Saturday, 2 January 2016

6b. Harnad, S. (2003b) Categorical Perception.

Harnad, S. (2003b) Categorical PerceptionEncyclopedia of Cognitive Science. Nature Publishing Group. Macmillan.
Differences can be perceived as gradual and quantitative, as with different shades of gray, or they can be perceived as more abrupt and qualitative, as with different colors. The first is called continuous perception and the second categorical perception. Categorical perception (CP) can be inborn or can be induced by learning. Formerly thought to be peculiar to speech and color perception, CP turns out to be far more general, and may be related to how the neural networks in our brains detect the features that allow us to sort the things in the world into their proper categories, "warping" perceived similarities and differences so as to compress some things into the same category and separate others into different categories.

Pullum, Geoffrey K. (1991). The Great Eskimo Vocabulary Hoax and other Irreverent Essays on the Study of Language. University of Chicago Press.


  1. This comment has been removed by the author.

  2. This comment has been removed by the author.

  3. If anyone is looking for some interesting articles on how language might shape categorization and general cognitive abilities...

    (Note: You’ll still have to find these articles on McGill’s online library to access them)

    Boroditzky mentions this article in her speech in the video above which is really interesting on english vs mandarin speakers perception of time (which is more about thought in a broader sense, though can be tied into categorization):

    This article is a great example of how language affects categorization. It compares English and Korean speakers and literally gets the to categorize and group objects together based on certain functions. The two languages afford different groupings based on the prepositions they use to describe objects being on/in/through other objects.

    Lastly there is this article by Brown and Levinson that describes the mayan language that Boroditsky talks about above that uses cardinal terms (north/south) for spatial orientation instead of ego centric terms (left/right) that most languages use. The language Tzeltal describes direction in terms of being uphill or downhill and so upon reading the article you can judge for yourself how this affects categorization.

    My confusion with language and categorization then is with this quote at the end of Harnad’s article

    “to show that it is a full-blown language effect, and not merely a vocabulary effect, it will have to be shown that our perception of the world can also be warped, not just by how things are named but by what we are told about them.”

    Can vocabulary alone not warp our perception of the world? I’ll take this in the form of the Korean vs English paper above. In the paper the kids group objects differently because the words they commonly use to describe how they interact with objects are different from each other. (I don’t recall the specifics) but say for example Korean uses a word that doesn’t just mean “on” or “on top of” but instead more specifically means a certain type of hanging or draping action when put on an object. We, in English, have all the words to describe this yet we choose to use different words. Clearly we have the vocabulary but we still categorize differently. Would a vocabulary effect be that If we had simpler words that described the action the way Korean has it we would then suddenly be grouping objects differently? Admittedly this experiment was done on children and showed the effects of acquisition rather than full comprehension of language. But with acquisition being so important, and vocabulary being so important to acquisition, can you really detatch “vocabulary” effects from “full-blown language effects”?

  4. “Membership in the category may be (1) all-or-none... membership might be (2) a matter of degree… the category is "continuous””
    - Why not have/create a separate term? Why use the term for both all-or-none and matter of degree?

    - I thought the motor theory of speech perception was pretty cool because it reminded me of something heard in linguistics, which is that if one tries to pronounce from “ee”, as in feet, to “ah”, as in opera, continuously, it’s possible to produce the entire possible vowel sound spectrum along the way.

    “… with colors, it looks as if the [CP] effect is an innate one: Our sensory category detectors for both color and speech sounds are born already "biased" by evolution: Our perceived color and speech-sound spectrum is already "warped" with these compression/separations.”
    - I’m always a little iffy when it comes to the word “innate.” I’m really curious about the neural networks to model innate CP effects. When you talk about neural network modeling CP, what does it mean for the input units to be “categorically biased sensory detectors cells”? Are the input units discrete feature detectors in this case, and the “bias” a parameter that affects connection weight shifting? This sounds like the possibility of grandmother cells that correspond to detection of specific features, but the likelihood of these are extremely unlikely isn’t it? And does the hidden-unit activation patterns that represent within-category compression and between-category separation corresponds to the spectrum of an ambiguous stimuli between two extremes that we see in the CP effect? Would the output be the stimuli that we perceive then? So different outputs correspond to different categories and same outputs would be the same category?

    - I really don’t get what you mean by “our perception of the world can be warped.” What does it mean to be “warped”?

    1. Hey Oliver,

      In response to your question about categorically biased sensory detector cells, I believe this refers to cells that are "programmed" to preferentially respond to certain stimuli. For example, those visual cells in the inferotemporal cortex which become critically active to specific line orientations, contour or even particular textures (discovered by Tanaka et al). So, these cells are "innate" in the sense that they are there, at birth, to detect specific features that help us categorize our visual scenes.

      And I believe that "our perception of the world can be warped" simply means that our subjective life experiences can change the meaning we attribute to different things, altering the way we categorize them. So language helps categorize things, but to take it a step forther so does your own experience with objects. Joe believes that laundry detergent is edible (see the show "my strange addiction"!!) - so perhaps laundry detergent is a member of Joe's "edible foods" category but not yours.

    2. But to what extent are these input units innate? If you create these input units to detect all sorts of specific stimuli, not just specific line orientations but to color, to shapes, etc., to the most fundamental broken down features possible, are all these input units still modeling innate capacity? Are we assuming here that we are born with the capacity to identify all possible fundamental features in this world?

      And for the warped perception, what doesn't count as warped then? Is everyone's perception warped? How would we know how to classify a stimuli if everyone views it from a subjective perspective? What would be the fundamental classification then? Because by that definition of warped, everything we know is technically warped because there is no standard categorization of any stimuli, just a relative one based on what most people agree with.

    3. Think about “warped” as being a synonym for “abnormal” or “swaying from the standard”. For categories we (whether innately or learned or both) have standard and most prototypical kinds that we know represent categories. But some other people may consider something a part of a category that someone else or a majority of other people may not (Maya’s example of edible items). This could be related to cultural influences. Another example is that the category of breakfast foods does not in my mind consist of vegetables; it consists of oatmeal, cereal, and fruit. However, someone in the middle east, or I am specifically thinking of Israel, may say vegetables is a big proponent of the category breakfast food since in Israel the main breakfast food is cucumber and tomato salad. So warped/not warped depends on someone’s culture. Warped would be if everyone in Canadian culture seems to think one thing is the norm for a category and there are a few people who think otherwise: their perception is warped.

    4. I see what you're getting at Jordana, but the question I have is do we really have standards/prototypes we use in categorization? Problems like the CP effect makes this "standard" of comparison ambiguous, and the idea of prototype theory is based off the idea that we actually do have a basic level, superordinate level, and subordinate level of categories isn't it? Or else, how can one say there is a standard/prototype/typicality and atypicality?

      If talking about cultures and norms, then there wouldn't really be a "standard" would there? Because standards are constantly shifting from culture to culture, such that it's not standard, it's constantly shifting and changing. Standard/norms would just be the convenient words we use to talk about comparisons between whatever it is we're comparing. Categories are learned, but are these categories then constant forever? I think instead of standards, I would say schemas instead of standards we use to represent categories since schemas can change based on interaction with the environment, but standards really should be "standard", because then the changing standard from culture to culture really contradicts what a standard really means. If asked about what schemas mean, I have absolutely no definition but a concept of it so... ya.

    5. Oliver, to get back to your question about feature cells and whether it's all a slippery slope - i.e. that since we have these feature detector cells (line orientation, contour, texture) in our IT cortex, then maybe we have an innate capacity to categorize ALL our sensory visual input.

      Consider this: We are able to categorize things either from language or sensorimotor perception. Both of these require us to use features, whether acquired from our definition of "edible foods" or from the perceived contour, line orientation and texture of a visual stimulus.

      Notice that in both of these cases, we are born with an innate tool box, capacity. For language one could say Universal Grammar (Chomsky), for vision one could say pattern responsive cells. Having this tool box has allowed us to excel as a species for it has allowed us to categorize our world and presumable cognize within it. But keep in mind that no one is stating that you are born with an entire dictionary, nor are you born with (as you say) cells that could respond to every type of visual stimulus (house cells, bug cells, food cells). I believe we are born with the basics, the tool box that bestows us with the CAPACITY to navigate our world and build more cell-specificities as we do so.

    6. Ohhh ok I understand now! Thanks Maya!

  5. ‘’Can categories, and their accompanying CP, be acquired through language alone?’’
    I do not think that categories and their accompanying CP can be acquired from language alone as as it can be seen through examples of feelings and some categories, language is not necessary to acquire them for a certain individual. Language is probably necessary to communicate the boundaries for these categories and what pertains to them between individuals. A person or an animal, without using language is able to categorize what feelings made him of her feel good, the food he or she ate and that was fulfilling or toxic. Animals, and people to a certain extent are able to know what predators and dangers they have to beware of, after a bad experience or through observation of happenings. In these cases, language is not necessary to categorize these things to the categories they belong to.
    Furthermore, categories that have been established as categories based on the description that things that belong to this category have to have, and that require language to get classified can be viewed as ‘’artificial’’ categories. They are categories that are simply valid through their definition and are not necessary for everyday functioning.
    Some categories that seem to only be definable through language, in most cases can be defined in terms of sensorimotor experiences and observation, showing that language is not necessary to define these categories, if they have some sort of utility and that the things that are in this category really have a legitimate reason to belong to the same category.

  6. “So there is no Whorfian learning effect with colours: Or is there?”

    I think this comes back to my last point in my discussion of the first paper. I think that the compression and expansion of ranges of light frequency occurs because we detect boundaries in colour due to something innately encoded. However, as the author suggests here, while the boundaries are innate, the names are not. So we recognize a certain wavelength as a specific colour but the names of the colours differ based on our experience (learning).

    “Can categories, and their accompanying CP, be acquired through language alone?”
    Here I think it is also to important to refer to the previous paper. We discuss learning categories through hearsay.
    “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“
    Categories cannot be named exclusively through hearsay. Hearsay alone implies no sensorimotor experience. Initial sensorimotor experience provides learners with a foundational vocabulary that can then be used to teach/learn new categories. For example you cannot explain the “new” category “bachelor” as “man” and “unmarried” unless you know what those two categories are. You cannot understand what those two are without some sensorimotor experience (sooner or later). But the initial experience, matching names to other (arguably more fundamental categories) provides us with building blocks with which we can describe and acquire new ones and is therefore a necessary part of categorization.

    1. Since you mentioned Whorf's hypothesis..

      To summarize the Sapir-Whorfian (SW) hypothesis would be to say that mankind’s perception of colour is reliant on the categorical structure of our language. I do not believe that the strong version of the SW hypothesis (linguistic determinism) is true but do recognize that our language plays a strong role in the way we see the world and classify things.

      I recently came across that Business insider article “No one could see the color blue until modern times” It talked about how the word “blue” only appeared relatively recently in some languages and that even Chinese, Greek, Japanese and Hebrew languages used to not have a word for this colour. The article also drew on Gladstone’s analysis on Homer’s The Odyssey in which he describes the sea as “wine-dark” instead of deep blue or green (although I believe it is intended as an allegory). Gladstone’s analysis went on to say that the Greeks seemed to live in a world that was “void of colour”. It was the philologist Lazarus Geiger who went on to later study this phenomenon across cultures to see that the word blue only came into existence not too long ago in some languages. The article seemed to suggest that only language could separate colours – something that I am in great disagreement with. It is not that any colours are “missing” per se; it is well known that we are trichromats that can distinguish different wavelengths of light from each other but perhaps how we view them in relation to each other is in part dependent on social culture/language. Of course the title of the article is misleading by first making readers believe that the actual colour blue did not exist. Similarly, Whorf agreed that our ability to place things into categories originates from our particular language. This language is what we use to conceptualize the environment around us. I guess I am not sure as to what extent our language and its structure determines perception? Although right now, this makes me think of a quote from Shakespeare’s Romeo and Juliet... “What's in a name? that which we call a rose, By any other name would smell as sweet?”

      What do you guys think?

    2. Apologies for the super late response! I think that this information on the history of colours in language is really interesting, especially since they seem to be used quite prevalently nowadays.

      I have a couple comments in regards to your query about the extent to which the structure of language determines our perception. I agree with your analogy using Shakespeare's quote, in that names are relatively arbitrary. If the colour red was instead given some other, arbitrary name xyz, it is doubtful that this would ultimately change the way we perceive colours. Similarly, as the article suggests, perception of primary colours in a categorical manner is most likely "evolutionarily biased," so lack of a name for red may not affect one's direct perception of wavelengths of light that correspond to "red". However, without a definitive 'red' category, this could change the way one categorizes colours, perhaps altering the boundaries of the category 'orange' to be inclusive of 'red' (though I'm unsure whether or not this would be true in reality). Like the article states however, in creating new, arbitrary divisions within the spectrum of visible light, new secondary colour categories emerge (like crimson, scarlet, magenta, etc). To compare this to the Shakespeare quote, all flowers under the category of rose would have the quality of 'sweet' and certain types of roses may have a slightly different 'sweetness' (if we categorize roses based upon scent) much like how certain types of red have a different quality of 'redness.'

  7. “CP occurs whenever perceived within-category differences are compressed and/or between-category differences are separated, relative to some baseline of comparison. The baseline might be the actual size of the physical differences involved, or, in the case of learned CP, it might be the perceived similarity or discriminability within and between categories before the categories were learned, compared to after.“

    What would be the mechanism to establish a “baseline of comparison” for CP to occur? A way of explaining it that had been presented to me in other classes was for the baseline of comparison to emerge as a result of statistical learning. This was discussed in the context of speech acquisition of the sort occurring when we are learning the “pa”-“ba” distinction. If we take “pa” and “ba” to be two ends of the same continuum of voicing, then all instances that we hear can be placed at specific points on that continuum. The way we could learn to distinguish between the two extremes would be a result of how the instances of “pa” cluster at one end and the instances of “ba” at the other. In a language in which such a distinction does not exist, there would be no clustering at the ends of the spectrum and so “pa” and “bad” end up being part of the same category. So, for a child to learn the “pa”-“ba” distinction, he/she would have to keep track of all the instances heard as well as of were they each fall on the spectrum. In doing so, the child can then establish a baseline of comparison, a point on the spectrum from which the instances of “pa” and the instances of “ba” start clustering in different directions.I don’t know if this is correct or if it can help explain other examples of CP, but I think it is interesting way of approaching the problem of establishing the baseline of comparison.

  8. Just a question I've always wonders that was sort of sparks in this papers...
    Do you think that everyone seems colours the same way? What I call red and what Joe Doe calls red look the same? Are there at all the same or just some shades off? Is everyone's colour perception the same? take someone who is colour blind and cannot see green. What they've been learned to know is "green" since a child is their version of green but they don't know otherwise... or do they?
    This is just something I've wondered before and is more philosophical...

    1. Hi Jordana,

      I’ll try to give a (non)answer to your question.

      This problem is usually discussed as the 'qualia inversion problem.'The answer, in short, is that it may very well be unknowable whether any two people, or animals, or robots experience colours, or any other feeling, the same way. It is really a variant of the other minds problem that we have discussed in class. Harnad contends—and I tend to agree—that we can never be certain whether others feel anything, let alone precisely what they feel. However, behavioural standards like the Turing test(s) provide our best clues, and, certainly, this doubt is no grounds on which to abuse the robots in our class, or any others who behave as though they have certain feelings. As to the way those feelings feel to those who are feeling them, and whether they feel the same or different, it seems we are in the dark. There is no way to compare feelings except by feeling them, and you cannot feel someone else’s feelings unless you are them.

      Which is all just to say that you are not alone in wondering about this question, and it seems we can do little more than wonder.

    2. Hey guys,

      I've always found this issue fascinating as well. I would say that everyone sees colors them same in the sense that everyone will detect light with a wavelength of 650 nm as light with a wavelength of 650 nm. However, whether you and Joe both see this as the exact same shade of red, we will never know, because as Timothy points out, we have reached the other minds problem.

      However, we all know people can classify the same object as different colors. I have had many arguments over whether something is actually red or orange, and neither my friend nor I can ever convince each other that we are right. Probably because we both are, in our own minds. So it is clear the category boundary is at different locations for different people. While we all have the same color categories, those categories are not all exactly the same, allowing different people to categorize the same object as different colors. And this demonstrates just how important language and past experience is in shaping how we perceive the world. These color categories are continually modified as we learn and experience new things, shaping a (presumably unique) sense of color perception for every individual.

      Ultimately, we don't know if the color you and Joe each picture in response to the word “red” is exactly the same. What we do know is that we can show you a cup that you call “red” while Joe calls the exact same cup “orange.”

    3. Hi guys,

      Just to add to your point about how past experience shape the way we perceive colour (or anything for that matter). For example, consider the neurological phenomenon Synesthesia. It is defined by wikipedia as "Synesthesia is a neurological phenomenon in which stimulation of one sensory or cognitive pathway leads to automatic, involuntary experiences in a second sensory or cognitive pathway”

    4. Also, take note that it is very difficult to determine exactly how many people have some form of synesthesia, since many individuals would not self-calssifly that their perception is ‘abnormal’ since one cannot compare ones own perception to someone else's.

  9. I found the youtube talk given on language shaping our thoughts to be very stimulating. Language seems to really help define the specificity with which we can categorize. I think it would be interesting to consider categorization in the absence of language. When we categorize all things that are birds as birds, we use the label of “bird” to describe this category. When I see a pigeon, my mind thinks “bird” ie symbol grounding occurs between the features of that pigeon and the word bird. But if we had no word for bird, what would our mind use to label categories? We would obviously still categorize, but I feel like our categories would have to be more abstract groupings labelled as/ defined by the features that make them up. So “birds” to a brain without language would be “all things that look, act and sound in this particular way”. Language allows us to be more precise and organized in the way we go about our categorization

  10. "These language-induced CP-effects remain to be directly demonstrated in human subjects; so far only learned and innate sensorimotor CP have been demonstrated (Harnad 2003).
    Categorical Perception is one of the more interesting topics we’ve encountered in Cognitive Science because it provides a mechanistic explanation for why we are capable of categorizing things easily. By increasing the distance between dissimilar groups psychologically and increasing the within group similarity, it becomes much easier to categorize them and, by extension, do what we need to do with the kind of thing we have in front of us. Perhaps my favourite application of categorical perception were the abstractions that allowed Tajfel to come up with Social Identity Theory. At first he did categorical perception with groups of lines; grouping them increased the perceived similarity within the group and increased the perceived difference between the group. He later showed this applied to social categorization as well, where people perceive outgroups as more different than the in-group they belong to.

    I wonder if language induced CP has not already been demonstrated with such studies on social categorization. Tajfel and Turner’s minimal group paradigm showed simply using arbitrary words to group people into two separate groups (i.e.categories), people would demonstrate a maximum in-group bias. Simple questionnaires could have been used in conjunction with these experiments to ask participants to ask how similar they perceive themselves to the members of their own group and how similar they are on average to the other group. If done as a within-subject experiment, one could easily demonstrate that the similarity between the subject and a member of his group had increased and the difference between him and out-group member had increased. I believe we see language induced CP when we discuss ethnicity all the time. For example, at a South Asian gathering I would say that I am a Sri Lankan to increase the perceived difference between I and other Indians, though we look very similar. I then go on to perceive myself as more similar to other Sri Lankans. However, if I am in a different context, say in a predominantly white gathering, I will use the label South Asian and the perceived difference between I and other Indians would disappear and I would perceive them to be very similar to myself. However, this would increase the psychological difference between my group and the predominant white group at the same time. Language induced CP, especially in the context of social dynamics, has very powerful implications for many fields including political science, sociology and psychology, validating the need for a field called Cognitive Science.

  11. Expanding and contracting categories boundaries offers an interesting explanation that can be operationalized by our machine learning algorithms today. I do wonder, however, if this can not just become another version of the symbol-grounding problem. We say that these areas expand and contract, but to understand and investigate this necessitates taking a “snapshot” at a given time, whereby a set of neurons will presumably encode the boundaries as they exist at that time. I can’t quite wrap my head around the implications of this, but it seems to me like we will still run into the same un-investigatable problem.

    1. Hey nirtiac,

      I think a key difference in the case of categorical perception is that there is empirical data associated with the operational definition of CP (see the page 3 explanation of a learned CP experiment and Figure 2 in Reading 6a). Without needing to take a snapshot of neural nets, we can effectively determine what stimuli are perceived categorically (like blue and green) and what stimuli are not (like black and white). We can teach CP, we can recreate CP-like effects with computational modeling, and we can determine if someone perceives a stimulus categorically, even if they do not speak our language, or any language at all. CP is functionally defined.

      However, this is all different than from the symbol grounding problem because symbol grounding requires consciousness. And because of the other minds problem, we can never know if symbol grounding is actually occurring like we can with CP. As such, it is unclear if we will ever be able to determine how it is happening. However, since we can determine if CP is actually occurring, it's more reasonable to assume we will eventually be able to determine how it works as well.

    2. Functionally defined however, which still means we are doing black-box experiments.


  12. I have a few confusions in regards to the Whorf Hypothesis.
    While I think there may be some truth in the sense that language does influences how we perceive the world, I disagree with the idea that “colors are perceived categorically only because they happen to be named categorically” (p. 3). I think his statement is backwards. We don’t perceive colors categorically because we name them categorically. Rather, we name them as we do because that’s the way we see them, no? Regardless of the name we attribute to a given color, we all see colors more or less the same way. As Harnad says, wee all know red when we see red and blue when we see blue and the fuzziness between colors regardless of having a name for labeling it. So I come to understand that our color perception is not a Whorfian effect, but perhaps our cultural differences in color perception are? But at the basis, colors are not perceived categorically because they’re named categorically.

    Since we use language to label concepts, it’s interesting to think about categorization without language. Is language necessary to categorize?
    I think we have a basic set of concepts that we need to ground in sensorimotor capacities. Once those are grounded, we can start grounding other concepts through hearsay. But is grounding one concept considered categorization?
    I mean there are quite a few categories that are learned before they are linguistically labeled. I think language can certainly help us categorize, whether it is necessary though that is something else.

    1. Hi Cait,

      I was actually thinking the same thing while reading about the Whorf Hypothesis. I agree that it seems more likely that we have the ability to categorize and that plays a role in how our language categorizes things. Just to add onto this thought, I wonder if part of the reason why different cultures categorize things like colors differently could be due to a genetic difference in perception?

    2. Hi,

      I definitely feel that our language has a lot to do with how we perceive.

      For instance, our ancestors did not have a word for 'blue' and hence could not understand the colour the way we do (source: I'm sure they saw blue but didn't know that they were seeing blue.

      I do not think this is because of genetic differences. A quick google search will give you more examples. I guess we understand 'blue' better than our ancestors because we are aware of the fact that we are perceiving blue because we have a term for it.

      Is it possible to extend this argument to other areas such as emotions? For example, Korea has "han" which means feeling sad and hopeful at once and German has "Schadenfreude" which means feeling joy at someone else's failure. We all feel the same emotions, but if we have terms for them in different languages, will that affect the way we think of them?

  13. According to the Whorf Hypothesis … colors are perceived categorically only because they happen to be named categorically: Our subdivisions of the spectrum are arbitrary, (page 3)

    I had some trouble understanding where the Whorf Hypothesis fits into this whole argument, especially considering how many different ways we approach it throughout the article. Dr. Harnad immediately provides evidence to the contrary, proving that people still categorize (in a functional sense) the color spectrum the same way regardless of how they name or subdivide.

    So there is no Whorfian learning effect with colors: Or is there? (page 3)

    Well, we know the Whorf Hypothesis is false, but this seems to suggest that there could be a “Whorfian learning effect” going on. Which means the categorical naming of the color spectrum influences (but by no means exclusively determines) the CP of the color spectrum. How would this work? Learned CP is one answer Dr. Harnad provides, citing recent experiments showing that learning can change or even completely eliminate color category boundaries. Okay, so learning has the “Whorfian effect” of influencing the CP of color, but only because the boundaries exist in the first place. However, the Whorf Hypothesis comes back in a totally different way at the end of the paper:

    so far only learned and innate sensorimotor CP have been demonstrated (Pevtzow & Harnad 1997; Livingston et al. 1998). The latter shows the Whorfian power of naming and categorization, in warping our perception of the world. (page 4)

    So evolved CP is relevant as well, meaning that Whorfian effects, which are facilitated by learned CP, must have some way of becoming genetically cemented over time. It is certainly feasible that the names and divisions we assign to categories of the color spectrum would strengthen those categories over generations of repeated instruction. However, this is “merely a vocabulary effect” and not a “full-blown language effect” (page 5), as the categories already exist, even for those who don't linguistically subdivide the color spectrum. They would still exist, even in the absence of language.

    In summary, the Whorf Hypothesis is clearly false; CP is not all about language. However, there is a “Whorf effect” on CP in the sense that learning new names and subdivisions can modify pre-established categorical boundaries, and these modifications can be genetically preserved, strengthened, and further modified over time. So the Whorf Hypothesis just went a little too far. While it's not all language, language still plays a major role in how we perceive the world. What do you think? Am I getting this right?

  14. I have a couple of comment on different aspect of this article:
    “Liberman et al. (1957) reported that when people listen to sounds that vary along the voicing continuum, they hear only ba's and pa's, nothing in between.”
    When the sound heard is in between ‘ba’ and ‘pa’, I guess there is a decision(probably unconscious) that has to be made whereas if it is a ‘ba’ or a ‘pa’. It is a discrete feeling. As they mentioned, there is nothing in between. What could we possibly hear in between?
    “He suggested that CP was unique to speech, that CP made speech special, and, in what came to be called "the motor theory of speech perception," he suggested that CP's explanation lay in the anatomy of speech production:”
    The motor theory of speech perception suggests that CP is unique to speech. I can understand why one would suppose this, even if I don’t adhere to this theory. There is some categorization, some discreetness in speech. For example, for when you learn a new language. At first, the foreign language feels like a continuous gibberish of sounds. But, when you become better at that language, you start grabbing some common and recurrent morphemes; and eventually some words. You become able to discriminate the word. Could we say that you shape new categorization for that novel language? Maybe if there is an increase in language abilities in human(as well as an emergence in music perception), it is due to an enhanced ability to categorize.
    “It is just that the within-category differences (pa1/pa2 or red1/red2) sound/look much smaller than the between-category differences (pa2/ba1 or red2/yellow1), even when the size of the underlying physical differences (voicing, wave-length) are actually the same.”
    I found it interesting that the physical differences are the same, but that the brain is able to catch those differences. Is this discrimination ability innate for the brain? In the example of color discrimination, can we argue that there is actually a genetic basis for it? I mean, nature/genetic provide us with three cones that are sensitive to different length waves. They are then process on different brain pathway, which could account for their different felt experience... maybe!
    “According to the Whorf Hypothesis (of which Lawrence's acquired similarity/distinctiveness effects would simply be a special case), colors are perceived categorically only because they happen to be named categorically”.
    I don’t think this is an unfounded hypothesis since culture and language seems to affect most of our (learned) behavior. However, for color, there is a discrete and universal physiological categorization system. I do not say we all perceive the color in the same way. But that our brain process color information based on the same physiological basis. So the brain has biological foundation for this kind of categorization, which may account for our perceived distinctness.

  15. As I'm reading these papers on categorization I keep thinking about the ways that the categories change with respect to the question we're asking ourselves about the thing. Just as categories determine "how we see and act upon the world" isn't it also accurate to say that our personal perspectives/histories act upon the way we categorize things? I'm thinking of instances in which we have good and bad connotations or words based on experiences.

    Also, the fact that babies seem to have speech CP before they speak, is related to some biological predisposition to picking out differences in basic sensory inputs—this we can understand. If, though, "our sensory category detectors for both color and speech sounds a e born already 'biased' by evolution: Our perceived color and speech-sound spectrum is already 'warped' with these compression/separations", could an argument for learned categories be that the building blocks of the ‘not possibly inborn’ categorizations are the inborn abilities/biases to discriminate (as in the biases we are born with are what lead us to be able to learn these categories as we develop)? In that sense, we use the inborn abilities we have with things like speech and colour discrimination in order to apply them to learned categories? I feel like this is a logical pathway.

    1. Essentially, I suppose I'm saying that in the process of 'bachelor' inheriting the compression/separation of 'unmarried' and 'man' we had to utilize inborn biases at the core in order to understand what, for example, 'man' was before we could take the next step. This is what I'm seeing as the warping of our world- a sort of combination of inborn biases and experiential biases for categorization.

    2. I agree. I feel that context and cultural biases do have a fair amount to do with categorical perception – adding power and utility to the inborn categories. In this way, there is a symbiotic relationship between the learned and innate categorical perceptions. That through context and experience, the innate categories such as colour perception can be modified.
      “CP seems to be a means to an end: Inputs that differ among themselves are "compressed" onto similar internal representations if they must all generate the same output; and they become more separate if they must generate different outputs.”
      I feel that this sums up the purpose of categorical perception perfectly. We are constantly trying to simplify the processing of the huge amounts of information we perceive. Our brain makes differences within categories appear smaller as in most cases these inputs will require the same outputs. On the other hand, for inputs that require producing different responses, we need to make these differences between categories greater.

    3. "As I'm reading these papers on categorization I keep thinking about the ways that the categories change with respect to the question we're asking ourselves about the thing. Just as categories determine "how we see and act upon the world" isn't it also accurate to say that our personal perspectives/histories act upon the way we categorize things? I'm thinking of instances in which we have good and bad connotations or words based on experiences."

      I think it's definitely true that our personal perspectives and histories inform the way we categorize. Isn't that how we categorize in the first place? We learn to 'do the right thing with the right kind of thing' through trial and error supervised learning, which teaches us how to distinguish members of a category based on their invariant features. Doesn't this constitute a form of 'personal perspective/history'? Indeed the addition of certain connotations, quirks and prejudices passed on by the people who teach us categories (ie, our parents) will probably be learned in a way that informs our own categories.
      It's easy to imagine how subtle connotations (like a racial prejudice for example) could colour one's category learning, and because of categorical perception, colour the way we see the world.

      I'm not sure if that's what you were getting at, so feel free to correct me!

  16. Thinking about the relationship between categorization and language is interesting, and I can't make up my mind as to which affects which more. When I see a pencil or a ruler or a book or a backpack, my mind thinks of "school supplies", so I automatically group them together. If asked to circle which thing didn't belong on a sheet consisting of a pencil, a backpack, an encyclopedia, a banana and a ruler, I would circle the banana. However, that is only because in my mind exists the category of "school supplies". If we didn't have language, perhaps I would pay less attention to the practical use or purpose of the objects, and I would circle the encyclopedia and the backpack as outliers because they aren't elongated, narrow objects like the pencil, ruler and banana. Perhaps my thinking of the objects in terms of their use as opposed to their shape, colour, and any other descriptors is BECAUSE we have existing categories like "school supplies" which is a more commonplace category and more salient than a category such as "elongated objects that are narrow".

    The other side of the argument, however, is that language exists in order to specify these categories, and to make them more accessible and communicable. For instance, perhaps the reason I think of the encyclopedia, banana, backpack, pencil, and ruler in terms of their practical purposes is NOT because of the existing category of "school supplies", but rather because of a natural human inclination towards utilitarianism and then, because of that inclination, we came up with a way to identify that inclination, and so arose language.

    This all being said, I disagree with the Whorf hypothesis that colours are seen categorically because they are named as such. Excuse my following generalization (or my following attempt at political corrected-ness), but it is no secret that colours like pink and red and purple are associated with femininity (consider baby's clothing in stores) and colours like blue and green are associated with masculinity. Consider the fact that nothing about the word "blue" is any more masculine than the word "red" and, similarly, the word "blue" is not similar at all to the word "green". These colours are seen categorically for reasons other than language, and their names are only seen as part of certain categories because of the colours themselves.

  17. “The network's "bias" is what filters inputs onto their correct output category. The nets accomplish this by selectively detecting (after much trial and error, guided by error-correcting feedback) the invariant features that are shared by the members of the same category and that reliably distinguish them from members of different categories; the nets learn to ignore all other variation as irrelevant to the categorization.”

    What I find really interesting here is the nature of the neural network’s “bias”. Could such a “bias” reside principally in the brain’s ventral attention network? Could other biases conducive to categorization also exist in other networks, and if so, which ones?

    A “biased” attentional system seems to me like the closest argument to something regarding an innate conception of certain categories. For example, how human infants will preferentially look at human faces, or preferentially attend to human voices over other vocal stimuli. This doesn’t mean that the categories themselves are innate, only that we have some neural mechanisms that facilitate their acquisition. Furthermore, this doesn’t come close to arguing that all categories would be innate. Just that a great deal of within-category compression takes place for these specific objects, like human faces and voices.

  18. Could someone clearly denote the differences between categorization and categorical perception?

    1. Hi Naima,

      Categorization is 'doing the right thing with the right kind of thing'. It can be the result of either supervised learning (involving yes/no feedback from a teacher or from the object/environment itself), or unsupervised learning (which is less common, but does not require any error-corrective feedback, and is the result of repeated exposure to and processing of a 'kind' with very salient affordances e.g. valleys vs. mountains). Thus, categorization sort of encompasses everything that we do in interactions with our world, including naming (which is simply one of the possible 'right things to do'). It is clearly indispensable when it comes to survival, and enables us to make sense of our world/ not see it as a "blooming buzzing confusion".

      CP is quite different - it is a unique kind of sensory perception wherein categorization affects how we actually perceive certain things. The best illustration of it is the colour spectrum. So of course the different colours we see are the result of their different wavelengths, with blue at about 400nm and red at ~700nm, with all the others in between. When the average person views the spectrum, CP results in us perceiving a difference of 50nm (completely arbitrary number I chose) between two colour categories (e.g. blue and green), as considerably larger than the same difference of 50nm within blue category. Thus, the two blues seem more similar than the blue/green pair, even though the actual quantitative difference between them is the same. Does that make any sense? Cause that's really the essence of CP: differences within categories are made smaller by our brain, and differences between categories are made larger.
      Some CP, such as colour perception, is an inborn trait that presumably enables us to more easily/effectively categorize and had an evolutionary advantage. Other types of CP can be learned, as discussed in some experiments from the readings.

  19. After reading this article I have a question about the Whorf Hypothesis. From what I understand, it is saying that language is what causes us to categorize things the way we do, and that it is not something inborn or evolved that affects our categorization. I am just wondering if the Whorf Hypothesis includes culture with language?

    By this I mean that the culture we are in, and the environment, affect the development of our language and probably the way we categorize too. If I am raised in a topical rainforest and need to know which berries are safe to eat and which are poisonous, I would probably learn to categorize all types of berries into more specific labels from a very young age. My culture would probably develop names for these berries, because the berries are so important in this environment. If however I live in a city, my language probably would not develop a name for every type of specific berry, and if it did I wouldn't need to know them all. If you were to put a tropical rainforest berry in front of rainforest Amanda and city Amanda, rainforest Amanda would probably categorize it as safe or dangerous first, but would have more possibilities for categorization due to her culture and environment.

    All of this to ask if the Whorf Hypothesis would recognize this difference in categorization, due to an environmental difference causing a difference in language and associations with certain berries? Or would it ignore the fact that the environment affects the perception and language of items and attribute any difference to only language?

  20. About the Whorf Hypothesis; I felt like it was specifically about naming (vs categorizing). As in two systems (eg languages) may have the same category boundaries but different ways of naming them. Because naming or renaming still wouldn’t change inherent properties of the category or reshape it in any way. So language isn’t actually reshaping categories or a difference in perception, it’s only the labeling that varies between languages. Therefore the naming differences can easily be linked to a difference in language evolution/culture etc… rather than an actual perceptual difference (although that can be the case too, e.g. color blindness). It sort of feels like automatically dismissing the hypothesis comes from a misconception of what it actually is, because its’ meaning behind “perception” isn’t a physical change, but instead some sort of mental perception, how perception and associations can change based on culture and language (e.g. red being associated with anger).

    1. The dismissal of the Whorf Hypothesis does not come from the fact that we are misunderstanding perception as a physical change. The whole point is that physical difference aren't perceived as what they are due to change in (mental) perception (in the case of colours, this is innate and not culturally affected). The different names given to colours in different languages/cultures don't affect how we perceive them, but our inborn feature detectors in our retina do.

      "Our subdivisions of the spectrum are arbitrary, learned, and vary across cultures and languages. But Berlin & Kay (1969) showed that this was not so: Not only do most cultures and languages subdivide and name the colour spectrum the same way, but even for those who don't, the regions of compression and separation are the same. We all see blues as more alike and greens as more alike, with a fuzzy boundary in between, whether or not we have named the difference."

      So if you give people from different cultures the same colour spectrum, they will all 'misperceive' the physical differences between the colours the same way. This is what he means by "the regions of compression and separation are the same": compression meaning colours that are in fact further apart physically are perceived as more similar (i.e. in the same category), and separation meaning that colours that are in fact closer together physically are perceived as more different (i.e. in a different category). So all humans will put two shades of blue in the same category and a shade of red in a different category even though the shade of red may in fact be physically closer to one of the shades of blue.
      This is what we mean by physical differences and perception. I hope this makes sense.

    2. Wow thanks so much for this clarification, I was clearly very confused and wrong about this. So if I understand it correctly now, what you're saying is that the dismissal is based on some sort of perceptual bias? We don't perceive things the way we should, and because of that we might group things together that might not be as similar as one of those to a third thing?

  21. There were several points I just wanted to clarify.
    "The motor theory of speech perception explained how speech was special and why speech-sounds are perceived categorically: sensory perception is mediated by motor production. Wherever production is categorical, perception will be categorical; where production is continuous, perception will be continuous. And indeed vowel categories like a/u were found to be much less categorical than ba/pa or ba/da.”

    I am still having a bit of difficulty fully grasping this. The theory states that speech sounds are PERCEIVED, therefore it seems a bit ambiguous to clear-cut state that they can be viewed categorically or on a continuum. Then it makes me wonder if this is specific to humans, could other animals, perhaps chimpanzees, who were trained a new language still follow through with the same theory? In addition, what if the motor aspect was impaired and someone was unable to speak, how would this affect the categorical/continuous perception? Would accents contribute in any way?

    "Our subdivisions of the spectrum are arbitrary, learned, and vary across cultures and languages.”
    Also, I am going to have to disagree with the Whorf hypothesis. I think research is advancing and with the current generation, it seems like there is a lot more ambiguity with colour categorization. Where pink and purple were once seen as feminine, and blue and green were seen as more masculine, it seems as if the colour preferred divide is disappearing. Personally speaking, I love the colour blue and I never really liked pink or red. If the subdivisions of the spectrum are arbitrary, , learned and vary across cultures; it seems very hard to maintain a consistency with the ever-changing categorization of colours.Therefore, it seems like the categorization of colours is becoming much harder according to the Whorf hypothesis.

  22. I found that a lot of my comments on this article also applied to 6a, so I struggled to create commentary that has not already been hashed out in the previous thread. I do have some questions/comments about the class discussion however:

    In Monday's class there was a bit of a divide between whether dynamic interaction with the world (dancing specifically) is in fact just another categorization. A lot of confusion between the perception of the self while dancing vs. the perception of a dancer to an un-dancing outsider occurred. I don't personally think that continuous movements in dance are categorization (to the dancer) per say, but does one not have to categorize their intention to dance as dancing before they make such movements? There seems to be some fuzziness here for me. If I'm making the same "dance-ish" movements as a real dancer but I myself am saying that through my own perception I am not in fact dancing, can anyone argue and tell me that I am? Isn't the categorization happening because of my intentions? Or can this be counted as categorization because it can be corrected by other people when they tell me I am in fact dancing? Dynamic interaction with the world seems to fall on the border between uncorrectable feelings and correctable categorization. I am clearly a little confused about all of this.

  23. I agree completely with this article. The fact that a French speaker vs an English speaker may hear the distinction between the sounds ‘ba’ and ‘pa’ at a different point along the spectrum of voice-onset is simply a different manifestation of categorization by sound exposure, not actually by language. An example of this can be that if a 50-year-old French speaker learns English, they will maintain the voice onset that they are accustomed to through French exposure, and as such, if they were to then teach a child English, this supposed distinction that comes by language learning would be different (it would be according to the French-onset rather than English, even though the child is learning English, assuming this adult is their main / only point of exposure to the language).

    A possible example of how language influences our categorization directly at a higher order, not through sensorimotor experience, could be in the form of more theoretical ideas and their media representations within a culture. For example, the concept of terrorist is highly influenced by media, such that the language and culture that you are exposed to will thwart your categorization of the word, without thwarting the actual definition. For a current example in western culture, the term ‘terrorist’ is often only used to describe very specific acts of aggression against a population by a very specific genre of people. The CP of this genre of people then gets easily gets inherited by the more abstract notion of ‘terrorist’, regardless of being completely unrelated to the actual linguistic definition of the word, and as such we end up with the entire basis of Donald Trumps (horrific, in my opinion) presidential platform.
    I would argue that this is not merely a vocabulary effect, but is a full-blown language effect resulting from exposure to people that are named as terrorists, but also what else we are told about them.

    1. Hi Esther,

      I like the example you used about how a cultural milieu can affect the way we categorize, thereby affecting our categorical perception.

      "A possible example of how language influences our categorization directly at a higher order, not through sensorimotor experience, could be in the form of more theoretical ideas and their media representations within a culture. For example, the concept of terrorist is highly influenced by media, such that the language and culture that you are exposed to will thwart your categorization of the word, without thwarting the actual definition."

      I think we have similar ideas, but I'm a little confused about what you're proposing here. I'm not sure what you mean when you say that language influences our categorization. These two seem inextricably linked in the sense that categories are like the currency of language... They are the symbols which language permits us to manipulate and make propositions about. I'm not sure I understand what you mean when you say that the culture we are embedded in "thwarts" a categorization without "thwarting" a definition.

      If I could run with your point a little bit though, I'd say we're both thinking about categorical perception -- the process by which things in the same category appear to be more alike, while things in different categories appear to be more different. We could imagine that the culture we are embedded in provides us with feedback about categories, feedback which suggests that members of the category "terrorist" are also probably members of certain "race" category, rather than the definition you gave. For some individuals, this may result in an update to their category for terrorists which now contains an inherent racial prejudice. This sort of unfortunate category learning would also result in a learned categorical perception by which in-group individuals are perceived as more similar to one's self, and by which out-group individuals (ie individuals of the given race, or hailing from the country of putative attackers) are perceived as dissimilar. This sort of perception can lead to the sort of unfortunate divides you described.

      Then again, I could be totally off base. Let me know whatcha think, or if I misinterpreted anything?

  24. "According to the (now abandoned) motor theory, the reason we perceive an abrupt change between ba and pa is that the way we hear speech sounds is influenced by the way we produce them when we speak."

    I think that the motor theory of speech perception assumes a very culturally blind (or deaf) position. In his theory, Liberman overlooks the fact that other languages use various phonemes that the English language does not carry (e.g. there are two individual letters for /b/ sounds in Korean, as in bay and bird) and vice versa (e.g. there is only one letter for the /r/ and /l/ sound in many Asian languages). Because of this difference, many adult Koreans, for example, cannot physically pronounce the /r/ sound. Although these Korean speakers do not have the motor capacity to pronounce the different /r/ and /l/ sounds, they are able to distinguish between the two sounds when they are presented. This demonstrates that sensory perception is in fact not mediated by motor production.

    Granted, this theory was published in 1957 and has been since abandoned, but considering different languages provides more insight into how language shapes categorical perception.

  25. “Membership in the category may be (1) all-or-none, as with "bird": Something either is a bird or it isn't a bird; a penguin is 100% bird, a platypus is 100% not-bird. In this case we would call the category "categorical." Or membership might be (2) a matter of degree, as with "big": Some things are more big and some things are less big.”

    I understand the distinction being made here between the two types of memberships but this seems slightly contradictory to discussion we’ve had in class. When we categorize, isn’t it always an all or none distinction? To say something is in a category means to differentiate it from other things based on a specific quality, thus it is either a member of the category or not. Yes, it is true that certain things can better represent a category, but in the end, the “thing” in question is either in the category or not. With this in mind, there shouldn’t be a distinction here regarding something’s membership in a category because it always is an all-or-none situation.

    "Some of our categories must originate from another source than direct sensorimotor experience, and here we return to language and the Whorf Hypothesis: Can categories, and their accompanying CP, be acquired through language alone?"

    Is this question asking whether there are categories that can be acquired with only language, and that acquisition cannot involve sensorimotor experience? As in the explanation cannot draw upon sensorimotor experience to develop the category? If this is the correct interpretation do these categories exist? For example, our categorization of numbers in mathematics such as prime, rational, even, imaginary etc. are not learned through sensorimotor experience but rather through language and supervised learning. With this in mind, can we consider the categorization of numbers as something originating independently from sensorimotor experience, according to the criterion outlined above?

  26. I disagree that identification is an absolute judgment. The capacity to tell whether or not a given input is a member of a category requires a relative judgment as to whether the features of the input match a category or not. Even if these iconic representations are not prototypical in nature, a relative comparison is still required for their identification.

    A concern brought about by the symbol grounding problem is in learning through the Chinese-Chinese dictionary, concluding that simple manipulation of symbols cannot lead to understanding. Since the purpose of manipulating those symbols is through the necessity of having to describe another iconic image, or a new characteristic, beyond the elementary few, symbol manipulation must be at least originally grounded in something meaningful.

    A final concern is that perception of a thing is automatic and internal, and manipulating syntactic symbols into a semantic category is also internal. “Behavioral capacity” is mentioned in the final paragraph, but I am unclear on how it relates to the symbol grounding problem.

  27. These language induced CP effects remain to be directly demonstrated in human subjects; so far only learned and innate sensorimotor CP has been demonstrated .”

    After reading this article one of the questions I am asking myself is if it might be valuable to try to study what happens when humans miscategorize versus what happens when computers miscategorize?
    I think I understand the difference between how computers caterogize and how we categorize for the most part. When a computer engages in CP, it does so only through trial and error (which it can be trained to do). So it doesn't have anything to do with sensorimotor experience. It has more to do with the learning of YES or NO to a given member of a given category, which seems to follow the Whorfian hypothesis that we build categories out of language. When humans are engaging in CP, it about the sensoriomotor feedback being matched with a linguistic category (more interactive and dual processed it seems to me).
    So when mistakes happen for a computer or a human, it interested me to know whether it is a mistake linguistically attributing it to the category (like being taught the wrong name for a category, I used to think magenta was an orange but I guess it is a pink) or if it is a mistake in paying attention to the wrong parts of a member of a category, causing you to miscategorize it. Like the example at the beginning of the article, if you see a platypus and only see that it lays eggs and has webbed feet you might think it is a bird even though it is a mammal.

  28. When talking about categorical perception, I think about it in it’s simplest form the sensorimotor boundaries between categories. I feel this makes sense when talking about a rainbow, you can see the categorical perception in the bands. Colours are separated. However, is all boundaries categorical perception? Are the boundaries between textures or animal species categorical perception? Does it have to be ever present boundary like sea and sky? And if not, where does the expansion and compression come in.
    Colours as an innate category does bring up the idea of colour blindness. Nothing can make someone that is colour blind be able to distinguish the difference between red and green. It does make me wonder if colour blindness was evolutionarily favoured, and a majority of people were colour blind, how language and culture would be affected. Thinking about the Whorf hypothesis in its reverse. Would we still have these categories in language?

  29. It is really good that we can additively combine the categorical perception from two or more grounded categories linguistically. This is the true “nuclear bomb” that language offers us as a species. Without this capacity, it would all be for nought. The computer simulations are empirical support for the minimal grounding set; it makes the dictionary a lot easier to learn than it would be if we could not build combine categorical perceptions linguistically. How could we demonstrate this ability in humans, considering it has not yet been directly observed?

  30. So I would like to talk specifically about the snow problematic. So like, the idea is that a particular language may have more words for a particular object so people using that language may perceive more categorical distinction in the object of their perception. If I have 50 words for snow, then I am seeing more snow qualities. Since I have more words, my abilities to distinguish characteristics of the snow has improved. I think Stevan rightly puts this idea down. Just because I might not have 50 single-words to describe something does not mean that it is impossible to come up with 50 multiple-word-phrases to describe something.

    I think I would like to expand my understanding of what Stevan says about the 50 snow-words. Have I learned to perceive distinctions to a greater ability simply because I have more words at my disposal? Is it the grounding that comes first or the word? Could it be that learning language is kind of like learning faces? I personally cannot create the image of a human face that I’ve never seen before. There is a theory that all the faces I ever see in my dreams can only be faces that I’ve actually seen in real life. Is categorization like faces, in the sense that I cannot create new words which are then grounded?

  31. I’m still wrestling with how do we separate between what is innate and what is learned? For example, there are differences with different countries that lead to humans having different speech sounds; such as with laughs (A british laugh sounds different than a Canadian laugh). Could this difference be innate, or learned because our laughs can be influenced based on our parents upbringing or the environment we live in. Thus, how does learned vs innate categorization in language apply to different languages?

    1. No one knows exactly what is innate, but laughs certainly aren't: British babies are not born with British laughs; they learn them; and if adopted by Canadian they learn Canadian laughs. Also, part of phonology is categorical and part is a continuous motor skill.