Harnad, S. (2003b) Categorical Perception. Encyclopedia 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.
“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 there is no Whorfian learning effect with colors: Or is there?”
ReplyDeleteI thought it was interesting that this came up, since my Language Acquisition seminar also discussed the Whorfian Hypothesis a few weeks ago. I read this article by Winawer, Witthoft, Frank, Wu, Wade, and Boroditsky (2007) about how the Russian language utilizes different words for lighter blues (“goluboy”) and darker blues (“siniy”). The Russian speakers showed a ‘category advantage’, meaning that they distinguished lighter blues from darker blues crossing the boundary faster than an English speaker, who categorizes lighter and darker blues as “blue” (Russians do not have one term for all blues). This would argue for the Whorf hypothesis that language does affect perception – since the Russians had different categories, this ultimately lead to different discrimination times. I would say that language does allow us to categorize more effectively.
“That is enough to rehabilitate the Whorf Hypothesis from its apparent failure on color terms (and perhaps also from its apparent failure on eskimo snow terms, Pullum 1989), but 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.”
I don’t understand this last sentence. Does this mean that language must also affect our world based on what we’re told about these perceptions? It seems to me that the vocabulary of any given language is tied in so closely with the language that determining the effect of its influence would be extremely difficult, since stripping away the vocabulary and words would leave you with grammar and other various aspects of the language (which we conceptualize with words). I don’t quite see the difference between vocabulary effect and “what we are told about them”; they seem like the same thing to me.
As far as that last sentence, I think he means to distinguish named categories that have been picked up through ostension from described categories.
Delete"But when we look at our repertoire of categories in a dictionary, it is highly unlikely that many of them had a direct sensorimotor history during our lifetimes, and even less likely in our ancestors' lifetimes. How many of us have seen a unicorn in real life? We have seen pictures of them, but what had those who first drew those pictures seen? And what about categories I cannot draw or see (or taste or touch): What about the most abstract categories, such as goodness and truth?" (Harnad)
At least, that's what I'm gathering from that last section.
What I'm wondering is "how that would be tested?" Not sure. If one were described a unicorn, would that person be more likely to classify horses with a horn as more similar to each other relative to other features, same color skin, having or not having a mane, etc? I'd assume so. It seems like categorical compression is a given, but doesn't seem as impressive as CP of colors or vowel sounds, since we don't have conscious access to why we tend to compress colors or vowels, while we do have an explanation for why we grouped those horses that way ("because those looked like unicorns"). Does the triviality of this result diminish its import? I don't even know.
Should we be looking for compression and separation that is somehow unconscious? How could we even do that with described categories?
With regard to the last sentence, I think it's important to refer to the computer example in the second-to-last paragraph:
Delete"they can be combined into Boolean combinations (man = male & human) and into still higher-order combinations (bachelor = unmarried & man) which not only pick out the more abstract, higher-order categories much the way the direct sensorimotor detectors do, but also inherit their CP effects, as well as generating some of their own."
In the example of 'man = male & human' a computer combines terms from a dictionary and relates them to create higher-order terms in a dictionary. In this example there are two dictionary terms (male, human) and relationships (&, =), which are assigned to the higher-order term 'man' by writing the statement 'man = male & human'. In computer science, there are a variety of relationships (=, AND, OR, NOT, XOR, etc) that can be used to relate two terms. We cannot use these relationships in statements if we do not have terms to use them with, but when they are combined with terms they create a statement that has more meaning than just the terms themselves. In this example, 'man = male & human' means 'a man is a male who is also human' to a computer whereas simply listing the terms as 'man, male, human' doesn't tell a computer anything it didn't already know and the term 'man' is still undefined (side-note: I obviously am not making the claim that the computer actually understands meaning, just using these terms for clarity of thought).
I think what Harnad is getting at is that there may be a similar mechanism in humans whereby humans may define new categories by using words to describe the relationships between existing categories and assigning these descriptions to a new word which can be used to define a new category (e.g., 'a zebra is a horse with black and white stripes' --> we can use this sentence to form a new 'zebra' category even if we have never seen a zebra). Note that this would mean that this mechanism would necessarily depend on words (e.g., 'zebra', 'horse', 'black', 'white', 'stripes', 'a', 'is', 'with', 'and') and relationships (e.g., 'a', 'is', 'with', 'and', the order of the words used, the number of times a word is used) to define new vocabulary terms and categories. He says this 'full-blown language effect' hasn't been shown in humans yet.
In terms of how to test or determine this effect in humans, Harnad doesn't really say how this might be done. I think it could be done by creating an experiment similar to those that have already been done for learning but instead of using a training task, use a description. For example, show participants a set of paired stimuli and estimate the pairwise similarity/discriminability. Then, instead of putting the participants through a learning task, tell half of them about a new category by using both words and relationships (e.g., 'a zebra is a horse with black and white stripes'), and tell the other half about a new category by using only words in a random order (e.g., 'black', 'zebra', 'horse', 'stripes', 'white', 'a', 'is', 'with', 'and'). Then test the pairs for similarity/discriminability after they know about this new category. If the former shows a change in similarity/discriminability and the latter does not then we might be able to say that simply having the word 'zebra' and a set of words is not enough to convey the category zebra, but by using language to it's full capacity we can see there is a 'full-blown language effect'.
It has been hard to find something new to say from what has already been said, seeing that this reading follows very closely to the first one, and re-reading it after having done my first commentary shed some light on the distinction I tried to draw between categorical perception and categorization. Here the focus seems to be on the question “how do we show a full-blown language effect for categorical perception?” and I agree with Ethan’s and Jessica’s proposed experiments would be very interesting to make. My guess though, would be that the word to be defined, could not be part of the vocabulary of the subject. In particular with Ethan’s example I don’t think the word Zebra could be used with adults (perhaps with kids yes, and let me present some personal experience: during the winter break I gave a stuffed animal (a zebra) that belonged to me to a friend’s son, Santiago (he is 3 years old) and immediately after getting it he said ‘un caballo!’ (horse in Spanish) and of course we immediately proceeded to correct him that no, it wasn’t a horse but a zebra. Now that I think about it, Santiago’s categorical perception was very close to the truth). Back to designing this experiment, the task would require an arbitrary word that doesn’t belong to the subject’s vocabulary; say: ‘melaggle;’ then half of the subjects would be presented with the new category: ‘a melaggle is a camel with white and orange stripes,’ and the other half would have the same words presented, as Ethan said, in random order. The only problem though I see with the random presentation is how you “keep” the subjects from automatically being able to re-order the words, or being implicitly influenced by the random presentation (where perhaps some unconscious ordering happens on its own) and does this even matter?
DeleteIn his article, Harnad states that “stimuli to which you learn to make a different response become more distinctive and stimuli to which you learn to make the same response become more similar”. This statement raised some questions to me regarding allomorphs and their perception.
ReplyDeleteAn allomorph is a set of different speech sounds that are perceived as a single morpheme (a meaningful unit in a language). For example, when you say the words “walked”, “begged”, and “banded” you understand that these words are verbs and the verb ending “-ed” signifies that these actions happened in the past. However, you produce these allomorphs in different manners; in “walked” you produce /t/, in “begged” you produce /d/, and in “banded” you produce /ɪd/. These allomorphs are produced differently and sound differently, but are perceived as meaning the same thing.
I understand that language induced categorical perception (CP) of phonemes is due to innate factors and learning. When you learn a language, you learn that there are different phonemes (basic speech sounds) that can be based on place (where in the mouth), voicing (whether or not your vocal chords are vibrating), and articulation (how you produce the sound). These phonemes can be combined to make different words. You know these words are different because they contain different phonemes. For example, you know rod and rot are two different words because one ends with the phoneme /d/ and the other ends with the phoneme /t/.
We have learned to react differently to these phonemes in certain cases, but we have also learned that we must act the same way when these phonemes are combined in specific ways. But what is the point in distinguishing these phonemes into different categories in the first place and then perceiving them the same way when these distinct phonemes are combined in certain ways? Wouldn’t it be more efficient for a morpheme to have a single manifestation instead of multiple?
That is enough to rehabilitate the Whorf Hypothesis from its apparent failure on color terms (and perhaps also from its apparent failure on eskimo snow terms, Pullum 1989), but 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.
ReplyDeleteAlthough not directly related to linguistics and cognitive science, I've heard of social psychology experiments that seem to prove these concepts. For example, we'll perceive individuals differently based on what we have been told about them beforehand, and based on this impression, we'll interpret all further actions based on this framework.
If we are told Joe is: (1) a heavy drinker and partier or (2) a scientist on the cusp of ground breaking research, when we meet him we'll perceive his actions differently. In the first case, if we see him yawn, we may assume he has been out the night before, though in the second case, based on the identical behavior, we might assume he spent the previous night at the lab. These opinions may change our affect toward him. We may be resentful of Joe based on (1), and sympathetic towards him based on (2). In this example, his behaviors are judged in accordance with what we've been told about him, leading us to categorize him as a party animal/dedicated scientist, and this alters our perception.
Although this is a simplistic example, I think it can be further extended in relation to the Whorf hypothesis, and in how categorization in different ways can alter our perceptions.
That is an interesting example, even though it isn’t linguistics or cognitive science. Your last sentence “I think it can be further extended in relation to the Whorf hypothesis, and in how categorization in different ways can alter our perceptions” and it made me think of how perceptions also alters categorization. We can see this in the McGurk effect. If we look at a video a person mouthing the sound “ba” (lips roll in together), we hear the sound “ba” repeated over and over again. However, if the person starts mouthing the sound “fa” (lower lip under upper teeth) we start to hear the sound “fa”, but when closing your eyes, you can hear that the sound that is actually played is still “ba”. Our mind is therefore changing our perception of a sound based on the visual information given to the brain. Therefore, depending on the visual information, our categorical perception either associates the visual perception to the sound “ba” or “fa”.
DeleteTherefore to make a link with your first paragraph and Harnad’s ending sentence, it will have to be shown that perceptions of the world are indeed influenced by how things are named and what we are told about them as well as all the visual and sensory information we gather about them.
I don't think that categorical perception refers to the way categories change the way we think about things though. I understood it more as the categories literally changing our perception of objects With Joe, we're still seeing him the same way perceptually, but now attribute his actions to different causes, which i think is a pretty indisputable byproduct of categorizing people. The confusion probably arises from social scientists using perceive in a more general sense
DeleteYes, we have to distinguish what "merely" changes our interpretations and responses and what changes our perception. The jury is out, but the border between the two does sometimes get blurry.
DeleteCategorization Perception Effects in Humans
ReplyDeleteIt feels as though I’m continuing my Skywriting from the last topic here, but I thought it relevant to bring up some interesting research I’ve found with regards to categorization perception’s effects in “the way we see the world”: my conclusion is that categorization (schemas) has little to no effect on perception, but has a large effect on memory. All humans can tell apart red from orange. However, some languages do not have separate words for the two – and although they see the perceptual bands just fine, learning a word for “orange” allows them to abstract the two categories separately, and, in recall, more accurately remember whether the color of the thing they were describing was closer to our prototypical "red" or our prototypical “orange.” Conversely, if the language has a word with an associated prototypical shade of, say, blue-green, they’ll do much better at remembering whether things were of that colour than the average English speaker. Moving on to a more interesting example (we've all heard a lot about Whorf by now), a 1986 study by Hoffman, Lau, and Johnson with Chinese-English speakers found that when Chinese-English speakers were presented with stories involving “stereotypes” (think “jock” or “redneck”), the words and associations towards which vary widely across the two languages, individuals were able to remember more attributes relating to a character in a story if the character met the criteria for a stereotype in their given language. I assume they were also more likely to describe features that weren’t present in the story, but are associated features with the given stereotype. None of this is surprising, given an understanding of how chunking works – to economize, we assign labels to stuff – categorizing kinds of kinds. We lose some specificity but gain an enormous amount of manipulation power through this process. And we see evidence of this process in the way our memory works – forgetting the specific referent in favour of the label; ignoring the trees for the benefit of the forest, so to speak.
What is categorical perception (CP)? I think the easiest way to get your head around this idea is through an example. Colours are on a continuum, a spectrum. The difference between a very dark yellow and a very light red is about the same the difference between a light red and a slightly lighter red (in terms of degrees of difference). And yet, I see dark yellow and light red as TOTALLY different colours, whereas I categorize both light red and slightly lighter red into the “red” category. Weird! In more technical terms, Harnad writes “CP occurs whenever perceived within-category differences are compressed and/or between-category differences are separated, relative to some baseline of comparison.” So, in the above example, the difference between light red and slightly lighter red is compressed, whereas the difference between very dark yellow and very light red is separated.
ReplyDeleteAccording to the Whorf Hypothesis, our subdivisions between different colours are learned and these subdivisions vary between cultures. Turns out he was wrong: our ability to categorize colours is innate (but we probably evolved, through natural selection, to be able to categorize colours).
At the end of his paper, Harnad suggests that we may be able to rehabilitate the Whorf Hypothesis, writing that while this hypothesis has failed on colour terms, he thinks we can show that it is “a full-blown language effect, and not merely a vocabulary effect” if we can show 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.” What on Earth does this mean? Here is my stab at it.
First, I think Harnad shows in his paper that the Whorf hypothesis is a vocabulary effect for some categories. The Lane/Lawrence experiments show that “CP can be induced by learning alone.” Our perception of the world can be warped by how things are named. Hypothetically, I think this means that we could (potentially) be taught to sort reds into “crimson” and “non crimson” categories through trial and error. An experimenter would tell us when we have sorted the red into the right category, and when we have sorted it into the wrong category. Eventually, we would get really good at telling crimson from other reds. The within-category differences would be compressed (between the reds in the “crimson” category), and between-category differences would be separated (between “crimson” and non crimson reds).
But It has yet to be shown that the Whorf hypothesis is a “full-blown language effect.” To show that the full-blown language Whorf hypothesis holds for the crimson category, I think we would need to show that the difference between “crimson” and other reds could be learned not only through trial and error, but also by being given a linguistic rule that would allow me to categorize reds accurately. For example, if you told me that crimson is a strong, deep red color, inclining to purple, and I already knew what strong, deep, red, and purple meant, the full-blown language Whorf Hypothesis would hold if I could categorize reds accurately into the crimson category merely after being told this rule, with no trial-and-error feedback.
I don’t have much to say since I view this article rather as an extension of Harnard’s “To Cognize is to categorize”.
ReplyDeleteHowever, Harnard prompted the question on whether “categories, and their accompanying CP, be acquired through language alone?”. This raised me the same question I had in the previous article.
What about modelling and vicarious learning?
I do feel that a LOT of our knowledge today comes from social factors rather than “trials-and-errors” factors. Phobias are a good example. Someone could be deadly afraid of spiders, despite having absolutely no prior experience with sharks. Thus, the spider would be categorized as a dangerous stimulus just because the social environment is shaping one’s expectations of spider.
(Although one can argue that it would be natural – or innate – to fear spiders, I could think of a lot more instances of irrational phobias)
Is this the kind of “supervised learning” that Harnard was arguing about in his last article? Or is this another type of acquired categorization (language-acquired categorization)? Harnard wants “to show that it is a full-blown language effect, and not merely a vocabulary effect”. How is a vocabulary effect different from a full-blown language effect?
I would argue that a true phobia is often, like you said, irrational, but that it would be an example of categorization that is innate and can be changed with trial and error factors. Phobias are classified as irrational fears. So in a way, you are innately classifying something as "scary". But you do get feedback from your brain and nervous system that make you feel afraid when you encounter the object you are afraid of. Because it is an unpleasant feeling, you continue to feel afraid. So perhaps it is not a learned categorization, but it is reinforced by "trial and error". With therapy, these phobias can also be learned to be non-aversive. In a way, this is an example of categorization through trial and error. (Realizing that nothing bad will occur when you encounter it after repeated trials results in "re-categorization" of the target of the phobia as non-aversive or not scary).
DeleteIn terms of the vocabulary effect vs. the full-blown language effect, I do have difficulty grasping this too. But from how I understood it, the vocabulary effect has to do with the combining the Whorfian hypothesis i.e. colours are perceived categorically because they are NAMED categorically, with the ability to indirectly categorize abstract concepts that we do not have sensorimotor contact with (eg. the peek-a-boo unicorn) through only LANGUAGE or a description, and not just the named category.
“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?”
ReplyDeleteWe discussed in class last week how grounding can occur indirectly for categories that we do not experience directly. One example is understanding a zebra, which may understood by someone who has never seen a zebra as horse +stripes. However, this method would seem to be less ideal than sensorimotor grounding, since we all agree that a horse painted to be striped would be distinguishable from a zebra for anyone who has grounded horses, stripes, and zebras through sensorimotor experience. This situation leads me to respond with that even if the Whorf Hypothesis holds true, and categories can be acquired through language alone, it should be the second choice method for categorization. In other words, categorization through sensorimotor experience should always be one’s top choice in categorization. Another example to illustrate the limitations of categorization through language alone comes with our difficulty in categorizing intangible concepts like love. Although we all know what it feels like to love someone, many people struggle with romantic love in particular, because they are tasked with distinguishing when “liking” someone becomes “loving” someone. Now, liking-love can be appreciated as existing on a continuum, however even on a continuum, labels can be added to verbalize the different degrees. Romantic love is poorly explained to others, and many people simply concede that you need to feel it to know it, and more importantly to know why it is different than mere infatuation. Situations like this one demonstrate why even if categories can be acquired through language alone, it is a messier and second-rate way of acquiring categories.
I agree with you, and I would add that if you first acquire a category through language, it isn’t really felt so it isn’t perfectly grounded until you experience it with sensorimotor interactions. With language, to define a new category you can use already grounded symbols (meaning, you have already had sensorimotor interactions with symbols’ referents), however you can never be sure how the features the grounded symbols refer to actually interact when put together on one object. So you need actual sensorimotor experience with objects of the new category to perfectly ground the category. Experience can confirm or correct the hypothetical category that had been established by language only.
DeleteBut there is still the unicorn problem.
“The frog’s brain is born already able to detect “flies”; it needs only normal exposure rather than any special learning in order to recognize and catch them”
ReplyDeleteI’m always skeptical of the idea that evolution stamps us with innate abilities like this. I’m curious to learn more about Universal grammar and even more about our innate color perception abilities. I haven’t yet wrapped my head around how capabilities like this could be inherited in any way other than learning. Certainly, an organism must be born already equipped with the physical and cognitive capacity to identify flies or see certain wavelengths of light. I struggle to take it any further than that as it seems to discount the effects of culture, experience, upbringing, etc.
Harnad talks about frogs catching flies as an example of innate category detection. Does this example merely refer to the capacity of a frog to detect the category of flies, or does it assert that frog cognition comes pre-equipped with the ability to identify flies? Furthermore, what are the benefits of distinguishing between innate categories and learned categories in a given organism?
I am not sure either how the capacity of identifying a fly can be innate but I think taste would be a good example of innate categorization capacity. We have receptors for certain tastes and there are certain tastes we innately interpret as being correlates of harmful food. Is it what we call instinct?
DeleteI am asking this question because it seems like instinct could be a very good example of inherited, innate capacity of knowing how to interact with objects and that would prove that we have innate categories.
About the benefits, I would say that finding innate categories could help explaining how are formed the first categories formed experience. Especially, if ‘’when we learn to categorize things, we are learning to sort the alternatives that might be confused with one another’’. ‘Alternatives’ implies that we define category a by comparing it to category b which is not category a. Therefore, there must be at least a first category at some point.
I agree with Marion in terms of the taste thing! Except for I would say that "taste" although innate (whether something tastes likeable or not) also plays a huge role in learning (we only learn that we don't like brussel sprouts after we taste it and realizing that we find it unpleasant).
DeleteBut it does seem that there is an inheritance or genetic component to this innateness...which kind of suggests to me that at some point it could have been an evolutionary advantage to eat certain things and not others.
"Harnad talks about frogs catching flies as an example of innate category detection. Does this example merely refer to the capacity of a frog to detect the category of flies, or does it assert that frog cognition comes pre-equipped with the ability to identify flies?"
DeleteI have a feeling the reason Stevan put "flies" in quotes was to imply that the frog doesn't actually detect flies, but simply detects something like flies. Flies just happen to be most of the little dark things that fly by in front of frogs (they might make a false positive error if you threw a little rock). I doubt he means "identify", unless you mean it in the vaguer sense of "detect" (which I doubt by the context of your question). I think detection is implied by the fact that frogs react in a specific (categorical) manner to flies (whipping their tongue and eating them).
"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. What is varying along this continuum is voice-onset-time: the "b" in ba is voiced and the "p" in pa is not. But unlike the synthetic "morphing" apparatus, our natural vocal apparatus is not capable of producing anything in between ba and pa. So when I hear a sound from the voicing continuum, my brain perceives it by trying to match it with what it would have had to do to produce it. Since the only thing I can produce is ba or pa , I will perceive any of the synthetic stimuli along the continuum as either ba or pa, whichever it is closer to"
ReplyDeletealthough it is true that our apparatus is not capable of producing all sounds, isn't it nontheless interesting that some language's boundary differ. for instance, the b/p boundary in English is around 20-40ms conflicts with that of French which is around 0ms (e.g., Caramazza, Yeni-Komshian, Zurif, & Carbone, 1973)
i don't think anyone would argue that the physiology of French- and English-to-be speaker differs... still, interestingly what is perceived as a BA by English speakers (e.g., 20ms vot) is a PA to an English speaker.
another quick and (i think) interesting fact:
Deletealthough i agree that the strong version of motor theory may not have many supporters, there is some recent studies that have hints of motor thery… for instance, Yeung & Werker (2013; if you're interested ... http://mcgill.worldcat.org/title/lip-movements-affect-infants-audiovisual-speech-perception/oclc/843922489&referer=brief_results) show that infants' perception of sound is influenced by the current shape of their mouth. so even though they would usually prefer to look at the i-making face when hearing the sound i, when given a choice between a u-making face (note that's the french sound in dessUs). they actually stop showing this matching preference if their mouth is shaped to make the u sound…
even though i am not a supported of motor theory that is still quite impressive and potentially difficult to explain...
Discussing speech perception and creation, I am still confused as to how exactly the philosophers in question separate what is innate and what is learned. For example, I wonder what these philosophers would have to say about the way babies and children create speech sounds; for example, babies cries have been found to display marked differences between countries with different languages (ie a crying Canadian baby sounds different from a crying Japanese baby). This could be described as an innate, natural difference in how speech is produced, but it could also be described as learned, because infants hear the language sounds of their mother and their surroundings from inside the womb.
ReplyDeleteThis problem is perhaps more complicated still when you consider dogs; it has been shown that does perceive and create different different types of sounds depending on the county and language that they are immersed in. This leads me to wonder: How do the theories of innate versus learned categorization in language apply across languages, and even further, across species?
I think this is a super interesting point, and thinking about your example has lead me to believe that language is perhaps more learnt than innate.
DeleteI would argue that baby cries are definitely learned. If a woman who had been born and raised in Japan, moved to Canada, and had a baby that was only exposed to Canadian-English sounds, even in the womb, I'm guessing that that baby's cries would follow the Canadian-English pattern as opposed to the Japanese pattern.
Perhaps one could distinguish between language and vocalization? Or as having the capacity to learn a language as being innate, but having to act upon that innate capacity to learn to be able to actually speak a language?
There have been cases of children who were isolated from speech so never learnt how to talk (see: http://en.wikipedia.org/wiki/Genie_%28feral_child%29)
I agree that language is learned but its foundations are innate and I believe that was the point being made. It is the universal grammars we are born with that allow us to learn a language. Studies have shown that it is not just mimicking, that is to say learning, that allows them to understand what is right and wrong in sentence structure. For example, one study in particular examined children just starting to make 2 word combinations and presented them with nouns that should be preceded by "a" or "the". Most people use "the bathroom" more often than "a bathroom". Children were able to accurately match a determiner with a noun much like adults could and because of the variance in their responses, it was clear that this could not simply just be something they picked up.
DeleteLearned CP:
DeleteTo me, this is the most interesting section from a phonetic and broader cognition standpoint. Harnad is writing that it is possible for CP to be induced from learning alone. It makes sense that, “our categorically biased sensory detectors pick out their prepared color and speech-sound categories far more readily and reliably than if our perception had been continuous,” but I can’t help but consider second language acquisition here: if you look at French vs. English VOT, a native English speaker will have to learn (if they want to sound native-like) to minimize VOT on their ‘pa’ and ‘ba’. Wouldn’t this result in someone acquiring learned categorical perception? They are able to generate “weaker secondary boundaries” but they will still (probably) have a sharper discrimination for VOT length that matches their native language more.
I think that the role of learned CP in categorization could be to allow a cognizer to adjust their perception of new object/stimuli if necessary and if their innate CP is not efficient or accurate enough.
I really like Shrinkhala’s point about the difference between language and vocalization. I feel like this approach fixes a lot of issues I had in understanding innate versus learned language, because it makes a valuable distinction between the creation of sounds and the deliberate creation of purposeful, directed language. Surely some sort of sound creation is innate, as we are born with the ability to produce sounds, but I agree with the idea that language is different from this because you have to act upon your innate capacities in order to produce meaningful language.
Delete“Inputs that differ among themselves are ‘compressed’ onto similar representations if they must all generate the same output; and they become more separate if they must generate different outputs. The network’s ‘bias’ is what filters inputs onto their correct output category” (Harnad, 2003).
ReplyDeleteHarnad explains a possible computational model of categorical perception by illustrating the idea that inputs that are sufficiently different than one another will be separated from one another if they generate different outputs, but are arranged together if they generate the same (not similar) output. The brain’s “bias” to discriminate between these “things” is what filters inputs onto their correct output categories. This explanation of categorical perception makes intuitive sense to me, but I am wondering how the network’s “bias” is created and whether it is an innate feature of every individual or if it is the product of learning? It seems that we all tend to categorize things in the same ways in North American society, so it is perhaps the case that the discrimination process that goes on in our brains in something that is inborn. Also, I wonder which features of the shared members of a particular category are most significant for this discrimination process. Ultimately, I suppose what matters is the output, but it seems strange that the inputs can just be similar to be “compressed” into similar representations for a particular output, and not the other way around. Certain inputs, additionally, can create a variety of outputs, which suggests to me that the process may be a lot more complicated. Also, we still don’t know much about the brain mechanisms that could account for these processes, which is a problem in itself.
"But, as with colors, it looks as if the 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."
ReplyDeleteActually I am pretty sure our category detector for speech sounds is learned. Categorical distinctions along the speech continuum is based on the language and is learned as an infant based on the auditory inputs received. For example, if the infant's language contains 'pa' and 'ba' but does not contain 'da', then when the voice onset time of 'da' is absorbed into those of 'pa' and 'ba' based on whichever it is closest. Speakers of different language have different categorical perceptions because of the auditory environment they are raised in.
I think categorical perception also serves a useful purpose. It lumps together items that are very similar so that our brain can more easily discriminate--that way we do not have to pronounce a sound exactly the same every time in order to be understood the same every time. Because of this the computational model makes a lot of sense.
Categorical perception (CP) is defined as perceiving differences as “abrupt and qualitatively”. On the other hand, continuous perception (CoP) is defined as perceiving differences as “gradual and quantitatively”. (Harnad 2003)
ReplyDeleteI really find it interesting where one draws the line between these two different processes. They can be often conflated, and sometime even hard to untangle.
For example taking temperature as a CoP, yet still being able to distinguish the CP of when the water is cold or hot.
Another example of the is the McGurk Effect (https://www.youtube.com/watch?v=G-lN8vWm3m0)
Although we categorize the different sounds as ‘ba’ and ‘fa’, it is so strange to see that what the eyes perceive can have a greater dominance in terms of categorizing the sound. However, when one switches to hear the two sounds across the two different faces with different mouth movements, the two sounds almost seem as if they were working along some sort of continuous perception.
I guess this example somewhat ties in with motor theory of speech perception. The visual image one sees and the sound which accompanies it is what seems to follow a clear cut CP. However, when I closed my eyes between the two pictures to try to hear the sound as purely the sound, it felt as if ‘ba’ and ‘fa’ followed along CoP where one sound “morphed” into the other sound, something more like the “voice-onset-time”.
This comment has been removed by the author.
ReplyDeleteI really enjoyed reading this paper as it touched upon several interesting issues. However, I found the Whorf Hypothesis difficult to grasp. I initially assumed that this hypothesis was specifically about naming and not about categorizing. Referring to the example provided (i.e. color spectrum), two different cultures may have exactly the same category boundaries but they may name these categories differently. It would be absurd to think that one could reshape the color spectrum and its categorization through language alone.
ReplyDeleteTherefore, I feel as if the Whorf Hypothesis can be easily misinterpreted, in that one could easily assume that language was reshaping the categories. However, I think the Whorf Hypothesis was meant to illustrate how our mode of labeling changes across cultures but not necessarily our perception of them.
As a result, if the above is true, I don’t think we should automatically dismiss the Whorf hypothesis. It’s possible that our various naming decisions are in fact derived from our communities and societies but they may not necessarily be linked to a perceptual difference. Therefore, I feel that our dismissal of the WH may be based on a misconception. When the WH speaks of perception, it’s not talking about physically changing how we perceive things, but rather how we mentally perceive things and how our ideas can change based on language and other social factors.
“And what about the very building blocks of the language we use to name categories: Are our speech-sounds -- ba, da, ga -- innate or learned?”
ReplyDeleteWouldn’t it be great to spend some time in an infant’s mind? It would definitely make the lives of developmental psychologists a lot simpler. Instead, infants’ minds are black boxes and so we need to use indirect methods to determine what they are thinking. One of the most interesting things I have learned while studying cognitive science is what happens when infants start distinguishing phonemes.
Two languages can have their own typical voice onset times (VOTs) for a given consonant, and this is possible because VOT is on an acoustic continuum. All healthy babies start life with a fresh phoneme-distinction slate: whatever VOT distinctions between voiceless/voiced consonants with the same place and manner of articulation they hear are the ones they will eventually know, and those they don’t hear will never again be so easily learnable. This means that consonant distinctions (ex. pa/ba) do not have one universal VOT-switching point, but instead this switching point has to be learned. If an English speaker and a Spanish speaker were presented the same phoneme recording for which the VOT-switch point was different enough to have clear perception, they would hear different phonemes! But once you deviate further away from this switch-point, both will likely agree on having heard the same phoneme. It seems the innate human language system likes two clear categories for similar phonemes (pa/ba), but it is willing to be nudged up or down the VOT continuum based on what it hears from the external world.
"Some of our categories must originate from another source than direct sensori-motor experience, and here we return to language and the Whorf Hypothesis: Can categories, and their accompanying CP, be acquired through language alone? Again, there are some neural net simulation results suggesting that once a set of category names has been "grounded" through direct sensorimotor experience, they can be combined into Boolean combinations (man = male & human) and into still higher-order combinations (bachelor = unmarried & man)"
ReplyDeleteI feel that the Whorf Hypothesis answers some of the questions I had in my skywriting for the previous article we were assigned-how categories based on conception, such as conservative or communist, created?
"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: 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 color spectrum the same way, but even for those who don't, the regions of compression and separation are the same"
Could the Whorf Hypothesis be talking more about symbol grounding? I agree with the contemporary definition of categorical perception, which is that items within a category are more similar to each-other than they are to items in a different category. So even though most cultures see various shades of red as being closer to each other than various shades of yellow, different cultures could ground "yellow" as very different things.
"These language-induced CP-effects remain to be directly demonstrated in human subjects; 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. That is enough to rehabilitate the Whorf Hypothesis from its apparent failure on color terms (and perhaps also from its apparent failure on eskimo snow terms, Pullum 1989)"
ReplyDeleteThe example I will discuss is not strictly language-induced, but it's a categorization effect that's shaped by language use. (Does this classify it as a learned CP-effect?) According to the innateness theory -- which has been shown in newborns to be true -- we are all born with the same innate ability to categorically perceive, among other speech sounds, voiced and voiceless consonants based on their differences in voice onset time (VOT), but this membership criterion can be shifted and mediated by later language use. For example, English is known to have phonologically voiced but otherwise phonetically voiceless stops. (Background info: voiced stops have negative VOTs, and voiceless stops have positive VOTs.) In other words, what English speakers define as a voiced bilabial stop /b/ can (in certain words) have a VOT close to zero milliseconds (it may even be positive), which speakers of other languages might categorize as a voiceless stop. If you were to play the sounds of English /b/ and /p/ to, say, French speakers (who are used to /b/ with very negative VOTs and /p/ with VOTs around zero) may judge both of them as voiceless /p/ sounds. This CP-effect is interesting in that it's similar to what the Whorf Hypothesis assumes. The way people categorize the distinction between voiced and voiceless stops can be affected by how they use these consonants contrastively in their native languages.
Outlined in this article are examples of two sensory perceptions, sight and sound, and there follows an overview of the extent to which we can categorize or discriminate things that we pick up using these senses. What about the other three senses - touch, taste, and smell? I think we can learn a lot from how those senses work, particularly touch, to inform our research into sensorimotor categorical perception. Our skin contains sensory fields which, when used very often, grow smaller and more densely filled with nerve endings, and thus more capable of discriminating; it's why we can tell quite well if we are being prodded with a small object on one point on our thumb compared to a nearby point in the same general area; whereas along the length of our arm, we are less able to discriminate different places being touched, because we don't use that part of our body for tactile activity nearly as much. So this is quite an elegant and straightforward example of the difference between inborn and learned CP - we are born with the ability to perceive touch, but we learn, or more precisely, hone our ability to perceive touch through repeated use. Intuitively, it seems like the rest of the senses must have an analogous process of inborn / learned ability, and maybe language too should be classed in the same general category, though that remains to be seen.
ReplyDelete“But, as with colors, it looks as if the 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.”
ReplyDeleteThere have been studies that have shown that people with different first languages have different voice onset time boundaries, so when someone whose first language is English and someone whose first language is Portuguese listen to the same continuum of sounds ranging from extreme ‘ta’ to extreme ‘da,’ they will perceive the ‘ta’ changing to a ‘da’ at different times. I’m wondering how this fits in with category detectors for speech sounds being born already biased by evolution.
“There are even recent demonstrations that although the primary color and speech categories are probably inborn, their boundaries can be modified or even lost as a result of learning, and weaker secondary boundaries can be generated by learning alone.”
Does the first language you learn slightly alter where your inborn speech category boundary will be because you realize that the adults you’re getting language input from mean something slightly different when they say Ten vs Den, haT vs haD, Train vs. Drain etc.?
I'd also be really interested in hearing about cases where boundaries between colours have been modified by learning.
Harnad’s paper “Categorical Perception” deals with how objects are categorized, whether it is on a continuous scale or an all-or-nothing basis. The article deals particularly with the most basic categories perceived by our senses: colours and phonemes. These can be harder to distinguish than object categories. The line between blue and green is more blurred than the line between dog and cat, for example. I find the Whorf hypothesis very interesting. It states that colours are only perceived categorically because the happen to be named categorically. This turned out be not true, as Berlin & Kay showed that colours are perceived about as similar or different whether a given culture has a name for them or not. At the end of the paper, Harnad suggests we revive the Whorf hypothesis under the rubric of language-derived categories. While perhaps our senses are the most important when we are first learning to categorize as toddlers, language is much more important in our grown-up life. Many things are categorized according to social influence. Many people now believe gender, which once seemed to be a cut-and-dry category, to exist on a spectrum. Continuous perception therefore does not only apply to our senses when we look at them on a microscopic level, but to larger categories too on a macroscopic level.
ReplyDeleteThis comment has been removed by the author.
ReplyDelete''If motor production mediates sensory perception, then one assumes that this CP effect is a result of learning to produce speech. Eimas et al. (1971), however, found that infants already have speech CP before they begin to speak. Perhaps, then, it is an innate effect, evolved to "prepare" us to learn to speak. But Kuhl (1987) found that chinchillas also have "speech CP" even though they never learn to speak, and presumably did not evolve to do so. Lane (1965) went on to show that CP effects can be induced by learning alone, with a purely sensory (visual) continuum in which there is no motor production discontinuity to mediate the perceptual discontinuity. He concluded that speech CP is not special after all, but merely a special case of Lawrence's classic demonstration that stimuli to which you learn to make a different response become more distinctive and stimuli to which you learn to make the same response become more similar.''
ReplyDeleteI'm not sure I understand exactly. So, in the case of speech, motor production does not mediate sensory perception because even species that have not evolved to speak can percieve the differences that constituted evidence for the motor theory of speech perception? So how to link Lawrence's classical theory, that CP is a means to an end so that input differences are compressed if they need to produce the same output and separated if they need to produce a different output, with the the baby (1971) and chinchilla (1987) evidence? They are two examples of CP without apparent previous learning (except maybe evolutive in the case of the infants).
It is the distinction between evolved CP, as a compression/separation scaling factor who’s “effect is innate”, and learned CP, whose “boundaries can be modified or even lost as a result of learning”, which confuses me (Harnad, 6b, p3). Especially when I consider terms from reading 6.a that seem to be quite relatable, i.e. unsupervised learning and supervised learning.
ReplyDeleteIt seems as though evolved CP results from the innate properties associated with the biological genotypes that are naturally selected for, resulting in “biased sensory detectors” (Harnad, 6b, p4). Similarly, unsupervised learning seems to categorize sensory input based only on “internal competition” (Harnad, 6a, p6), as opposed to context based external interactions. Input can only be clustered according to how it is initially encoded, because there is no way for the animal to then interact, in a trial and error manner with the environment to account for certain inconsistencies related to context. On the other hand, the description of learned CP, from Harnad’s Categorical Perception (6.b), shows that there seems to be vast array of categories that cannot be inborn, and must develop through various learning processes. Here, my ultimate concern, and question, is whether both forms of CP fundamentally require learning, and if so, why is the distinction between evolved CP and learned CP made? It seems as though evolved CP still requires both unsupervised and supervised learning. In the case of unsupervised learning, the selected photoreceptors, ear structure, taste buds etc… discriminate categorical precepts purely based on the sensory input. Supervised learning also demonstrates this because humans, as well as more primitive organisms, have evolved feedback systems for contextualizing sensory input through trial and error. In the context of color perception for instance, trial and error learning might influence the designation of secondary boundaries of color categories such as magenta, and cyan. If evolved CP seems to evolve both unsupervised and supervised learning, does learned CP then imply a higher level form of learning, one that is not grounded in sensory motor compression and separation?
The main discussion here is on categorical perception which is defined as an effect "in which a perceived quality jumps abruptly from one category to another at a certain point along a continuum, instead of changing gradually."
ReplyDeleteI take this to mean that at some point along a continuum of sound or color perception, we decide that the category in which to classify the stimulus changes to another. This seems very black and white to me. If forced to classify colors into either green or blue, yes, this would occur, but seeing something in between these two categories would most likely prompt me to at least say that it is a "bluish-green" or a "greenish-blue" or even perhaps a more specific word like turquoise.
But the main example here was on sound perception. While yes, in speech we do tend to correct people's mistakes in speech production (in our minds) and make assumptions on whether something had a "ba" sound or "pa" sound at the beginning, the same concept applies as with the colors. Playing with the VOT (voice-onset-time) here and not forcing a person to decide whether it is one or the other may actually prompt the person to not be able to discern the difference. Context generally plays a large role in these distinctions. Say you do this experiment with “bat” and “pat” as the two ends of the spectrum, context would dictate which is more likely. Otherwise, I’m pretty sure I would just ask someone to repeat themselves...
Still, there do exist languages in which there do exist more than two bilabial stops. In these languages there may be one that has a very high VOT, a second that has still a positive VOT but closer to zero, and then a third with a negative VOT. The distinctions in such a case along this continuum are much more specific. A native speaker of such a language may have no trouble with this, but someone without a trained ear will still only hear two general sound types.
So the newer definition of categorical perception then is “CP occurs whenever perceived within-category differences are compressed and/or between-category differences are separated, relative to some baseline of comparison.” Parsing this sentence took some time. So CP occurs when you see two types of “tables” and still label them both as such. It also may occur when we use some sort of guideline to distinguish and differentiate between something like a french press and an electronic coffee maker? They both make coffee… This second part of the definition still confuses me a bit.
What are the implications of having this label of categorical perception? Is it actually how we perceive things or does it happen after we see something? When does the compression or separation occur? What aspects do we utilize first? Do we first give a general compressed category and then separate it by its distinctive features? Perhaps the compressed categories are more likely to be innate, while differentiating comes with experience in the world.
What i think is interesting here is that is almost dances around a notion that is mentioned briefly in “To Cognize is to Categorize”, which is the idea that in any moment of perception, there is an internal conflict between potential categorization models of the thing being perceived, and final categorization is given to the model that best fits. This explains the trainability of CP with regards to colour and speech, as certain categories are strengthened against natural categories, our innate categorization patterns can shift, especially at the margins. This also applies to language/combinatorial categorization. If these categories are built from perceptual categories, then, when presented with a stimulus, whichever constructive category set fits the perception best will likely trigger categorization into the category which they construct. This also explains why we usually don’t perceive someone as a “bachelor” and a “man” at the same time, but rather one or the other, with some sense of the other possibility as present, and an ability to switch our focus. The man category is activated, but if the unmarried one is active as well, and being factored in, the “bachelor” combined category will be stronger, and compression occurs into the “bachelor” category. If the mans marital status is not in consideration, the man category activates alone, and we compress our perception of him into the “man” category.
ReplyDeleteIn the paper Categorical Perception, Harnad focuses more on the perception part and language part of categorization, and I think this is where everything gets a little bit more complicated and harder to understand. I agree that some perceptions are innate like color perception: naturally our eyes see some colors as so different while others as very much close to each other, and we do not learn this, but were born with this ability. Does this implies that categorization may have some parts that are innate in us? We already know that categorization is not all innate; for some, we have to learn. But besides perceptions like color perception, when we look at words in dictionaries and see what categories those words belong to, is this process merely caused by learning and natural exposure or there is something to do with some kind of ability that we were born with? Is it possible that we are not born with the ability to categorize but born with the ability to learn categorization? If that is so, is it possible for computers to have our ability to learn categorization and then do categorization? In this paper, Harnad says that people are trying to model human brains and let computers do categorization. The result seems uncertain right now, but I believe computers can do a lot more than we think they can do, but there is something they are unable to do.
ReplyDeleteIn this paper, Harnad also talks about the language-induced categorization. This is the part that is confusing but at the same time very interesting. I was intrigued when we discussed the pikabu unicorn in class. Language seems to have a large power in our lives. Though sometimes we never see something or touch something, we know what that is. I have never seen Mars, but I know Mars is the closest planet to Earth or whatsoever. Language can induce categorization I think because we do not actually categorize on real objects, but instead we categorize on their representations. If the representation has a referent (the real object) in the world, we link them and are able to categorize them both. If we do not see or touch the referent, we only categorize the representation.
While reading this paper, a certain article by Frank Jackson came to mind. This is famously known as the knowledge argument, or Mary’s room argument. It is a philosophical paper based on the notion that if someone (Mary) spends her entire life in a black and white room and studies all the physical aspects of the colour red, she would still be missing the conscious experience of the colour and would therefore not know what the colour really is until she is exposed to it.
ReplyDeleteWith respect to categorical perception she would never attain the category of the colour red until she exits the room. Exiting the room she would be able to induce the category based on learning and sensorimotor experience and be able to compress her category of what red really is compared to members of a different colour category. I believe it shows one of this article’s main arguments, that “some of our categories must originate from another source than direct sensorimotor experience”. I think that once she exits the room she would still be able to correctly recognize that the colour red is a different category than other colours even though she has never had any sensorimotor experience with any colours. This would seem to be an innate capacity she would have for recognizing the different physical aspects of colour.
One of the questions in Harnad's paper was if “categories, and their accompanying categorical perception, can be acquired through language alone”. I believe this thought experiment shows that this is not the case. Even if someone was in the room with her and trying to describe the conscious experience of the colour red she would still be no closer to knowing what it is and how to categorize it differently than other colours. Yes she would know the physical differences between them in terms of wavelengths, but until someone actually presented the colours and allowed her to consciously perceive them I think there would always be something missing to her categorical perception.
I have a question that arose while writing a paper for another course. I would appreciate any input on it. I am writing about video game players' identification with characters during game play and have been puzzling over how the process of self-identification fits within the framework of cognition as categorization. I've read a few (fairly old/outdated) papers about self-categorization theory, or the idea that we categorize ourselves within social groups if we match the prototypical member of that group. Specifically, I've been considering the example of self-identifying as a leader if one fits the prototypical schema of a leader (having particular traits, actions, etc). Would self-identification in this sense be an example of the cognitive process of categorization?
ReplyDeleteSelf-Categorization
DeleteCategorization means doing the right thing with the right kind (category) of thing. Doing can be eating vs non-eating something, approaching vs avoiding something, or calling something "X" vs "non-X." The thing that is categorized could be someone else (that's Bob, that's Bill) or oneself (that's me at age 4, that's me at IGA last week). The categorization can be social (that's a citizen, that's a foreigner) and it can be biassed (("I don't trust people with turbans" or "I don't trust people with dark skin.") The categorization can be based on "mirror neurons" ("He is/isn't doing the same movement as me") or it could even be based on mirror self-recognition ("that's me in the mirror" and "that's you").
Now I suppose those components could add up to "self-identifying" as the leader or the follower of a group, since categorization just means doing the right thing with the right kind of thing.
But with categorization there's also always the distinct possibility that you're wrong.
In which case the corrective feedback (from the consequences of mis categorizing) will not be far away...
(Dictators don't have that problem, because everything goes their way until...
A propos, this (late Tuesday pm at Concordia) might be interesting not just to András...
Thanks for the response! Do you think that means that the process of developing a new identity is a cognitive process, i.e. practicing leadership skills and becoming more comfortable as a leader? It seems to me that you are donning traits and a self-conception that makes you closer to the prototypical leader.
DeleteThe concept of language-induced categorical perception is really interesting. It is the idea that we may have acquired certain categories not through sensorimotor experience, nor innate origins, but through language. As a result, these categories may actually warp our perception of the world.
ReplyDeleteI can think specifically of cases in which scientific language, which has the goal of bringing us closer to things as they are (nature), actually estranges us from our sensorimotor experiences, our perceptions. Friedrich Nietzsche argued this saying that categories are "the graveyard of perceptions" and to a certain extent, I agree. I think that sometimes words that cause us to (maybe unnecessarily) categorize the natural world, disjoints our holistic experience of things. But how can this be avoided? And why did language-induced categorical perception evolve? And how?