in the world that can be defined. However, I doubt that anyone would ever want to talk to a person with a vocabulary of merely 500 words (excluding function words, predicates/propositions), and it's not only because this person always takes forever to describe and define any content word outside of the MGS. Like all of you have pointed out already, the other tens of thousands of words each -- or at least the majority of them -- carries their own unique meanings. They are not just used as perfect substitutions for longer, more tedious definitions (with MGS). These words are nuanced and carry information about categorization, beyond their definitions in a dictionary, shared among users of a particular language.
DeleteSince the paper brought up the definition of "fruit", I'd like to use the distinction between a fruit and a vegetable as an example. Botanically speaking, "fruit" is a subset of "vegetable", so all fruits are vegetables but not all vegetables are fruits. Culinarily speaking, however, fruits and vegetables are mutually exclusive sets of edible food. This works fine when we are trying to categorize apples and spinaches, but what about things that are botanically categorized as fruits but commonly/culinarily categorized only as vegetables? I don't want to get into another debate about tomatoes on a late weekday night, so…what about eggplants? They are botanically considered fruits, but people rarely call them that because we don't eat eggplants like other fruits; we cook them like vegetables. Now try explain all of this without using the words "fruit" and "vegetable". It's hard enough to discuss this with words (such as botanical/culinary/edible) that are definitely not a part of MGS, therefore it must be extremely difficult -- although not impossible -- to make anybody understand it when you are only allowed a vocabulary of 500 words.
I would argue that we have evolved to have a vocabulary far bigger than the MGS not only for the brevity of communication, but more importantly, for the expanded lexicon that allows us to express even the subtlest differences between categories.
The mushroom forager simulation suggests that creatures who can learn by instruction out-reproduce and out-survive those who learn by induction only, despite the fact that when a category is learned by instruction it takes longer to apply the instruction to each instance encountered than when the same category is learned by induction. This paper goes on to put forth the theory that the cooperative societies discussed at length in the Pinker and Bloom article would motivate group members who already know categories to try to teach them to learners. The authors suggest that proposition was born after group members started to attempt to draw category combinations to the attention of learners by miming the actions that you would perform on such a category in the absence of the actual objects. I understand that it would be advantageous for human ancestors to be able to instruct each other about combinatorial categories in the absence of the objects themselves because this increases the number of situations in which one could learn a category by instruction (and we already know it is likely evolutionarily advantageous to be able to learn categories by instruction). I agree that is logical for human ancestors to have started communicating propositions gesturally but eventually shifted to communicating them vocally so they could free their hands to do other things. However, I’m a little unclear as to at what point miming combinatorial categories in the absence of objects transitions from just miming an action to a symbolic ‘name’ for a category.
ReplyDelete“We conjecture that language began when attempts to communicate through miming became conventionalized into arbitrary sequences of 2 shared, increasingly arbitrary category names that made it possible for members of our species to transmit new categories to one another by defining and describing them via propositions.”
Does the miming just gradually become more and more abstract until its ‘shape’ no longer mimics its meaning? How exactly would mimed symbols have been systematically translated into vocalizations to free the hands? Would human ancestors have tried to teach each other ALL categories INCLUDING the ones that can only be learned by induction or would they have known to limit their miming instructions to categories that can be learned by instruction?
Miming probably has to become more abstract because you need to save time when communicating. Therefore, you need to use gestures that are shorter and simpler. Moreover, abstraction is intrinsic to mime because someone miming an action needs to select its most salient features.
Delete“However, I’m a little unclear as to at what point miming combinatorial categories in the absence of objects transitions from just miming an action to a symbolic ‘name’ for a category”
DeleteI agree with Marion. It seems to me that the advantage of miming a name for a category instead of miming the symbols that when combined would describe that category is the time saved when communicating. In addition, having a single name for a category reduces the memory load compared to a long description of that category. For example, to describe a table, one would only need the single label “table”. This makes it easier on the memory because one would only have to search for an iconic figure of a single category in their memory. Having a long string of words that when combined describe the table forces one to first retrieve all the different, relevant components and then combine those components.
“Does the miming just gradually become more and more abstract until its ‘shape’ no longer mimics its meaning? How exactly would mimed symbols have been systematically translated into vocalizations to free the hands?”
I’m not sure if a direct translation from mimed symbols into vocalizations is the only way to (or the most appropriate way) to describe the emergence of vocalizations for communication. Is it not possible that the words we use to describe things are independent of the physical gestures that we would use to mime them? Obviously there is no way to keep the “shape” of a category in a word when one vocalizes it (with the exception of onomatopoeias). To be completely honest, I have no idea why we use “this” arbitrary word to label a category instead of “that” one. Some words do have some reasoning as to why they are used as the label for a category. For example, the words to label a maternal figure (“mom” in English or some variant of “ma” in many other languages) have many similarities cross-linguistically. This may be because when a baby begins to produce speech sounds, this combination of syllables and vowels is one of the easiest to produce. However, why “table” is used to label the category of what we know as tables is still a mystery to me.
I like your discussion of mimed symbols and vocalizations and I'm inclined to agree that the words we'd use to describe things tend to be separate from the physical gestures we'd use to mime them. I get that it is adaptive to free your hands and have one symbol like "table" to designate something rather than a series of them. I'm just stuck on the question of how we got from gestures to vocalizations.If it wasn't direct translation, what other steps might have been involved?
Delete“Because we are 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” (Massé, Harnad, Picard, & St-Louis, 2010).
ReplyDeleteTo begin, I want to mention that I really liked the ideas proposed in this paper by Massé and colleagues, particularly because of the integration of categorization in the development of language. Also, I thought it was important that the individual’s motivation to learn, acquire, and share categories was what may be linked to the transition between language that was based on “show” to language that was based on “tell.” The “tell” language conferred a huge advantage over “show” language, as it allowed the acquisition of new categories without the time, energy, uncertainty, and risk of obtaining them through simply sensorimotor induction. The motivation, which is the key factor in the development of “tell” language, allowed for language to develop in such a way that was useful and incredibly beneficial to humans. Scientists often attribute the development of language to the increase in brain size among humans in comparison to their predecessors, but motivation seems to be a more plausible explanation. Brain size doesn’t really tell us much anyways, as it not even linked to intelligence. The efficiency of the brain’s energy metabolism and regulation (the efficiency of supporting glial cells and the amount of sulci and gyri) seems to be a better indicator of intelligence. The transmission of new categories by instruction (instead of sensorimotor induction) is a very likely and believable theory as to how language developed in humans as we know it today, especially because of its numerous adaptive advantages mentioned in Massé and colleagues’ paper.
If I understand correctly, one argument is that the intentional nature (or propositional attitude, caused by our innate sociability) of actively teaching another about a new category (which would then become a proposition and eventually a content word in verbal language) is a possible definition of communication itself. And without that intent, or motivation, we would be stuck 'producing a string of associations for instrumental purposes' like our unmotivated chimp friends. Words then range from the minimal grounding set, to the core, to the grounding kernel, and then to the rest (the more abstract conceptual words that do not explain other words).
ReplyDeleteThe fact that within the simulation the instruction learners (those that learned through observation and imitation) had out survived the induction learners (those that simply learned by trial and error, or sensorimotor induction) provide an interesting theory behind the adaptive usefulness, and thus the origin, of language and complex communication, which seems simply be convenience picked up by our motivated ancestors.
The theory that instruction learners out-survive induction learners is really interesting. The fact that symbolic instruction can be learned through observation and imitation is crucial--this does not involve direct language. But it also clues to the necessity of imitation to language; not only are we imitating the actions of others, we are also imitating their symbols. Could this theory explain why modern day humans outlived the Neanderthals?
DeleteI do have a few questions about the reading:
ReplyDelete"Instruction requires far fewer training trials (in fact, a good instruction should allow correct categorization from the very first attempt), but it does take longer to apply the instruction to each instance encountered, at least initially; the reaction time for categorization when it has been 24 learned by induction is faster than when it has been learned by instruction."
What causes the difference between the passive observation of another's behaviour in learning a new conjunction of categories and learning the conjunction when another is actively teaching it to you? Is there an inherent difference in the processing of the same information between these two categories? I'm finding this hard to grasp without the details of the underlying biology.
"It could well be that our ancestors had the power of communication by pantomime before the advent of language; but you can’t convey new categories by pantomime alone. All you can do is mime objects, events, and actions, and, to a certain extent, properties and states (including requests and threats)."
I do not understand why not? Is this statement limited only to composite categories, concepts that cannot be understood in the physical world?
"Chimps have categories" (…) "They get the associations and the contingencies; but do they get them as propositions, with truth values? It is hard to say. What’s most puzzling is why they don’t seem to pick up on the power that is within their hands."
According to this statement categories can also be presented as associations, which we can observe in chimps, I assume, when they behave in ways that show their understanding of consequence (ex: if I pull this lever, I get a treat.) However I am confused. How does this automatically count as combining symbols to define new categories? Is the new category simply the existence of a new association? And how do these associations become propositions? "A proposition is any statement, subject plus predicate, that has a truth value, true or false." Is the only thing that's missing propositional attitude?
I am beginning to believe that there must be an infinite number of categories that can never be completely discrete. Our 'categorical decisions' (meaning, I believe, any accurate recognition which serves a purpose cognitively or behaviourally) might not necessarily be a product of 'categories', but a continuous stream of consciousness containing countless comparisons and influences that cannot be understood separately. There exist too many physical objects and possible concepts which we experience in our everyday life for categories to be finite and discrete.
"What causes the difference between the passive observation of another's behaviour in learning a new conjunction of categories and learning the conjunction when another is actively teaching it to you? Is there an inherent difference in the processing of the same information between these two categories? "
DeleteIt is probably a problem of direct vs. indirect grounding. In the two cases, you are not processing the same information because you don’t get the same kind of input: one is visual and you are directly perceiving the thing that is categorized; the other input can be heard (listening) or seen (reading) but it is not the thing that is categorized, but a description of it. Things you have directly grounded (even by simply watching someone else interacting with them) are faster to recognize. Whereas, because indirect grounding through instruction is never as precise as direct grounding by induction, it takes more time to match category you have learned by instruction with a thing you encounter for the first time. You can think here of when in class, we compared the efficiency of showing a picture to recognize someone vs. the efficiency of simply describing the person. Plus, how can you even be sure that when you are ‘instructing’ a new category to someone, this person gets exactly the same category.
I agree with Marion that there is sometimes some uncertainty in the way things can be described. If someone tells me to go find a particular can of soup in a pantry by describing it, there is always uncertainty of whether or not you are looking at the right can, but if someone shows you where it is, you know exactly which can is the right one. Elaborating on just the rationalization of why reaction times are faster for induction compared to instruction, I think it has a lot to do with memory encoding. As with remembering past events, they are often encoded more strongly when you have a strong autobiographical tie to that particular instance (you have better memory for things that occur to you, personally, than what someone tells you). I would assume that learning by induction vs. instruction works in a similar way.
DeleteAs with the other paper, I really have nothing to say.
ReplyDeleteThis paper demonstrates all of the difficulties that I have/had experienced while studying in this semester at McGill. What is very intriguing to me is the timing of the paper. There is a question that is remaining regarding the timing and the spacing and the languag-ing. When/where/who/what/when/how are all of the important questions, and they are important due to the fact we have (at somepoint) begun to find the necessary. The ideaology that goes into categorization is difficult to wholly understand yet it is impossible not to understand it once the concept has been mastered. There are many mysteries of cognitive science that remain unseen, we must continue in the current pursuit.
In the very beginning of the abstract, Harnad posits that “Most species can learn categories (1) by direct experience (“induction”). Only human beings can learn categories (2) by word of mouth (“instruction”).” This brings to mind several animal studies that I’ve seen where animals are taught to understand english words and commands, specifically commands such as ‘pick up the blue stick’ or ‘press the large button’. Obviously the animals had to have been taught what the colour red is, what large it, what a button is, and what a stick is. However, I wonder: does this count as categorizing? Have these animals not been taught categories by word of mouth (“instruction”)? I’m not sure whether or not this qualifies as genuine categorization, because the primates obviously don’t have a fluent understanding of english, but I do think this counts as instruction, or teaching by word of mouth. This then leads me to question where we draw the line at what intrinsically counts as categorizing and as instruction, and where we make the distinction here between humans and non-humans.
ReplyDeleteWhile non-human animals can in animal studies be taught commands by instruction, I think the complexity that Harnad is trying to get across is that non-human species do not have the motivation to adopt language as a social learning strategy. He does point out that “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 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.”
DeleteIn other words, a chimp can be taught a category through induction and even respond to instruction when they feel like it. The animal will respond to an instruction to pick up a blue stick by either picking up that stick, picking up the wrong stick or not picking anything up at all. But, the chimp will not respond to that instruction by signing back to you that he has no desire to pick up the stick. Furthermore, the next generation of chimp will not learn the category blue by instruction but from induction and may be taught once again that this category is named blue.
"Anything you can say in any natural language can also be said in any
ReplyDeleteother 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. 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—what that something is, and
why one cannot say it. You will find that you have generated your own counterexample."
To me, this passage says that although different languages possess different symbols, the meaning is the same. Differences in wording, precision and nuance are therefore irrelevant because the thing that the words represent are the same. I just have so much trouble with this. How can we explain the evolution of more and more words to describe 'the same thing', if not to refine categories, and to become more and more precise.
"Despite its dramatic benefits, however, learning categories through language, by symbolic instruction, is not completely equivalent to learning them by sensorimotor induction"
I interpret this passage as saying that symbols will never be fully sufficient in conveying understanding, because understanding is more than categorization. Understanding involves a 'what it's like-ness' that can only be experienced through the senses.
Stephanie, I too struggle to accept that anything you can say in one natural language you can automatically say in another natural language. While the argument of finding a counterexample and then giving it in English (countering your counterexample) is amusing and cheeky, I wonder if there is some nuance that is lost here. Often individuals who are fluent in multiple languages fall back on lone words from other languages to describe something. Yes, this word can be defined in English, however these individuals feel that the English translation does not get to the root or feeling of the word. It does not take into account all the contextual, and emotional aspects of the word—in other words, it does not actually define EVERYTHING attached to that symbol. I would argue that these contextual and emotional aspects are part of the thing that is being symbolized, and not just social or cultural components (maybe not in all cases, but certainly in some). In a way, this relates to Stephanie’s second paragraph about the benefits of sensorimotor induction. Reading a definition off a page does not match the experience of being learning the definition of a word through sensorimotor induction, because by using other words to define a word some parts may be lost, which is what occurs when we attempt to translate certain unique words between languages.
DeleteThis article was very well written and brought up some interesting ideas. In terms of learning how to categorize objects, the distinction between induction and instruction was very clear. However, the article lacked an explanation for one main component of language in trying to determine the origin of it, this being the phonology of language. Phonology is a fundamental aspect of vocal communication and does not have “any arbitrarily shaped symbols” that determine how to categorize words based on phonological elements. I found it a little unclear how we are able to learn to categorize the sounds for words based on instruction or induction when people vary so widely in their speech e.g the difference between a male and female voice, difference in terms of tone, accents etc. This was touched upon when Harnad made reference to categorizing proper names to individuals and that the category must be learned so that we are able to recognize the same person in all instances as being a member of the same category i.e. his proper name. When there is no shape or symbol to make reference to, which is the case with phonological sounds, I think it would be near impossible to learn how to categorize properly. It seems as though we would need direct experience with every instance of the sound first in order to do this sufficiently and then would require to remember them all – and yet we do not need this extensive amount of learning. I feel as though when there is a physical manifestation of the shape, such as a printed word, it is much easier to categorize because we can repeatedly look back at it and its symbol remains relatively constant – but when it is just a spoken sound it seems much harder as the stimulus is much more varying and instantaneous.
ReplyDeleteI am no expert in phonetics, but I think that there is enough similarity in the compression air patterns i.e. the wave properties of the sounds, so that words/letters that are emitted when speaking can be categorized without hearing all the variants. An analogy of this could be morse code: whether you're using an official morse code instrument or taping your fingers against a table, it's still morse code, there's something fundamental in the compressed air that remains distinguishable. Or like a piece of music that's being played with different instruments... you can still categorize that piece of music.
DeleteI think I understood your comment!
Danielle,
DeleteI think you raise a really interesting point regarding sounds. Maybe one way to look at them, in terms of this article, would be to think of the acoustic properties as being the physical manifestation of the symbol. Sounds do have shapes! And our ears are designed to categorically perceive the shapes of these sound waves. A sound with features for place, manner, and voicing is still a phoneme no matter the speaker, and our auditory system can discriminate those features regardless of who is saying them.
“The Kernel is unique, but it is not the smallest number of words from which all the rest can be reached by definition alone: That would be the minimal grounding set (MGS), which is not unique, and we are just beginning to extract it in real dictionaries.”
ReplyDeleteThis is an especially interesting aspect to consider, especially because Harnad goes on to say that only about 500 content words make up the smallest set. Cutting a language down to 500 basic operating items shows just how generative and powerful our language is, considering that tens of thousands of words are in the English language alone through some combination of the smallest set.
I also strongly agree with the point that language developed long before vocal communication; non-verbal language is just as valid as verbal language and I find it is intuitive to think that the verbal modality was found to be a beneficial means to convey language whereas there had once been only gestures.
In the paper Symbol Grounding and the Origin of Language, From Show to Tell, Harnad helps a lot with some ideas discussed in class. I think there are two major points in this paper: the origin of language and the symbol grounding problem, and I can agree mostly with Harnad regarding to the origin of language, but I still can hardly see a solution to the symbol grounding problem.
ReplyDeleteAccording to Harnad, we have many cognitive categorizing capacities other than language before language was born, and it seems that language "thieve" the capacity of categorizing, and that by evolution, we gradually use language to categorize. I really appreciate where Harnad kind of suggests language learning and language evolution are two different things. He says categorizing is not inborn; we learn to categorize after we are born. However, on the other hand, the process of language development is related to evolution; some of us, human beings, discovered the ability, and from generation to generation, we take it for granted. I could not agree more with this point of view: we are not born with the ability to categorize everything, but I think we are born with the ability to learn to categorize. It is still a little unclear here. What I mean is that for some things, we cannot categorize them because we do not know them; however, we can categorize them after a learning process. For example, a child looks at a chimp and says it is a bear, and this is an example of not born with the ability to categorize everything because otherwise he or she would say it is a bear. But later after the child is told that it is a chimp, he or she would say it is a chimp because he or she has the ability to learn to categorize and eventually he or she successfully categorize this furry, black and brownish, giant creature into chimps.
However, although I could understand and agree with the argument about the origin of language, it does not seem to explain the symbol grounding problem. According to Harnad, if we try to reduce the amount of words in dictionaries to the extent that we are able to explain every word in world, there are still hundreds of them called kernel or core words. This also seems very true to me because for every word, if we want to explain it in vocal language, we have to use some other words. However, if language was born after other cognitive categorizing capacities and it thieved the capacity of categorizing, does it mean if we really want to solve the symbol grounding problem, we will have to look at other cognitive capacities in order to explain the kernel or core words? Will this be a "real solution" to the symbol grounding problem?
This paper was a really enjoyable read. Aside from resonating with my own biases, the style was fun. So, let’s see if I could expand or disagree with certain claims.
ReplyDelete“What is certainly true is that, when we manipulate natural language symbols (in other words, words) in order to say all the things we mean to say, our symbol manipulations are governed by a further constraint, over and above the syntax that determines whether or not they are well-formed: that further constraint is the “shape” of the words’ meanings.” (p.10)
Out of curiosity, does ‘the “shape” of the words’ meanings’ imply that semantics is really a special sort of syntax? I remember Dennett describing brains as syntactic engines that approximate a semantic engine. I guess I’m asking whether this would be a “Harnad says”? To me, this seems like the only way to approach the problem of meaning.
“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.” (p.11)
“The” might not be a category in itself, but I’d argue that even though most linguists see “the” or “a” or “if” as grammatical devices, for sake of understanding language acquisition, function words still have content which can be exposed by their shaping of new categories when applied to words and phrases which are already imbued with content and are thus already categories. “The X” (as opposed to “a X”) implies there’s only one X such as in the case of “the 3rd planet from the Sun” or, when in physical presence of an X, “The X” refers to the specific X which is around such as in the case of “the cat is hungry” when noticing a cat meowing, or when in the linguistic context of an aforementioned X, “the X” refers to the X previously mentioned. If our task in cognitive science is to explain, understand and possibly build cognition, I feel like drawing analogies between seemingly distinct phenomena in need of explanation (in this case content and function words) is always a worthy exercise.
“They just don’t seem to be motivated to do so, even if they sometimes seem to “get it,” locally, for individual cases.” (p.17)
I’m presently taking Psyc 318 (Behavioural Neuroscience 2) which focuses on motivation and reinforcement. I’ve been wondering of late whether language acquisition and language comprehension and production are really the same process at different stages. Essentially, I’m wondering whether categorizing the world and labeling it is part of a sort of human drive state like hunger.
That’s pure conjecture and here’s another one. Is part of the reinforcement that seems to be lacking in language acquisition (The Poverty of Stimulus) actually manifested internally through expectations coming to fruition or not and calibrated through the dopaminergic prediction error system? When one hears “a phenomenon” a couple of times and has learned the plural rule of adding “s” and then one hears “those phenome…”, is the brain expecting “...nons” and when it doesn’t and hears “na”, does the dopamine signal die and thus inhibits the neural connection between “phenome…” and “...nons” which was initially wired together?
Is there any concrete evidence connecting the dopaminergic prediction error system and language acquisition?
For your first question, I don't think Prof. Harnad means semantics here as equal to the shape of the words' meaning (I might be wrong). Semantics is just the meaning, regardless of the shape of the words. So you always have two different things: syntax and semantics. You can swap out any of them, for example, English syntax with Chinese syntax, while preserving the same semantics (i.e. UG).
DeleteThe ability to determine well-formness, however, has to do with both semantics and syntax, that is, shape of the words' meanings (symbol grounding). Semantics is not needed for a symbol system, after all, symbol system is just a system with a bunch of arbitrary symbols (shapes). But without semantics, we cannot tell which proposition/statement within the symbol system is well formed. So to have well-formness in a symbol, you need two things: 1. Semantics 2. Grounding symbols onto the semantics. For example: How can I know "The cat is" is not a well-formed sentence? If I don't know English, I wouldn't know (symbol not grounded). If I don't
understand meanings at all (Searle's CRA), I wouldn't know (no semantics).
It is in assessing how organisms evolved the capabilities to detect “the members of [category] C … by simply combining the feature detectors of [categories] A and B” that concerns me when considering the evolutionary production of language.
ReplyDeleteHarnad, in his paper “Symbol Grounding and the Origin of Language: From Show to Tell”, outlines a general mechanism by which instructive combining of symbols, whose semantics have been induced by means of trial and error learning, into propositional language acted as a highly adaptive survival technique for early “Homo” cognisors. Harnad notes that “the capacity to construe a proposition with a truth value, may either have been a mutation, or primates may already have had the capacity and only needed to have it amplified motivationally and behaviorally” in order for the full power of language to take effect. In my opinion, the mechanism seems highly plausible, but ultimately DOES rely on natural selection. If, for example, the capacity to make a proposition via instructive categorization did result from a mutation, then it would be the addition of observational learning through genetic variation which occurred at some point in history. A true explanation, it seems, requires a deep functional understanding of learning mechanisms in our ancestors, and how a mutation might result in a new form of learning.
On the other hand, I do not quite understand why natural selection would not automatically put pressure on primates capable of propositional capacities. Why is extra motivation required? Yes, it seems as if this might accelerate the adaptive pressure, but it seems as though natural selection would still run its course on a primate who utilized instructive categorization, even if he did so every once and a while. Wouldn’t an unmotivated propositioning primate still be selected for more effectively compared to one categorising by inductive methods alone?
Harnad argues that language is simply a type of symbol manipulation. But the symbols are not just manipulated according to their shape but also by the shape of their meaning. I believe the source of my confusion lies here. If I have understood correctly, the shape of the symbols in language (words) is their part of speech. This intuitively makes sense, as it would mean that we put words together in ways that are grammatical. If the latter is accurate, then would “shape of their meaning” imply that we understand words and then organize them based on their meaning?
ReplyDeleteIf so, I don’t understand the difference between manipulating symbols based on the shape of their meaning and just by their meaning.
While learning a new language, we first learn how to manipulate symbols based on their shape (i.e. leaning the rules of grammar…) and then as our vocabulary grows, does our ability to manipulate symbols based on their meaning increase?
When someone teaches us a new language, we learn what new words are; that is, we learn what they mean in the language we have already acquired. This is completely different from our first-language acquisition process because we are learning what objects are for the first time. But when learning a new language, symbol grounding seems to be two fold: We understand what a foreign word corresponds to in our first language and we connect this first language to something in the world. My question then becomes, how much induction is necessary before language comes into effect in the form of instruction?
I have previously encountered the term "protolanguage" in use by historical linguists—where it seems to have different connotations from a cognitive scientist's definition of the same word—and am intrigued by the clear delineations between the two as well as where they might converge. In studying cognition, and language as a mechanism thereof: "Now, with this test for whether or not something is a language—a system in which you can say anything that can be said—let us immediately consider one of the notions that is repeatedly invoked in this volume: “protolanguages.” Remember that the theme of the Summer Institute was the origin of language. If a natural language is a symbol system in which you can express any and every proposition, what about a protolanguage? If you cannot express all propositions in a protolanguage, which ones can you not express, and why not?" categorization - some easier, harder?"
ReplyDeleteDubious examination of this term with regards to language as an organ of cognition is perfectly understandable, given the supposed validity of any system that allows expression of any and every proposition as a language, per the above and "Alternatively, if one can express all propositions in a protolanguage, then why call it “protolanguage” at all? For then the question we are addressing at this institute would simply become: What is the origin of protolanguage? Because, as they say, it can’t be turtles (or tadpoles) all the way down.". In fact modern linguistics also professes no language to be superior or more optimally developed than another, although what their definition of "language" is concerned with multiple more specific, concrete substrates than cognition's. Thus their "protolanguage" would address the predecessors of the definitive forms of groups of languages, most often via theoretical constructions of potential symbols of such a language given sound change, environment, civilisation, etc.: all, in cognitive science, still perfectly valid forms of language. I would like to agree that "protolanguage" in a similar sense is ill-suited for the investigation into cognition. However, it would seem both cognitive science and linguistics are interested in identifying the "core" (or "kernel", even) of what gave rise to language (by definition of either discipline) as we know it today; it would have had to arise from something, and I myself am instinctively reluctant to say that something would be as fully formed as the systems we know today (in terms of either propositional capability or specific lexicons and prescriptive grammars). The latter is less perplexing than the former—most certainly needing to be preceded by it—but it would seem the problematic usage of "protolanguage" stems from both confusion about when to determine a certain system to have less propositional strength, as well as whether it is more grounding categories or something else that is needed to bolster it.
"We have only begun computing the smallest defining set (because the general problem of computing it is NP-complete), but we do know that that smallest set, unlike the kernel, will not be unique. There will be many alternative sets of the smallest size. It looks, however, as if each of them will consist of about 500 words."
ReplyDeleteThis is perhaps a side note, I'm a bit curious to know more about how the smallest defining set is computed. Many NP-complete problems can be solved "well enough" by using approximation algorithms (in this case, finding an "almost" optimal set), randomization (shorter running time, with a small probability of failure), leveraging the structure of the kernel, or by using a heuristic that works reasonably well.
Aside from that, I find it a bit odd to think of language as reducible to 500 words. Others have raised this issue from the perspective of it being difficult to express oneself without more words, but I'm a bit more concerned about the effect of grounding terms that do not belong to the smallest defining set.
Clearly, humans ground more than 500 words over the course of their lifetime. My question is: how much does this grounding affect the meaning of the words? For example, one cannot learn to dance from instruction alone, even if all the words used are grounded. Even if a dancer has mastered all of the movements individually, they will learn new intricacies of the motions when they physically dance the motions strung together in a complex movement. Even though the movements themselves are dynamical, I would argue the way the dancers categorize the motions will change somewhat as they learn to master a new combination. Furthermore, when attempting to convey what they learn from the experience, words are often not enough and some combination of instruction with reinforcement learning is used to convey meaning to other dancers. A dancer might know what a "jete", "pas de bourre", "chasse" and "glissade" are, but the meaning of these words will change when they learn (via direct experience) to do a "chasse pas de bourre glissade jete" in a way that cannot be adequately described except through direct experience. Does this mean that everything a dancer can only learn from experience must be given a term that is added to the smallest defining set? Or do we say that the naive approximation lent to the dancers by their grounding of individual terms are a "good enough" approximation to the overall motion?
What I'm trying to illustrate here is that the meanings of words change with experience, even if they can be roughly approximated by other words prior to those experiences. If we reduce the entire dictionary down to a minimal set of grounded words, can all the other words actually be built from this set, or will the built words simply be rough approximations of the words as we know them? If cold is not in the smallest set, but hot is, does it suffice to define cold as "not hot" or "the opposite of hot" or will this simply be a crude approximation? If a person has never felt cold in their life, but they have felt hot, will they be able to actually understand what coldness is based off of one of these definitions, or would they only be able to parrot back the definition without any true understanding of the term?
Although I find the question fascinating from a computational standpoint, from the perspective of cognitive science it does seem a bit artificial to me to search for the smallest set of grounded words in a language. As far as I can tell, humans do not operate based off of any such set -- they operate based off of many words that are continually being grounded in many different ways. Why not aim for a "typically grounded" set instead? By "typically grounded" I mean a set of words in which the majority of humans have some direct grounding associated with the terms. To me, this would seem like a more useful set of words for cognitive scientists to study, simply because that's the grounded set of words that people cognize with more often than not.
Delete