Harnad, S. (2005) To Cognize is to Categorize: Cognition is Categorization, in Lefebvre, C. and Cohen, H., Eds. Handbook of Categorization. Elsevier.
We organisms are sensorimotor systems. The things in the world come in contact with our sensory surfaces, and we interact with them based on what that sensorimotor contact “affords”. All of our categories consist in ways we behave differently toward different kinds of things -- things we do or don’t eat, mate-with, or flee-from, or the things that we describe, through our language, as prime numbers, affordances, absolute discriminables, or truths. That is all that cognition is for, and about.
This idea makes sense when discussing the abstraction of categories with large amounts of sensory data (i.e: physical things) but less so when discussing how to abstractly describe categories of concepts. As Dr. Harnad pointed out, concepts like ‘goodness’, ‘morality’ etc. don’t leave sensory traces. And so, the question becomes: what exactly do we abstract from to create those categories? Are we drawing on the memories we associate with those kinds of actions and collating the commonalities between the actions to develop a concept or is it something deeper?
ReplyDeleteHere's a mouthful (and a test of your understanding of the course so far): Consider the expressive power of combining the names of grounded categories to define or describe new categories via words (e.g., the "peekaboo unicorn") through subject/predicate propositions with truth values. This is what the symbol grounding problem and the "minimal grounding set" are about. And remember the power of computation (the weak and strong Church-Turing Thesis), the thesis that language can express any possible proposition, and that although a picture is worth more than a thousand words, with more words, the verbal description can always catch up, as closely as needed to resolve any remaining uncertainty.
DeleteIt seems to me that the truth values are an important part of abstract categorizations. From my understanding of the reading and course content, it seems like we would be abstracting from either trial-and-error learning (listening to the vocabulary and scenarios described when the word morality is used) or through 'hearsay' (a well informed, trusted individual explaining complex concepts through more elementary concepts that you have encountered in life). Because more ephemeral seeming concepts don't have physical properties to encounter with our sensorimotor system, the power of language would be the only way that the realization of these concepts could occur and be theorized about in the first place.
DeleteIn the end, aren't all properties (features) of categories physical? And aren't all categories abstract?
DeleteTo be able to categorize (do the right thing with the right kind of thing) we have to be able to distinguish the properties of the category members from the non-members. That's what "abstracting" means: to pick out, ignoring all the other properties.
"Red' is a property (all red things have it), so is "round" (all round things have it). Our senses and measuring instruments can pick all that up. "Roundness" is more abstract, but it's just the property that all round things share.
So what about "truth" and "justice": aren't they the properties (possible multiple Boolean (and, not, or, if-then) properties) that true/false statements and just/unjust actions share (depending on your taste in theories of justice)?
Abstract? Yes. It's all abstract. The abstraction is just a matter of degree, But what do you mean "ephemeral"? (The dictionary will tell you, and some things meet that definition, others do not...)
Verbal definitions/descriptions are the way still more abstract categories can be defined/described (and grounded) through Boolean recombinations of already grounded categories.
But it's all physical, and sense-able, although way down to the ground.
”But of course our sensorimotor systems do not give equal weight to all features; they do not even detect all features. And among the features they do detect, some (such as shape and color) are more salient than others (such as spatial position and number of feathers)”
ReplyDeleteDuring the last class, we discussed the Great Eskimo Vocabulary Hoax in which Whorf stated that “Eskimos” are able to discriminate the differences between snow better than English speakers because they have > 400 words for snow. This argument was refuted, as it was proven that their language is rich in derivational and inflectional morphemes that seemed to create more words for snow.
I wonder if Inuktitut speakers are still are able to detect differences in different types of snow better than English speakers due to their different sensorimotor experiences that result from their environment. In this case, wouldn’t their sensorimotor systems weight the features of types of snow more than someone who is not exposed to different types of snow? Their difference in perception would not be due to their language, but their way of categorizing snow.
“Let us assume that if organisms can categorize, then there must be a sensorimotor basis for that skill of theirs, and its source must be either evolution, learning, or both.”
I think that in this case, the source for this skill would be learning.
This seems like it would be the case for any sort of expertise. If I got into a competition with a bird watcher over who could identify the most songbirds, they would beat me by a long shot even in the same language. They've been exposed to more examples and know what to look for.
DeleteWhat would be interesting to see is if, for example, you took a bird watching course with someone who is extremely experiences in bird watching and see if you could compare to the individual who has a lot more experience than you but learned to identify the songbirds via trial and error. The example used in the article shows that individuals who learn the Geon features can identify the sexe of the chicken and perform as well as an expert without having the trial and error experience of an expert.
DeleteYes, "difficult" (highly inter-confusable) categories (including snow, and wines, and cancer cells) might produce "weak" Whorfian effects ("Stevan Says"). That's why I'm still a Whorfian despite all the failures.
DeleteBut the three things to compare are learning cayegories (1) by mere exposure (no correction or feedback), (2) by trial and error with corrective feedback, and (3) by verbal description. (3) is the nuclear option, and only our species seems to have it.
I'm curious as to how these innate vs. learned categorization effects correlate with brain development. Take the FFA debate: is that area innately attuned to categorizing faces as faces, or do we just learn to preferentially categorize faces through increased exposure as our brains develop?
DeleteI found Berlin and Kaye’s Anti-Whorfian findings on the innateness of colour perception extremely interesting. They showed that most cultures divide the spectrum in the same way and that those who don’t still view the spectrum with the same regions of compression and expansion nonetheless. I am then wondering if those who do not name/divide the spectrum in the traditional ROYGBV manner have additional regions of compression/expression in accordance to their own culture’s labels (in addition to the innate ROYGBV regions of categorical expansion/compression)? I.e. are there still Whorfian effects that shape categorical perception?
DeleteAbout the face area: there is probably some plasticity too, because dogs have one too -- for people.
DeleteIf there had been "strong" Whorfian effects, the rainbow would differ across languages and cultures. It doesn't. But there are "weak" Whorfian effects of learning sub-categories, like scarlet/crimson.
Delete"That’s why the only way to learn them is through the months or years of trial and error reinforcement training guided by feedback under the supervision of masters."
After reading this article, it seems to me that there are 3 ways in which the human species learns categorization. 1. by trial and error with feedback, 2. by exposure, and 3. by verbal 'hearsay'. However, I wonder how much of learning categorization, specifically the ability to abstract in our everyday lives, comes from trial and error and how much from direct feedback, and if there is a cut-off in terms of development that these experiences are needed to happen? In other words, we know that children who are deprived of language are only able to recover and achieve the same levels as children who are not deprived up until a certain age. Does this inability to learn language at the same level as other non-deprived children play a role in other aspects of their lives later on. Would this affect their ability to categorize if they lacked the ability to verbally express their sensorimotor experiences, or to fully comprehend the verbal feedback they are receiving? If they experienced more instances of trial and error to make up for the lack of processing via verbal feedback, would they be able to advance to the same point eventually?
This reading was so helpful. It summarized and explained a lot of categorization-related concepts we had already touched on in class discussions. Still, some questions arose while reading it. One of my interrogations concerns the innate/learned debate in color CP.
ReplyDelete“Although basic color CP is inborn rather than a result of learning, it still meets our definition of categorization because the real-time trial-and-error process that "shaped" CP through error-corrective feedback from adaptive consequences was Darwinian evolution.”
Does this passage imply that categorization of colors is undeniably innate? Because if it is the case, I would argue that although we do have a specific sensorimotor system to perceive different hues of colors via different wavelength inputs, I do not believe the categorization of it is completely innate. My comment probably falls close to the Whorf hypothesis. To perceive the color “forest green” and to verbalize it as “forest green” (in your mind or aloud) are two different processes. My argument is that when you categorize this stimulus as “forest green” and verbalize it as such, you are relying on a learned category. On the other hand, when your eyes perceive the colored stimulus, the biological process of it is obviously innate.
I don't think the claim is that Forest Green, is an innate category, just the basic colors. Forest Green seems like it would be a sub-categorization, requiring the category green to already be understood before it's meaningful. The fact that we even call it forest green means that we recognize that it's more similar to our notion of green than it is to blue or red. These divisions of the category are probably not innate, and not every is going to be able to recognize them. I probably couldn't accurately point to forest green if i tried. But you don't have to be trained on the basic colors to recognize them as belonging to different categories. Anyone who has functioning color vision is going to perceive red as a separate thing from blue, just because that's the way our visual system works.
DeleteThe basic colors are innate; the sub-categories like crimson and scarlet might generate some weak Whorfian CP effects when we learn them.
Delete"[learned CP] works like a kind of input filter, siphoning out the categories on the basis of their invariance features, and ignoring or reducing the salience of non-invariant features."
ReplyDeleteThis paper really solidified my understanding of the (combinatorial) power of language, but this quote was the most engaging for me. Thinking about the mechanisms we use for selective attention (in dichotic listening for example), this explanation seems extremely parsimonious and like a very important key in understand the neural mechanisms behind categories. Because categories are such an important part of cognition and communication and acting on the world - this seems like an exciting avenue towards the hard problem.
Careful... just saying there are invariants does not provide the mechanism for learning to detect them. (But "deep learning" nets are the best current model.)
DeleteI thoroughly enjoyed this reading. Each new concept built effectively on the previous one, making the idea of categorization very clear. What I found specifically interesting was the Whorf Hypothesis, which states that how objects look to us depends on how we sort and name them. This concept of linguistic relativity really structured my understanding of how language affects our every day lives; often we don't think about it consciously, but in hindsight language is the driving force behind most of our experiences. Without language we would not be able to effectively form categories verbally to ourselves or to others. The idea that categorization also forms the basis of cognition offers further insight into understanding what cognition means. That being said, the hard problem still persists, as we still don't have a clear mechanism for 'how' we form categories and, thus, 'how' cognition works.
ReplyDeleteBut remember that none of the strong Whorfian effects (like color) turned out to be learned, hence they were not language effects. There might nevertheless be weak Whorfian effects, maybe even based on the effects of Boolean verbal descriptions of categories, rather than just the naming of categories.
DeleteWhere does Categorization end? In the reading it makes the distinction between sensory discrimination and categorization, but what about recognizing that an object does not belong to a category in isolation. Let's say you stumble upon a poisonous mushroom in the forest that looks wildly different from the edible ones you know of, but there's no edible ones around to directly compare it to. If you don't eat the mushroom, not because you recognize that it's poisonous, but because it is not a member of your current category of edible mushrooms and you're being conservative, is this categorization? Is the act of recognizing that an individual is not a member of a category in itself categorization, so the category would have the invariant "Not a member of X", or is it just a consequence of having already defined the category "Edible Mushrooms"?
ReplyDeleteLearning categories requires both positive examples (members) and negative ones (non-members), plus feedback about which is which. If you have not yet completely learned a category, you will make two kinds of mistakes: false positives (thinking it's a member when it's not) and false negatives (thinking it's not a member, when it is). That's what the corrective feedback is for...
DeleteThroughout the course I’ve always come back to the same question: how does abstraction fit into cognition?
ReplyDeleteFinally I’ve gotten my answer: categorization.
Categorization is not possible without abstraction. Abstraction allows us to single out a specific part of sensory input and selectively ignore the rest, which is an integral part of the process of learning categories.
I think that if we weren’t able to categorize we would be overwhelmed by each moment to the point where our way of living would be completely unrecognizable. Every new moment would be a barrage of sensation and it would be impossible to process it all. Thus, it seems that abstraction is responsible for our ability to not get lost in the potential sensory overload that each moment holds.
Yes, all categorization is abstraction. And William James also thought there would be "blooming, buzzing confusion" without categorization.
DeleteAfter reading the supervised learning section, I think I better understand how this relates to categorization. I think it makes more sense to view categorization as a learned ability as opposed to an innate skill because it seems as though the error corrections and miscategorizations contribute to our learning and capability to categorize in the first place.
ReplyDelete"An unsupervised learning mechanism could easily sort out their retinal shadows on the basis of their intrinsic structure alone". But this is not the case, some inputs require more than the shape of the shadow to allow us to categorize item, it is the external supervised input and life experience that allows us to be able to categorize in these instances, supporting the supervised learning hypothesis.
But this brings me to wonder about the mechanisms at play when patients who have experiences traumatic brain injury and can no longer categorize. Does this suggest that the injury has destroyed the "categorization" area of the brain in which they can no longer perform the skill (innate hypothesis), or are there other, more complicated, mechanisms at play.
I agree with you Rachael with how supervised learning is more of an explanation for categorization rather than an innate skill. Surely, experiences shape the way one categorizes. Trial and error based on sensorimotor experiences is what shapes how we categorize things which would help explain the variability of categorization between individuals.
DeleteExpanding on your point on a categorization area in the brain, I think this would be interesting to investigate. I wonder if neuroplasticity would influence this "categorization" area with our different perceivable experiences.
"Everything in nature is a dynamical system, of course, but some things
ReplyDeleteare not o n l y dynamical systems, and categorization refers to a special
kind of dynamical system." Is this akin to the square-rectangle phenomenon, where all squares are rectangles, but not all rectangles are squares? That is, there are purely dynamical systems and of course there are dynamical and categorical systems, but are there purely categorical systems? Would numbers count as one?
I think the answer lies in the part of the text you quoted. If categorization refers to a special kind of dynamical system, then dynamical systems subsume categorical systems (it’s like the square-rectangle phenomenon in that rectangles subsume squares).
DeleteI don’t think you can treat numbers as purely categorical (entirely distinct from dynamical systems) because they ultimately rely on some kind of sensorimotor interaction with the world. Obviously, as adults, having acquired the category of “two”, we don’t literally have to picture two of something to understand what is meant by the symbol “2”. But to have acquired the notion of “two” in the first place involves having seen, heard, touched, etc, various amounts of things, guessing at whether those amounts corresponded to two, and being given corrective feedback.
If we use Professor Harnad’s definition of categorization (“doing the right thing with the right kind of thing”), then some kind of sensorimotor interaction is always going to be involved. Even for the most abstract of categorizes, furthest removed from real life, you still have to talk about them to get corrective feedback. And to speak is to use categories acquired through sensorimotor interaction with the world.
Prof Harnad argues that categorical perception isn't innate-and if it is, it is at most so in a Darwinian way, by process of elimination. I find this very plausible: I don't believe humans are born with the intuition to decipher edible from non-edible foods, as toddlers will try to ingest just about anything. When fully-developed humans perceive the smell of a home-cooked meal, they may relate this feeling directly to their stomach by being conditioned to for many years. We don't smell an industrial work site and become hungry (unless it is a smell that has often been associated with food for a significant amount of time, you never know). Following the article, and various class discussions, I don't believe that we innately categorize things. Many animals certainly do, but as a matter of fact, humans are born with underdeveloped brains and bodies-which is why we need our parents in order to survive our first years, unlike other species in which the newborns are nearly independent from birth.
ReplyDeleteDoes categorization subsume feeling? Or, I wonder what the broader connection there is between categorization and feeling. Is feeling a special case or effect of categorizing? At the very least, I think they may be related (possibly relevant for explaining the weak Whorfian effect?). We do categorize feelings in a sense - we feel them as different kinds of feelings. And some can learn to better categorize feelings (Is that what master meditators do?). I’m not too sure where this thinking leads but it would imply a few things: feeling is not exclusively a passive thing that happens to us & that we could be “mistaken” in our feelings/experiences of the world as we could with other categorizations we make and could therefore correct/change our feelings. But that seems a tad strange to say..
ReplyDeleteI was wondering about this while I was reading the article too. In class, Harnad posed the question to us of “What cognitive function ISN’T categorization?” Answers presented included relative judgements that don’t have discrete categories, and continuous motor skills. At this point I asked “What about feelings?” and I wish I wrote down his answer because all I remember is that he waved it away and said something along the lines that although he’ll give me that, but it’s not in the direction of the conversation. I still don’t know where feelings come in either.
DeleteYou said that we can learn to categorize our feelings. Although true, if I look at the definition of “doing the right thing with the right kind of thing”, when I experience sadness, although I can label it, there’s more to my experience of that sadness than just recognizing it as such. If categories are learned, and you can’t just passively learn enough in isolation without supervision, and if feelings are just categories, then it would follow that a feral child or the isolated person on the island would not have feelings… which is so not true. My question is: has he revised his position in the years since writing this paper, or am I missing something? In my notes I have the phrase “to say cognition is categorization is an overstatement” and it seems like there should be more to cognition that categorization. Feelings, for instance.
I find this discussion really interesting. I think the language of feelings, i.e. the labelling, is definitely categorization. For example every occurrence of sadness is very different but we clump all sad occurrences into the sadness category. I think this is simply our way of talking about our feelings, whether to ourselves or to other people.
DeleteHowever I would beg to argue that our categorizations are moderated by our feelings. Since the process of categorizing is done through experience, experience is moderated by feeling, hence categorization is moderated by feeling. I don't mean that categories change based on your feelings, i.e. a child that is scared of dogs isn't going to have a different category for dogs. But they way the child interacts with this category is going to change. Or for more complicated things like grouping categories together, ones feelings about a category is going to be effected. I think this may have some effect for categorizations of bigotry.
Could someone explain the "vanishing intersections" problem, please? I'm not sure what it entails. Looking for the invariance, shared by all the sensory shadows, between some content words, we would find that's there no invariance --the "intersection" is empty. In the section, we learn that Fodor and others have suggested this to counter a learning or evolving categories. But what evidence to they have of this? Isn't it generally accepted that dictionaries are circular, implying that there is invariance between a lot of words.
ReplyDeleteAdditionally, the "vanishing intersections" doesn't refute innately learned categories, either. It seems to be suggesting that categories don't and can't exist at all.
DeleteThe vanishing intersection problem applies to categories which do not seem to be based on shared features, because the shared
Deletefeatures aren’t apparent, or don’t appear to exist. If you pick some content words and look for their invariance there is none (they have nothing in common). Fodor used vanishing intersections to explain why children cannot learn language by association alone since examples of language often have nothing in common. Perhaps I am incorrect, but I don’t believe that vanishing intersections refute innately learned categories. In fact, it seems to strengthen the evidence for them because children consistently form the same categories for which there seems to be no unifying features. As for your question of what evidence Fodor has, I am unsure (besides his example about boomerangs, tables, and chicken abdomens). I did a bit of outside research but only found that same unclear example.
Ah, thanks so much!
DeleteRE: Learned Categories
ReplyDeleteWhen someone looks at a photograph of someone and asks us who they are, most of the time we say “that’s Jim” (for example) and not “that’s a photograph of Jim.” Even though we know that the picture is not actually the person, we still lump it under the same category. With this in mind, it is clear that we are capable of learning that two different things can go into the same category if there is some resemblance. My question is why are differences (in this case the fact that the “thing” in question is not actually the real thing) are forgotten in favour of one category over another? Does it have to do with environment or learned importance?
I think it has to do with a hierarchy of affordances. In the ugly duckling section, Harnad talks about how our "visual system "weights" certain features more heavily than others". Probably because there is more to gain in recognizing that the ducklings are yellow and the swanlet is grey, than there is to know that spatially the swanlet and one of the ducklings are equal distance apart from their sibling, thus belonging in the same category. With the example of a chair, it’s a lot more important for you to know that it belongs to the category of things-you-can-sit-on than things-that-are-bigger-than-bread-boxes. And I believe this hierarchy is flexible depending on what you need to do in an environment. If you’re in a WWE ring and you’re presented with a chair… your first response won’t be to think of it as a thing you can sit on.
DeleteHarnad goes on to say that if we did weigh every category equally, we'd run into a similar issue to Fumes, in that every item would have infinitely many features that we hold to be equal, so everything becomes infinitely similar. By that story, one of the most important features of our memory is the fact that we forget/selectively ignore things. Shape and colour are examples of features that we are just predisposed to pay attention/give more weight to, because they generally give a lot more indication to the kind of thing we’re looking at than spacial proximity (and because we have innate colour feature detectors, vs. other features we have to learn).
In your example, it wouldn’t be conducive to the conversation for you to insist that the photograph you’re looking at is ONLY a photo of Jim, and that you can only use it as an aid to talk about photographs of Jim, but not Jim himself. It would make for a really insufferable conversation to be honest.
RE: Abstraction and Hearsay
ReplyDeleteThe jump from abstraction to ascribing names to things is completed with the intermediary step of recognition. Abstracting means to attend to some part of the stimuli while ignoring the rest of it, but what is the mechanism behind this? Abstraction is selective, but in order to be selective we need to have knowledge of both that which we are selecting for and that which we are not selecting for. After abstraction is complete (a comparative process in nature) recognition happens, which is an active step because it is based on experience. After we have recognized the stimulus, we can name it. This is a seemingly inclusive list of processes that could happen in a neat, linear fashion. But what processes or intermediary steps underly our ability to abstract? What is causing categorization?
This paper made me think about the act of mindfulness, or meditation. It's not an action that is innate, most people are unable to meditate more than a minute without thorough practice. You are learning by taking in all these sensory inputs (smells, sounds, touch, etc) and letting them "flow over you", kind of like how Funes does with his experiences. Like a master chicken sexer, it takes years of thorough practice from meditation masters to reach a state of pure mindfulness. So I guess in a way, chicken sexing can be compared to meditation?
ReplyDeleteI'm unsure if I think meditation is a total abstraction of everything or an act of concretizing, or as you describe it as "mere passive experiential flow from one infinitely unique instant to the next" that Funes does.
I find that this article shed some light on a question I had from last week’s reading on symbol grounding. I was confused as to how something that we cannot act upon with our senses could ever be given meaning which is what the symbol grounding problem is all about. In this article, the author explains that a way that we can give meaning to such things is through linguistic hearsay, people can just tell you about the properties of things without you having to engage with them directly. This saves time certainly but my question about linguistic hearsay is how did the person who is telling you about a category come to know about it? Isn’t it possible that they got it through hearsay too and perhaps the person before did so too and so on? Although explicitly telling people about categories certainly allows us to learn faster and can be very adaptive I can’t help but think that over time this transmitted message could become very flawed and deviate little by little from the original description of the category such that individuals’ categorization for the same object might be very different from one another. Just think about what happens when we play the telephone game….
ReplyDeleteI just wanted to comment on the idea you (Maria Cianci) mentioned, that there exists "something that we cannot act upon with our senses" - I think the idea from previous readings was that more or less everything requires some kind of sensorimotor interaction. The same thing goes for hearsay really - for you to understand the hearsay you will have to have understood, via sensorimotor experience, the various components being explained to you. For example, if the person giving the explanation, the hearsay was talking about looking the mushroom with the two spots on its head, well you'd have to understand all kinds of bits of that sentence, like for one, what it means to have two of something - which you've learned through sensorimotor interactions and grounded that way. Hearsay is like a shortcut to knowing what to abstract! And somewhere up the line there was sensorimotor interaction which lead to the learned abstractions and following categorizations. That being said, I can see why you would worry about it being hearsay for far too long and getting muddled along the way.
DeleteFollow up - I checked up in the reading and first of all, for what I was talking see this quote: "I will have had to learn some things ground-level things the hard, sensorimotor way, if the words used by the grandmasters are to have any sense for me". Then in addition to this, I found the line "Language allows us to acquire new categories indirectly, through "hearsay" without having to go through the time-consuming and risky process of direct trial-and-error learning" - which I think is important to clarify that there can an likely still is direct interaction with whatever the object or thing is, but that you no longer have to deal with the trial-and-error part (expanding on what I meant by hearsay being an abstraction shortcut).
Delete"Yet what explicit knowledge we do have, we can convey to one another much more efficiently by hearsay than if we had to learn it all the hard way, through trial-and-error experience. This is what gave language the powerful adaptive advantage that it had for our species."
ReplyDeleteWhat does this mean for animals who don't have language as humans do? They still communicate, albeit in a very different and (it can be argued) a less complex way. And we know that they certainly must categorize, as categorization is crucial for survival for all types of species. There are two perspectives I could take on this. Firstly, is it possible that animals categorize less efficiently, because their methods of communication do not have the same adaptive advantage? Or is it possible to argue that animal gestures convey just as much knowledge as language does for us? The reason I think this is a plausible point of reflection is derived from comparing spoken to sign language. Sign language is fully made up of gestures and conveys just as much meaning as spoken language does, so is it possible that animal gestures have the same power?
Your perspectives touch on what differentiates between human and other animal categorization, as key aspects of conscious interaction with the greater world. Harnad describes our capacities to interact in five ways: to see, recognize, manipulate, name and describe. In the case of most other animals (as far as we know) the two later categories are not afforded by their cognitive systems. This is one reason why animals "categorize less efficiently" as you say. Another shortcoming of animal learning which limits their capacity to categorize efficiently is their slower and more selective memory. As Harnad makes explicit with the tale of "Funes the Memorious", memory is inextricably linked with the capacity to abstract information and create new categories through learning. So here, relative limits on memory formation or memories dominated by one sensory system deeply affects the ability of an animal to abstract and hence efficiently create categories.
DeleteSpecies without language do not have the higher-order categories that require language to describe them. (No other species has gesture language either. All languages can express all propositions.)
DeleteThe role of memory in categorization is more in (selectively) forgetting what is irrelevant, rather than memorizing it all, as Funes does.
“ Now it might have turned out that we were all born with the capacity to respond differentially to all the kinds of things that we do respond to differentially, without ever having to learn to do so (and there are some, like Jerry Fodor (1975, 1981, 1998), who sometimes write as if they believe this is actually the case).”
ReplyDeleteI am a little confused about whether Fodor is arguing whether the categorizations that take place in our minds are innate or whether our ability to categorize is innate. It would make sense to explore the idea whether our ability to categorize is something we were born with or had to work to create. In most cases in our society we can see how the categories that we have are created through our environment and it is interesting to consider whether some categories have been formed inherently in our minds. On the other hand if we are examining the ability to categorize as innate we ought to look at the past interactions of our species. This explores a deeper question of what our ancestors had to gain from the ability to categorize- which if we are to consider this we stand to gain a lot (from this ability) and lose very little.
Most of our categories are learned. What is innate is not the categories, nor the "ability to categorize," but the ability to learn to categorize.
DeleteAccording to the Whorf Hypothesis, "how objects look to us depends on how we sort and name them." What can then be said about the "bouba-kiki effect" as first described by Wolfgang Köhler 1929, and then elaborated upon more recently by VS Ramachandran and Hubbard? According to their work, everyone may be a "weak synesthete" and has an innate mapping between sounds and shapes where a "bouba" is a more round, blob-like shape and a "kiki" is a more jagged and sharp looking shape. This effect is stable, and does not occur after learning of anything; all children tested consistently sorted the shapes the same way. Is there then maybe some innateness to how we sort and name things BECAUSE of how they look?
ReplyDelete"But to identify a shape in isolation is to make an absolute discrimination (i.e., a categorization), and Miller showed that the limits on absolute discrimination were far narrower than those on relative discrimination."
ReplyDeleteIs absolute discrimination truly necessary? As we process things by their properties (which can be described into features and sometimes dissected into affordances), is there not a great deal of absolute discrimination necessary. So I would argue that when demarcating categories we probably only really use relative discrimination which aids us in a useful functional demarcation towards the purpose that we want to achieve and that is available to us.
"This kind of extreme nativism about categories is usually not far away from something even more extreme than nativism, which is the view that our categories were not even “learned” through evolutionary adaptation: The capacity to categorize comes somehow prestructured in our brains in the same way that the structure of the carbon atom came prestructured from the Big Bang, without needing anything like “learning” to shape it. (Fodor’s might well be dubbed a “Big Bang” theory of the origin of our categorization capacity.)"
ReplyDeleteBut what was the equivalent big bang of the human mind? If it isn’t through evolution, then was it something that happened during evolution of the human or was it at the first birth of the very first most primitive human? How does one come prestructured with the capacity for categorization if we began at the very, very beginning of evolution as a single celled organisms?
This comment has been removed by the author.
DeleteFrom my understanding, the mention of the Big Bang is a metaphor saying that in the same way that carbon didn’t have to “learn” to be carbon/didn’t “evolve” into carbon (it was prestructured to be carbon), categorization (according to Fodor) was inherently prestructured in our brains, the same way that we would explain how Universal Grammar works. I think that truth is that we don’t really know how UG is prestructured in our brains, but we know that it is because we know that it is not something that we extrinsically learn how to do.
DeleteHowever, according to Harnad, Fodor is incorrect in his claim that categories are prestructured. While Chomsky and his theory about UG are supported due to the poverty-of-stimulus-evidence (which is the lack of negative examples when we learn language), Fodor does not have any similar type of evidence. This is because when we learn categories we often have negative evidence (this is a kind of a thing vs. something else isn’t a kind of a thing) which shows that all categories aren’t unlearnt/unevolved. I think in reality what is innate is not the categories themselves, but our ability to learn the categories that we do.
“Although categorization is an absolute judgment, in that it is based on identifying an object in isolation, it is relative in another sense: What invariant features need to be selectively abstracted depend entirely on what the alternatives are, amongst which the isolated object needs to be sorted...So when we learn to categorize things, we are learning to sort the alternatives that might be confused with one another.”I am wondering if the alternatives share many common features with the object, so they are similar enough to be called and recognized by one umbrella term (ex. table, chair, etc.). Does our categorical mechanism priorly tell us to further differentiate them apart or group them together as one category (Assume the situation does not matter)?
ReplyDelete“Language allows as to acquire new categories indirectly, through "hearsay," without having to go through the time-consuming and risky process of direct trial-and-error learning. Someone who already knows can just tell me the features of an X that will allow me to recognize it as an X.” Regarding this passage, I am wondering if a category is learned explicitly and quickly through verbal communication (ex. the distinguishing features are informed verbally by others), is it considered supervised or unsupervised learning?
I found Borges' Funes to be a particular important aspect of this reading. Funes cannot abstract: every instant is infinitely unique and different instants are incomparable. I think the most remarkable aspect of this short story is that it was published in 1942, i.e. far before modern neuroscience of memory. In fact, it was published 15 years before the first paper about patient HM came out. And, incredibly, much of what is said holds up against our modern understanding of memory systems. Namely, that forgetting is absolutely essential, and that without it our memory and cognition would be severely handicapped.
ReplyDeleteFirst, to break this down, we can examine the role of pattern separation in memory formation. Pattern separation involves reducing the overlapping features of memories encodings, allowing for the recall of individual events. We know that Funes has this ability because he is able to remember entire episodes, in fact he may have perfect pattern separation. This, it turns out, is dysfunctional. Perfect pattern separation would mean that Funes’ current memories would not interfere with the encoding of new memories. However, this interference is critical to the formation of conceptual representations, or abstract thinking. As interference occurs overlapping perceptual details of encoded memories are forgotten, resulting in sparser encodings of details and the formation of abstract representations. Dr. Harnad brings up the case of the dog. To have a concept of a dog we must ignore the variance between individual iterations of dog, and understand that this furry creature with four legs, despite breed differences, fits into the category dog. For Funes, every iteration of dog is infinitely different from every other iteration of dog. But, to create a representation of dog = furry creature… Funes would have to FORGET this individual variation tied to the many dog-meeting episodes in order to create a singular abstract representation tied to the animal.
I found the following passage of the reading particularly thought-provoking: "Dynamical systems are systems that change in time. So it is already clear that categorization too will have to have something to do with changes across time.".
ReplyDeleteFrom what I have learned in both my Neuroscience and Psychology classes, one of the most basic and fundamental examples of how we humans categorize is our ability to distinguish between faces and non-faces. Many experiments have shown that at birth, babies have a natural preference for face stimuli and as a result, much of the literature terms this as being an "innate" trait that we humans do not learn through experience. Professor Harnard would argue against the idea that these "innate" categorizations are not built upon learning or with evolution. With this and the above passage in mind (correct me if I am wrong), it could be postulated that we once, in the past, did not have an implicit ability to categorize face vs. non-faces and with trial-and-error mechanisms throughout history, we learned and came to be able to. This I'm also assuming is an evolutionary adaptation that occurred 1000's of years ago. What makes some of our categorizations more or less vulnerable to changes in time? Is it tied with social movements and values? Is there a threshold in which these categorizations must be reinforced in order to be encoded in our brains? Does this decide whether a category is worth creating? What allows sensory categorizations such as color perception to be a lot more salient to categorization than categorizations that we make for things influenced by our social surroundings?
"living in the world requires the capacity to detect recurrences, and that that in turn requires the capacity to forget or at least ignore what makes every instant infinitely unique, and hence incapable of exactly recurring." In our everyday lives, we are constantly categorizing. We need to be able to CORRECTLY select, attend to and abstract which features, objects, events are to be privileged as well as CORRECTLY filter and inhibit which features are to be ignored for one's goal, task at hand etc. We need to be able to modify our behavior and be flexible in our categories to learn, grow and adapt to situations.
ReplyDeleteIf categorization is in Prof Harnad's word "doing the right thing with the right kind of thing", then there is the possibility of making a categorical mistake which will have 'negative' consequences.
This may be tangential, but I am wondering if it is when the mind/brain somehow doesn't make the 'right' categorical judgment over time, when it doesn't perceive changes in/of categories, or does not correctly update or inhibit accordingly to situations that is becomes 'sick'?
Is it that schizophrenia patients have, inter alia, trouble abstracting ? Or that their categories are too porous and they "let everything in"; no longer being able to inhibit as much or what they are supposed to ?
This article also led me to think about the link between creativity and categorization. Could creativity, or making unexpected links between unassociated events, things, features, be related to categorical mistakes ? Is having little inhibition and seeing unimportant features as salient a "source" of creativity?
Harnad claims that categories are learned through sensorimotor experience. He also states that we can learn categories from others through language, a phenomenon otherwise known as “hearsay”, however category learning “cannot be hearsay all the way down”. Harnad states that this is also the case for abstract categories such as “goodness, truth and beauty”. However, it seems to me that a category like goodness is rooted in hearsay or, in other words, influences from one’s family, religious beliefs, cultural group, etc. Given the same sensorimotor experience, then, different people will likely form different concepts of goodness due to hearsay. I am not sure that there are sensorimotor invariants grounding goodness or other categories related to morality such as ‘right’ or ‘wrong’ unless a form of moral realism is supposed, i.e. that moral judgements exist in the world independently of human opinion, which I do not think is plausible.
ReplyDeleteFrom my understanding, abstract categories like goodness, truth, and beauty do in some ways depend on hearsay like your saying. For instance, “one’s family, religious, beliefs, cultural group, etc.” for sure do play a role on what one thinks (I would imagine at varying levels, depending on what one thinks is more important). But at the same time, it is not the only thing that determines our opinion of these concepts, because we, ourselves, interact with the world, and those interactions (sensorimotor experiences) shape our categories. It is from our own positive and negative feedbacks (supervised learning) and our observations from repeated exposure and internal analysis (unsupervised learning) that add to our perception of the world.
DeleteFor example, let’s say that when X was young, their parent told them that “telling the truth is always good”, and that person’s perception of “truth” was something based in reality and “good” was a positive experience. I would think that is hearsay. But let’s then assume that this person one time told a secret (that was true) to someone and had a negative consequence, which would be a sensorimotor experience. Although that person’s overall opinion that “telling the truth is always good” might still stand, they now categorize “secrets” (even if they are true) as something that was in a different category with “truth” since it led to a negative consequence. From my understanding, these abstract categories like goodness or truth get molded by all the different experiences and things we hear in our lives. I would imagine that a certain person’s reliance on hearsay would be determined by how much they value their own sensorimotor experiences or vice versa.
If categorization is cognition, then it would make sense for a T3 robot(whether it is a zombie or not) to have cognition since it almost certainly categorizes. A machine could not pass T3 without being able to categorize, and so it seems that we must infer that it somehow has cognition as well. Or does it not categorize at all and merely treats every single thing as Funes does, each in isolation, and all we see is that it produces the same output. Imagine a machine that is presented with a golden retriever and correctly names it a dog, then is presented with a pug and also identifies it as a dog; what allows us to infer that what they are doing is categorizing? Maybe it’s programming is made in such a way that when it is asked what animal a golden retriever is it replies with ‘dog’ and does the same with a pug or any other species or instance of what the human species calls ‘dog’. But if we apply Occam’s razor, it makes more sense for the machine to actually be categorizing, and therefore cognizing, when it is presented with a dog.
ReplyDeleteThe paper talked about the necessity of selectively ignoring certain differences and preserve certain invariants to achieve categorization, or we would stuck at infinite instances where sorting is impossible. However, if every instances are unique in its own way, how do organisms, especially humans achieve the ability to agree on things sharing the same features and bearing the same names? We must be genetically wired in a specific ways to collectively ignore features that are irrelevant, but if the essence of our cognition is just firing of a huge chunk of neurons in responses to stimuli, how do we achieve this agreement? In essence, how do we agree on anything at all? Evolution might be the answer for this question, for those who are unable to agree to the majority who has the collective uniformity on ignoring things that are not important in the matter of life and death, died due to inability to adapt. However, this only explains why, but does not tell us how.
ReplyDelete“But the categorization problem is not determining what kinds of things there are, but how it is that sensorimotor systems like ourselves manage to detect those kinds that they can and do detect: how they manage to respond differentially to them.”
ReplyDeleteAt first this confused me a little because I was thinking don’t we need to know the what to know the how but from my understanding of this explanation, categorization is being able to tell 2 things apart or detect how similar/different they are and since we can do that with things we have never seen before i.e. don’t know what they are, we do not need to be able to identify something to be able to compare it.
“but for a sensorimotor system, an individual is effectively just as much of a kind as the thing a content word designates”
Proper names name individuals e.g. Jim rather than the kind it is e.g. person/human. Using this example to explain how sensorimotor systems identify by grouping or detecting how similar things are and not by naming is interesting because for example, if we are blindfolded and told to touch water, what our senses “afford” us is that we touched a liquid i.e. the kind or category, we do not at first think or can convincingly verify through touch whether it was water or juice or any other liquid i.e. the individual. Rather the sensorimotor system focuses (primarily?) on the group as opposed to individuality.
“But if the animal had color vision, and we used blue and green as our inputs, the pattern would be different.”
This alludes that the “generalization gradient” is dependent on the inputs and as such we have varying levels of categorization occurring. Where do we decide categorization stops or starts then? Since it is supposed to be discrete?
“Although basic colour CP is inborn rather than a result of learning, it still meets our definition of categorization” and “Categorization is intimately tied to learning”
Does this mean there are different kinds of categorization itself? Categories of categorization? Or degrees of categorization? Because there are apparently innate ones such as UG and then ones we are inborn with such as CP but through evolution learn to do better/faster and most of the rest we learn through detecting similarities and differences since we have no knowledge of their existence from birth?
The reading begins with the idea that “everything in nature is a dynamical system… but some things are not only dynamical systems.” It also states that categorization is a special dynamic system, adapting to changes across time. You can’t have the same output whenever there is an exact same input, but rather categorization happens with the same “kind” of input rather than the exact same input. There is always variety with categorizations, and it “is intimately tied to learning.” However, as stated before, we might innately be born with the ability to respond differently without learning how to do so - so learning might ultimately be innate as well. The reading states that there are different types of learning, including supervised, unsupervised, and instrumental, and it is through these methods of learning that we choose to categorize what we learn into something meaningful. I don’t believe that the capacity to categorize is learned, however. Categorization as a sensorimotor skill, must definitely be learned, but the categorizing capacity that we have as humans is something that we are innately capable of doing.
ReplyDelete