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.
Pérez-Gay Juárez, F., Sicotte, T., Thériault, C., & Harnad, S. (2019). Category learning can alter perception and its neural correlates. PloS One, 14(12), e0226000.
Pérez-Gay Juárez, F., Sicotte, T., Thériault, C., & Harnad, S. (2019). Category learning can alter perception and its neural correlates. PloS One, 14(12), e0226000.
Pullum, G. K. (1989). The great Eskimo vocabulary hoax. Natural Language & Linguistic Theory, 275-281.
I think the link to the text is broken. Is this the one?
ReplyDeletehttps://web-archive.southampton.ac.uk/cogprints.org/3017/1/catperc.html
Many thanks, Gabriel, fixed now.
Delete"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." Because CP stipulates that perceived within-category differences are compressed and/or between-category differences are separated relative to before categories were learned, does this imply that we are born with a sense of categories? Or does it simply mean that in the case of evolved CP we should compared compression/separation relative to actual size of physical differences rather than in relation to "before categories were learned"?
ReplyDeleteAmélie, there is innate CP (between-category separation and within-category compression) in color perception, facial expression perception and phoneme perception. Innate CP is dramatic, but rare. Most categories are not innate, they are learned. When the learning is difficult because the categories are confusable, and it’s important to get them right, the learning produces CP to reduce the uncertainty. Learning produces CP in neural net models of category learning too (supervised/reinforcement learning).
DeleteWith innate CP, the compression/separation is based on comparing physical differences with the size of perceived differences along a physical continuum like color wave-length. Equal-sized physical differences in wave-length are perceived as being bigger when they cross a category and smaller when they are within a category.
With learned CP, the compression/separation is based on comparing between-category and within-category discriminability before vs. after category learning. After learning, between-category differences look bigger than within-category differences.
An explanation for the change is that the categories are learned by selectively detecting the features that distinguish the categories. The learned feature-detectors sharpen the perception of the differences between members and nonmembers, reducing, heling to make the category “pop out.”
Could we say that evolution leaves what can be learned to learning rather than innately precoding it which is more costly and less flexible. The only thing that's genetically pre-coded it the capacity to learn. In this case would critical, temporary "use-it-or-lose-it" categories be intermediate cases? If so, the only thing that would need reverse-engineering (for T3 or T4) would be the learning capacity itself that would automatically including the critical period capacities.
DeleteMelis, Lazy Evolution leaves most categories to be learned in the two ( or in humans, three) ways we’ve discussed. The biologically “primed” or “prepared” categories, including the ones with use-it-or-lose-it features and critical periods, are sort of in-between, as you note. But I don’t quite see how that reduces the tasks for T3 or T4…
DeleteIs the primary mechanism of innate CP compression/separation effects? At what level does compression/separation operate for innate CP? Presumably the vegetative functions of the sense organs could have evolved to create compression/separation effects that delineate useful categories (such as the amplification of sounds within the frequency of typical speech by the tympanic membrane). Is this the case or do we only talk of compression/separation effects as beginning at the level of sensory processing for innate CP (as I imagine it does for learned CP).
DeleteI found the discussion of the ‘validity’ of the Whorfian hypothesis really intriguing. Even though this hypothesis was refuted some time ago, thinking of it in connection with categorical perception gives the theory another chance. We have established that CPs can both be learned and innate, and are sensorimotor effects. But there is caveat to this: many words in our vocabulary couldn’t have been categorized through sensorimotor interaction, especially more abstract concepts. These could be acquired through language alone, and from my understanding this could be done through the process of grounding that we have read about in previous papers. We combine words that we have categorized (and grounded) through sensorimotor experience using logical symbol compositions in order to categorize higher-order concepts. Through this process, higher-order categories inherit the CP effects (compression and separation) of the combined words.
ReplyDeleteIn the closing sentences of the paper, it is mentioned that this effect should be demonstrated as a language effect rather than a vocabulary effect, shifting the belief that our perception of the world can be influenced by the way things are named (Whorfian hypothesis), but also by what we are told about them. Does this mean that we want to demonstrate that the way we perceive things in the world is impacted by the different grounded categories we put together though symbol compositions when grounding higher-order categories?
I would also like to further understand the link between the processes of categorization and of symbol grounding. Are these processes equivalent, or is categorization just one use of symbol grounding?
Harnad (2003) discusses categorical perception was helpful for my understanding. Categories can be categorical or continuous, and the discussion revolves around the inborn and induced learning capabilities. Mathilda’s insightful comments further helped me think through the validity of Wholf's hypothesis.
DeleteInterestingly, because both innate and learned categorical perception are dependent on sensorimotor effects, the difference lies in when the experience was had- for learned CP, life of the individual, and for innate CP, during the lives of the individuals in the species. So it seems that in the life of an individual, the way we perceive things in the world and interact with them (categorization) is probably due to innate aspects of vocabulary in addition to learned grounded language combinations.
It seems to me (I fully might be off base here, though) that symbol grounding is a necessary but not sufficient condition for categorization in the case of language and words (in a similar way that grounding is a necessary but not sufficient condition for meaning to be attached to symbols).
"…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."
DeleteHearing that our perception of the world can be warped, based on what we are simply told about things, is a daunting idea to accept. This is what is said to be required to show that it is a 'full-blown language effect' rather than mere vocabulary. To ‘warp our perception of the world’ may not be as momentous as it sounds, but I do not know how significant or strong of a change something must make to be considered a ‘warp’. For example, if I am verbally told that all of the world’s apples are toxic and thus no longer edible, maybe my perception of the world will now be warped. Would something of this magnitude be considered 'what we are told about' the thing or is this still an instance of 'how said thing is named'? If it's the latter, what would qualify as warping our perception of the world based on what we are told about them, since if we are told something contrary to what we have previously believed, are we not also altering the category of which that thing identifies?
“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.”
ReplyDeleteThis section made me think of natural selection and the evolution of species, as some baselines have endured the test of time and some have been man-made or developed later. Certain categories are physical and undeniable like blue and green, that depend on fixed wavelengths, or living versus inanimate, that is discerned by the ability to breathe. However, in the case of learned CP, such a discriminability within and between categories could only develop through trial and error, through adding new instances to each of our previously existing categories. That would mean there is a common category that derived into two, and then into many. Possibly the first category was “matter” and that the first category distinction was “living” vs “inanimate”, and so on. This would reflect evolution in that we all derive from a common ancestor, that then derived into new species.
Tess, I couldn’t quite follow your comment. We learn what’s called “green” and what’s called “blue”, but that’s easy because we have innate feature detectors for these categories already; we just needed their (arbitrary) name.
Delete“Living” vs. “inanimate” is a bit harder (the feature “breathing” won’t be useful with insects or plants), but the features are learnable (by trial and error, or with a verbal description of features that do work).
I don’t know what you mean by “the first category was matter”. There are an infinity of categories – those we actually learn and those we could potentially learn (but may never need them to “do the right thing”).
Yes, there are many ways to subdivide categories we already have, or super-join them into higher-order categories. And every sentence we speak, with subject and predicate, is describing another category by recombining features. for which we already have a category name, to define or describe further categories. (That includes the preceding sentence: Find its subject and its predicate. The members of the subject category are contained in the predicate category.)
PS What is "categorical perception" (CP)? And what's it's relation to category learning and categorization?
Delete
DeleteMaybe I could take on this question. Categorical Perception is the abrupt perceptual change at the boundary because its membership is either all or none. Concerning its relation to category learning, category learning can induce CP. For example, within-category compression and between-category separation can occur during category learning. On the other hand, categorization is doing the right thing with the right kind of thing, and we use it when we want to identify a shape in isolation. CP occurs when we need to make absolute discrimination(i.e., categorization).
Nadila, good, but what is the "boundary" if a category is not on a continuum, like "rabbit" vs. "hare"? And how might CP help us categorize?
DeleteTo my understanding, if the boundary is unclear, such as in the case of rabbit vs hare, then further trial-and-errors with corrective feedback, meaning supervised learning, are needed to learn the categories. The acquired distinctiveness further contributes to our categorical perception of these categories, enhancing within-category compression and between-category separation. I believe this is how the expert chicken-sexers learned and reinforced their understanding of the male and female chicken categories.
DeleteJenny:
DeleteI agree with your point that category learning, in this case, involves supervised learning -- in that the finer distinctions between a "rabbit" and a "hare" are acquired through repeated trial and error. I just wanted to add to your point that the distinction between a "rabbit" and a "hare" can also be learned through language-induced CP. It is accepted that both innate and learned CP have sensorimotor effects. However, in real life, some of our learned categories originate from sources other than direct sensorimotor experiences. Professor Harnad mentioned in his paper that "once a set of category names has been "grounded" through direct sensorimotor experience, they can be combined into Boolean combinations." So, in this case, if I have already had direct sensorimotor experiences as to what "rabbit" refers to; a "hare" could be a higher-order combination of a rabbit with "longer ears" and "longer legs" (granted that I have already had direct sensorimotor experiences as to what "long," "legs," "ears" refer to).
This reading emphasized the perceptive aspect of categorization. Categorical perceptions are innate or acquired through learning. They serve to sort out elements into categories. Some sensorimotor perceptions however are both categorical and continuous as they can be true at both levels. I found this idea interesting as in principles categories stem from this idea of differentiation and separation between elements. I found the Whorf Hypothesis interesting as it made me think of the symbol grounding problem. It states that our subdivisions are arbitrary, learned, and vary across cultures and languages. The proof to refute this hypothesis is that other languages subdivide and name their objects the same way. However, having categorization ways similar could potentially still make them arbitrary.
ReplyDeleteThis reading was fascinating on how categorization is learned and can vary depending on culture. Very much agreeing with what you wrote Ines! Even with the computational models being what we think the brain is doing, we are getting closer to understanding what it is doing for CPs even if we were heading in the wrong direction.
DeleteInes, I was a little lost with your comment too:
DeleteFirst, what is CP?
What are “elements”? Do you means distinguishing features allow us to do the right thing with the right kind of thing? That’s category learning and categorization, but it’s not CP.
Some category-features are discrete and some are continuous, but categorization is categorical (correct or incorrect). See other replies about cognitive capacities that are continuous (e.g. swimming),
Color categories are innate and universal, regardless of language or color. But most categories are not innate or universal.
Helen, Kid-sib can't understand what you are saying. Please read the other comments and replies on this and the prior thread.
This reading was interesting in that it elaborated more on the development of and questions surrounding categorical perception. In it, Harnad writes "If there is significant within-category compression and/or between-category separation, this is operationally defined as CP." Are the compression and separation effects separable? What are cases of categorical perception in which a separation effect exists without compression or vice versa? It states that "CP can be induced by learning alone," which leads me to wonder if the compression and separation effects of categorical perception varies at different stages of learning a category of if it only arises when the category is fully learned. Finally, it is mentioned at the end of the reading that categorical perception in higher-order categories has not yet been demonstrated in humans. How would it be experimentally tested? Has there been research on this since the publication of the article?
ReplyDeleteElena good questions. Since the purpose of categorization is to do the right thing with the right kind of thing, you have to learn to distinguish between members of different categories. So between-category separation is much more important than within-category compression. If you need to learn to distinguish rabbits from hares, it doesn’t help to see rabbits as more similar than they looked before the learning: what helps is to see rabbits and hares as looking more different than before the learning.
DeleteIt may be more accurate to describe the CP effect as between-category differences coming to look bigger than within-category differences after the learning.
The question about whether CP builds up as learning gets better is a good one, but the experiments are not yet fine-tuned enough and the results are statistical. Something like what you said is probably true, because in a population of learners, the size of the CP effect is correlated (positively) with the average percent correct performance per participant.
One way to study higher-order categories is to train the lower-order categories first (by supervised learning), say, A, B, C, and D, and responses Ra, Rb, Rc and Rd. Once that’s learned and you’ve measured CP, now train two new Categories, P and Q, so that the correct Responses are Pa, Pb, Qc, Qd: Are members of A and B further from C and D after the second training than they were after the first?
This is probably much easier to test first with neural nets rather than human participants.
As you mentioned, between category separation is more important than within category compression. Additionally, the reading mentions that CP occurs whenever perceived within-category differences are compressed and/or between-category differences are separated, but that this is relative to some baseline of comparison. This baseline could be the perceived similarity or discriminability within and between categories between categories are learned, compared to after. I am assuming that in your rabbit-hare example, the baseline here applies to the between category differences looking bigger, or more discriminable than the within category differences after the learning, which is indeed what makes it CP.
DeleteAriane, that's it.
DeleteIt's interesting that language has this CP nature due to the mental 'process of elimination' we do when hearing speech. What we hear can only be a specific thing based on how that person could have produced it, and if we can see their lips moving, it becomes more obvious, as demonstrated in the McGurk Effect. Again, if language has this categorical nature, it must have been an evolutionary advantage that allowed us to communicate more effectively. I think there are some innate aspects to language such as UG, CP, and even facial expressions as mentioned in the piece.
ReplyDeleteAlexander, but the main evolutionary advantage of language itself was not CP for phonology or UG for syntax, but the 3rd way of learning categories that language made possible.
DeleteI believe how words are compounded and folk etymology are powerful of 'warping' the perception of the world. E.g., violins are xiao3ti2qin2 (small-handheld-Qin) in Mandarin Chinese, organs are feng1qin2 (wind-Qin), and harmonicas are kou3qin2 (mouth-Qin), so naturally, by name, they can be collectively categorized as qin. Meanwhile, there is no same compounding structure in English for these three musical instruments, so they are likely to be categorized differently from each other.
ReplyDeleteHan, yes, but it would have worked almost as well if you just called them X, Y and Z. The extra cue from word morphology or even iconicity is minor, and hardly has an effect in discourse or thinking once the word is known. Arbitrary shapes are just as good. “Violin”, “organ” and “harmonica” are fine. The relevant features are not in the word but in the object.
Delete"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" (p. 3)
ReplyDeleteCould someone help me understand what this passage is saying? Is the claim here that inborn categories such as "ba" vs. "pa" – for speech perception – and "red" vs. "green" – for color perception, can be modified or even overridden by learning? For instance, could you train someone to perceive "ba" and "ga" sounds as being within a single (continuous) category, or certain "ba" sounds as belonging to the "ga" (discrete) category? If that's what's being claimed, it's quite remarkable!
Gabriel, the boundary between adjacent colors in the color wheel can be moved a little by repeated training, but it might just be a short-term habituation effect that returns to its original value unless the frequency change is permanent. If, for example, the color difference between green and blue grapes were moved a little further into the blue range, so that green grapes were more bluish green (AND if those were the only blues and greens there were in the visual world).
DeleteBut with phoneme boundaries the effect can be more dramatic. All babies are born able to produce and perceive RA (rolled R, not the English vocoid) and LA. There is a phoneme category boundary between the two, for both perception and production.
But if, in the critical period, one child only hears and speaks Mandarin (which uses only LA and not (rolled) RA), and another child only hear and speaks Japanese (which uses only (rolled) RA, and not LA), then the monolingual Chinese child will lose the RA, and will be able to hear and say only LA. And the monolingual Japanese child will lose the LA, and will be able to hear and say only RA.
Zhang, Y. (2016). Categorical perception. Encyclopedia of Chinese language and linguistics.
From the responses and article, could we summarize that even though both color and phoneme CP have critical, "use-it-or-lose-it" periods, the main difference between them is that phoneme CP are able to perceive and produce phonemes, while we can only perceive colors. But this is more than just sensorimotor associations, especially in the case of learned categories vs inborn ones (where mirror neurons are used).
DeleteMelis, sounds right. I couldn't quite understand your last sentence...
DeleteThis reading goes into the various theories of categorical perception, and discussions as to their nativity/variability. The discussion of the Sapir-Whorf hypothesis and language-induced CP, which is applicable to only some kinds of categories, leads to the problem of how CP actually filters our reality.
ReplyDeleteIf direct, innate categories like colour and face recognition are not subject to Whorfian rules, but our higher-order CP might be, we need to find out what 'the way we see the world' is, and how it can change. However, what language users mean when they say words is hard to put in a universal and manipulable form.
We can establish the validity of Whorf's hypothesis is possible via reverse engineering, but for humans, it's more qualitative and cultural. The genesis of linguistic categories is definitely social. It seems that this takes us back to the hard problem of consciousness, and some sort of phenomenological explanation which is rejected by CogSci. Perhaps via supervised learning in which realistic constraints are imposed, we can get closer to how humans use cultural/linguistic givens.
Jacob, learned CP effects can be induced by supervised learning of unfamiliar visual patterns.
Delete Pérez-Gay Juárez, F., Sicotte, T., Thériault, C., & Harnad, S. (2019). Category learning can alter perception and its neural correlates. PloS One, 14(12), e0226000.
From my understanding of reading 6B, the Whorf Hypothesis claims that language and the arbitrary labels we use to name things like colours is what allows us to perceive them categorically. Essentially, colours all lie on a continuous spectrum but these arbitrary labels allow for within-category compression/between-category separation. However, we can run into trouble with categorical perception when there are languages that label the same things differently…
ReplyDeleteAs someone who speaks four languages, I instantly thought about the difference between languages that assign genders to objects and those that don’t. Based on the Whorf Hypothesis, could we assume that people who’s native languages label things with a gender would experience different categorical perceptions compared to those who speak a language like English?
Anais, from my understanding Whorf Hypothesis was actually refuted many times. “Berlin and Kay 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.” I also speak 3 languages and I do encounter times where as you said, different languages name the same thing differently, however I believe that the nomenclature difference does not imply that those words are categorized differently. It would be ingenuous to say that culture does not play a role in categorization, since it clearly does but I think that it is important to distinguish language from culture and although culture influences how language is used, I do not think that language translation makes the same categories to be perceived differently.
DeleteAnaïs and Vitoria, color categories are innate and perception of colors is the same regardless of how your language divides the spectrum. But what is learned CP? Please read the other comments and replies, and re-read 6b.
DeleteMy apologies– I went back and re-read the reading as well as the other comments and I think I understand where I was getting it wrong. To make sure I’m understanding correctly, CP for colour perception and phoneme perception is rare and INNATE. This means that the sensorimotor detectors that we’ve been equipped with naturally help compress/separate the physical differences within/between categories. Learned CP, however, is when this compression/separation happens after learning– essentially learning will help make between-category differences look bigger than within-category.
DeleteI think my original question was trying to relate to the idea of language-induced CPs, and that’s where my confusion lies. Again, from my understanding, language-induced CP explains those categories which we may not have had any sensorimotor history with. Essentially, the theory is that if a category label has been “grounded” through sensorimotor interactions, then they can be combined into those Boolean combinations that are mentioned in the reading, and those combinations create those higher-order categories which still inherit the CP. This is where I bring up my original question again– in the case of languages that associate genders with category labels, how does this effect the Boolean combinations and higher-order categories? Would it even affect it all? I hope this makes more sense, and apologies for the lengthy response.
Anaïs, good confirmation of understanding. Now your question about gender:
DeleteA verbal definition of a new category tells you the members’ distinguishing features, making it unnecessary to learn them directly through a long series od trials, errors, and corrections.
The category’s features are described by grounded words, words that name categories you have already learned (objects, features, states) for which you already have feature-detectors.
Category names are arbitrary: “rabbit” does not look or sound like a rabbit. It may as well have been “gavagai” or X. It does not matter that it has two syllables and starts with “r” (in English).
“The MOON is a round or crescent shape in the sky” is a (very approximate) description of “moon” for someone who does not know the referent. It saves having to learn it by trial and error from many examples and feedback (correct, incorrect).
Unlike “hen” or “mare,” It is not a feature of the referent category of “moon” that «la lune» is feminine in French and “der Mond” is masculine in German, and has no gender in English. It is just a syntactic feature that allows you to find the right pronoun to refer back to it in a sentence: “When the moon appeared, IT was pale” «Quand la lune a paru, ELLE était pale» “Als der Mond erschien, ER war blass”
Language has a lot of faded iconicity and dead metaphors that do not affect reference or categorization. In English, “person” is gender-neutral. In French, «une personne» is feminine and «un individu» is masculine. So we say «une personne est entrée et ELLE a fermé la porte» but «un individu est entré et IL a fermé la porte». No native speaker thinks it’s a woman in the first case and a man in the second – though an anglophone learning French might make that mistake, or a French child, when first learning pronouns.
Syntax sometimes tangles with semantics. There is no literal, word-for-word translation for «une personne est entrée et ELLE a fermé la porte» in English (try it!), In such a case, English uses “he” in its “unmarked” form, which is generic, ungendered “he”, or the (slightly incorrect, but changing, because syntax can evolve) plural: “they,” which violates number. On the other hand, French, unlike English (whose 3rd person plural is gender-neutral) has to use «ils» generically when trying to say that a woman and a man came into the room and THEY sat down.
These syntactic quirks have since been politicized, with increasingly silly results. But that silliness is trivial in a word of anthropogenic climate change, poverty, war, exploitation – and anthropogenic animal agony.
Thank you for the detailed explanation! The examples you provided definitely helped answer my question.
DeleteThe “motor theory and speech perception” section of this reading reminded me of a linguistic class I took where there was a lecture specializing on speech perception. There is actually a category boundary based on voice onset time between the “ba” and “pa” sounds (the greater the voice onset time, the more likely the perception of “pa”), and contrasts between these phonological categories involve acoustic differences. Although it is possible to create a sound exactly in the middle of the category boundary of “ba” and “pa”, our perception is such that we perceive one or the other.
ReplyDeleteIn a perception experiment, a continuous manipulation from “ba” to “pa” was performed and participants were asked which syllable they hear. Since the manipulation is continuous, one might expect a linear pattern between voice onset time and % of hearing a “pa”, if perception is gradient. If perception is categorical, we would expect a sudden rise in the probability of perceiving a “pa”. The observed pattern was a sharp boundary between “ba” and “pa”, which is evidence that linguistic contrasts tend to be perceived categorically.
Alara, that was where the term "categorical perception" first appeared. But what is the "motor theory," and what is the connection?
DeleteThe motor theory is that the way we hear speech sounds is influenced by how we produce them when speaking; that’s why we can perceive an abrupt change between different sounds. But the new/contemporary definition of CP, which is within-category compression and between-category separation, no longer depends on motor theory.
DeleteI have a question regarding the concept and CP. How does categorical perception influence concept(mentally processed idea or notion), or is it the other way around — the concept that one has learned influence one’s CP? Or is it an interrelationship between concept and categorical perception?
Nadila, that explains the motor affordances that separate and distinguish ba, da and ga. There’s CP for vowels too, and the path from aa to ee to oo is more like the color continuum (except it has the perception/production mirroring).
DeleteWhat do you mean by “concept”? It’s another weasel-word, like “idea” and “notion.” “Mind” and “mental” are weasel-words too. “Mental processing” means cerebral processing, since our brains are doing it. But we don’t know how (till cogsci reverse-engineers it), so it’s not really “mental,” like doing long division in your head.)
Yes, CP influences how we perceive. It also influences how we categorize (helping categories “pop out” by abstracting and highlighting their distinctive features). Or maybe you mean social or gender biases (mentioned in some of the other replies)?
I was wondering about the "pa" and "ba" thing as well. By signal detection theory I would expect that sometimes people would mistake a "ba" for a "pa" or vice versa due to noise (random activity of the nervous system). But that doesn't seem to be the case. It's very interesting because this does suggest some unique qualities about language perception. Is this line so distinct between an aspirated "pa" and an unaspirated "pa"? Because, though they are distinguishable, this would be a "within" category difference rather than a "between" category difference such as "ba" vs. "pa". So my question is would we see more of the typical signal detection theory errors with the 2 "within" categories?
DeleteTeegan, from the signal-detection theory standpoint, “d-prime” (detectability) is greater across the ba/da (voicing) boundary than within. Same should be true across the subcategory boundary within pa boundary for aspirated vs unaspirated pa. It’s just that the d-prime for ba/pa would be bigger than for pa-asp1/pa-unasp.
DeleteThe last section, on language induced CP asks whether language alone would be enough to learn new categories. My instinct on this is it is not true. To a certain extent, the category must be grounded. For example, how do we come to learn what a unicorn is if it does not exist? We can conceive of a unicorn as a horse + horn + magic +...+ not real = Unicorn. It takes a certain number of previously grounded sensorimotor categories to create the abstract conception of a unicorn. In this case, Unicorn “compresses” certain grounded characteristics while also “separating” them into a distinct category. This would not be possible with language alone, as it would lack any significant meaning. If we were not able to have a sensorimotor conception of a horse, we would never be able to have an abstract conception of a unicorn. Even though we can never interact with a physical unicorn, we can distinguish it from other sensorimotor categories because we are able to ground said other categories in the physical world. Categorical perception is very related to the symbol grounding problem which I did not realise until reading this paper!
ReplyDeleteSophie, so what it the relation between CP and symbol grounding? (Read the other replies before replying.)
DeleteCP occurs when there is significant within-category compression and between-category separation and is a result of category learning, which can occur through both supervised and unsupervised learning. Through (1) unsupervised and (2) supervised learning, symbols (instances) that we use to create relations between categories and within their members are grounded in sensorimotor experience, as (1) exposure and (2) corrective feedback from the consequences of doing the right or wrong thing with the object requires the direct interaction with that object. However, arbitrary instances that we do not have direct sensorimotor interaction with (ex: "unicorn", "beauty"), are based on verbal grounding. Verbal grounding, or language-induced CP is the process of acquiring categories by language alone, whereby categories can be grounded indirectly through combining propositions that we have had past sensorimotor interactions with.
DeleteSophie, that section was also of particular interest to me. While I understand it may be possible to categorize through verbal instruction and in the use of language, I believe it will be difficult to say whether there is actually that compression and separation happening or is it just the distinction in the arbitrary names/symbols that we use to refer to things. Maybe that’s where that disconnect with grounding may play a role in actual CP.
DeleteI also found the last section of the paper very interesting. Although it asks, “Can categories, and their accompanying CP, be acquired through language alone?”, it raises the example of a related neural net simulation that makes the assumption that a set number of base categories have already been grounded. I would agree with Sophie and think that language exclusively (without any grounding) would not be enough to learn categories. Using ungrounded categories to verbally define other categories would not result in meaning. Instead, it is defining arbitrary rules combining symbols to create more meaningless symbols. Trying to find the meaning of any of these symbols would lead to an infinite regress because nothing has been grounded.
DeleteThe question of how much categorical perception can be learned through language is still interesting to me though. The neural net simulation (as mentioned previously) results suggested that categories can be combined and inherit CP effects from its base categories. The results also suggested these combined categories generated their own CP effects as well. It would be interesting to see how the separation and compression of categorical perception differs in magnitude through learning a category through language compared to direct grounding. It would also be interesting to look deeper into how these combined categories generated their own CP effects if it is not obviously inherited from the base categories.
Sophearah, good experimental questions.
Delete1 category context described was based on degree. Specifically, there is a spectrum that can be made in which things can be classified on a basis of this continuous spectrum. For example, a a person who is 7 feet is classified as tall, and a person who is 6 feet, although “less tall” is still classified as tall. Hence, all in all, both items would still be in the category of tall. This is what can be seen as a continuous.
ReplyDeleteThat said, I was wondering with this concept in mind, if it is reasonable to connect it to the Prototype theory. This theory works on the premise that a “prototype is just the general representation of an object/concept we have”. Here, recognition becomes built on whether what we encounter match the general idea of that prototype. For example, if we picture what a “general prototype” of a bird would be, even if the features such as colours arent the same, one can still say that a robin is classified under the category of bird. Furthermore, an owl which is bigger and quite different to a robin would still be categorized as a bird. Thus, surely it seems that the prototype theory has a role to play in categories and categorization?
Maira tallness is a matter of degree, so the “boundary” between short and tall is arbitrary, and also context-dependent. A tall person is short for a tree…
Delete“Prototypes” were discussed a little in prior threads (q.v.). They are not a very effective way of learning categories: A potential mechanism is feature-detection and abstraction by neural nets through supervised learning.
(“Representation” is a weasel word if we want to know how organisms are able to categorize things.) We can categorize owls as owls and robins as robins, and both as birds. But how? (In answering “how,” remember to keep in mind cogsci’s task of reverse-engineering the mechanism that produces our capacity to DO what we can do.)
Based on what you said, in this case of categorizing based on birds, owls, and robins, and referring back to the cogsci’s aim of reverse-engineering - does Quillan's semantic memory organization pose that same uselessness as the prototype theory?
DeleteWe have units (which are the main categories such as a bird) that can point to more specific units (robin and owl). Then you have the properties which can I guess be the features (wings, feathers, etc.). Hence, to really distinguish each object/concept is a matter of how far these units become in terms of response time.
For example, referencing an example from my cognition class by Dr. Sheldon: “Can a canary sing? Vs. is a canary an animal”. In this example, ‘can a canary’ is easily accessible compared to ‘is a canary an animal’. Hence, back to the whole connection before I stray too far away - when we categorize would this concept of what is easier to access in our brain when it comes to categorizing another useless concept? That is, we still would not know how the capacity of searching across networks to find closer relationships to concepts really works?
Maira, whether you use discriminability (d-prime), as in Teegan’s example above with ba/pa and pa-asp/pa-unasp, or reaction time, as with bird/fish and robin/owl, the performance data can be useful in testing the fine-tuning of your category-learning or categorization mechanism, but I don’t think Quillian’s model provides such a mechanism, as far as reverse-engineering and TT are concerned. Or maybe the way to put it is that Qullian models the fine-tuning of the capacity without modelling the capacity itself.
DeleteFrom this paper I understood that there are two types of categorical perception. One is rather uncommon, it is innate and present in color perception, facial perception, and phoneme perception and the other is induced by learning, by discriminating the different features that make up a certain category.
ReplyDeleteColor perception is said to be an innate principle and since it is not learned, I believe it cannot be a centralized process but rather a question of personal matter. Indeed, a shade of red might appear to be closer to crimson to me but closer to blood red for someone else. In this case, one could argue that we all have slightly different perceptions of colors.
This leads me to wonder how machine learning models can learn to distinguish colors. Surely there are some intelligent algorithms that can discriminate amongst the different colors. But how could these algorithms learn, through supervised learning, what the correct answer is if we all have different accounts for the correct answer? In other words, I could tell the program that it has correctly classified this color as dark blue when another person would have told it that it has correctly classified it as light blue. Color classification might be a learnable process to a certain extent, but I don’t understand how we could explore this highly subjective environment in a supervised learning setting for machines.
Étienne, for the “basic” color categories (the ones that “pop out” in the rainbow: red, yellow, green, blue) the identification and discriminability data seem to be the same across all languages and cultures tested. (So it’s safe to use them to train a neural net to categorize colors.) There might be some variability for some of the subcategories (crimson, scarlet), perhaps because of frequency of the color, perhaps because of the name (and this too might be a learned CP effect). There are also no doubt individual differences in color perception, as in every other biological as well as cultural trait.
DeleteHello Étienne and Professor, I find it important to note that the distinguishing features of categories (such as colour) are approximate, which leaves room for personal differences. Colour is innate and as Prof Harnad says, the difference between red, green and blue are pretty obvious, but more subtle differences between colours can absolutely be learned (and still allow for personal difference) thanks to brain plasticity!
DeleteFor the colour-differentiating algorithm, a few sky readings have pointed to the fact that a baseline (or reference point) would be needed to first teach the machine what light vs. dark blue is, by showing it many instances of blue and providing feedback on its brightness over several trials.
I found the Whorfian power of naming and categorization to be very intriguing. The paper illustrated examples of the categorization of colours and how it may be influenced by language and culture. Although, I wonder whether this idea can also be applied to other sensory inputs, like sound. Say I am given several different sounds of birdsongs to listen to, and I am asked to pick out which one is from a robin versus a blue-jay. As someone who is not highly knowledgeable in bird sounds nor has been exposed to such sounds in my life, chances are it will be much harder for me to distinguish one from the other. This seems like an intuitive idea, categories such as bird sounds may be influenced by experience and exposure to such sounds in one’s environment, and certainly by learning.
ReplyDeleteAdditionally, this could also be applied to our other sensory inputs, such as taste or touch. A chef who specializes in a particular cuisine may be better at identifying specific flavour palettes or spices in a dish. They may have a more nuanced exposure to certain flavours, but not everyone may be able to distinguish them. The overarching point from this thought is that the power of naming and categorizing should not be dismissed so quickly in how it influences how we see the world and either restricts or enhances our ability to understand it.
DeleteSara, what is learned CP? It is likely to happen in identifying bird-calls, and wines, and textures too.
DeleteFrom what I have gathered, learned CP occurs when the instances of different categories come to look more disparate and/or instances of the same category come to appear more related after a new category has been "learned". Both of these forms of categorizing members are stemmed from experience and information that is learned based on one’s perception of it rather than an innate understanding.
DeleteWhen you say that you haven't been exposed to the birdsongs of robins or blue jays, I take you to mean that you haven't been exposed to them in a context where it is relevant for you to focus on the stimulus nor to categorize it as one or the other (which is probably the case for most of us). This got me thinking about the role of attention, which has definitely shown to be important for learning, and whether that is a basic-level factor that a theory of learning, categorization, and cognition should explicitly account for, or whether it is already accounted for in a notion like affordance which already may narrow down the range of stimuli we consider to those that are actually relevant for our interactions with objects and the world as we go through our lives.
DeleteThe Whorf Hypothesis - about the fact that the differentiation of colors is probably not innate and depends on the culture we grew up in - is very interesting. Discriminating different shades of blue seem quite difficult if we never learnt to do that. I someone asked me to point out a “royal blue” and an “electric blue” among 15 different shades of blue, I don’t think I would be able to do so because I never learnt that. However, I am wondering if someone who works around colors might be better at differentiating different shades of a color… I can imagine that an artist is probably better than me at discriminating different shades of blue. Therefore, categorical perception can be innate, but it can also be induced by learning.
ReplyDeleteHi Charlene! When you say you would be unable to point out "royal blue" or "electric blue" - is this because you cannot distinguish between these categories or because you do not know the label for them? Artists may not be able to distinguish shades visually better than the average person, they may just be aware of more words and labels that they can apply to them. This does not necessarily speak to their actual perception. Also, if I am understanding correctly, the Whorf Hypothesis is more applicable to between-category differences (how did we decide that greens form a category and blues form a category?) rather than within-category differences (how do we distinguish shades of blue?).
DeleteHello,
DeleteI think in cognitive science when we talk about color perception it is usually used as an example of innate category. Newborns with normal color perceptions are just born with the ability to tell that red is not blue and yellow is not green. Because human visual system have color detectors to detect distinct features of colors (in this case the feature of colors are light frequencies?), humans can perceive colors whereas some other species can't. This is just because human visual system is evolved in this way.
What you mentioned about knowing what is royal blue and electric blue is an example of learnt categories. Because to presumably to correctly name which color is royal blue or electric blue requires training, which is thorough trial and error. Painting artists can do this simply because they did a lot of training (even without doing it with the intensions, presumably they just learnt it along the way. It demonstrates that learnt categories can be acquired unintentionally.), because their work as painting artists requires them doing so(knowing which is royal blue o electric blue).
Further, categorizing electric blue is also an example of categories which can not be described, and therefore can not be learnt verbally and have be learnt through trial and error. Simply because I can't really describe to you what the features of electric blue are......
DeleteI would also like to talk about another experiment about categorical perception and phonological development in children which I believe is worth sharing.
ReplyDeleteA typical experiment in phonological development uses the conditioned “head-turn paradigm”: an infant is exposed to a repeated syllable until a switch to a different syllable occurs. A short time after the switch, a stuffed toy appears on display on one side of the room. Infants naturally orient their gaze towards the activity and find it enjoyable to watch. Soon, they associate the switch in syllable to the activation of the toy.
Janet Werker and her colleagues tested English-learning infants’ ability to discriminate contrasts that are not present in English, particularly in Hindi and Salish. It was found out that younger infants could discriminate the two sounds in each pair, however older infants could not (older Hindi and Salish infants retained the ability).
I believe this experiment is a good example that we are born with an innate mechanism to discriminate between categories (phonological contrasts) in language, however we tend to lose this ability as we grow up because it is not useful in our language.
Alara, see other comments and replies about critical periods and "use-it-or-lose-it" feature-detectors.
DeleteVery little is know about brain mechanisms of category perception and learning. It was mentioned that our ability to detect a human face is because we have an innate category detector, can we or can we not reverse engineer this innate ability? This is not other mind's problem, this is ourselves.
ReplyDeleteMonica, see comments and replies on neural nets and category learning.
DeleteInteresting to note that the acquired distinctiveness of speech relies on speech production. I am wondering, when little kids cannot say certain words, for example, they pronounce library as "libawy" is this because they hear it as "libawy" when surrounding adults say "library"? Not the most intelligent comment.
ReplyDeleteI digress. I have been pondering the notion of our CPs being rooted in evolution and what our ancestors had encountered throughout history. At some point, man developed language, which means there was a generation (or a couple) that were born without the ability to speak that then developed it in their lifetime. So, language induced CP must have been, at one point, learned and is now considered innate. Considering that we evolved from unicellular organisms, could this model not be applied to almost all of categorization capacities that have developed since then?
Laura, it's more like kids hear it as libwawy because they can only say it as libwawy at first (at least according to the motor theory).
DeleteBut what if language did not begin from vocalization, as speech, but from iconic gesture and pantomime, then subject/predicate propositions, and only migrated to speech after? In any case, categorization came first, since no other species has language.
Yes, categorization came before language, but certain types of categorization have evolved from primal categorization throughout history. Is it not correct to say, in the case of language CP, that they were once learned and are now innate?
DeleteDespite being shorter than the first reading, this reading helped complete my new understanding of categorical perception. It was initially stated that differences (between categories) can be gradual and quantitative (continuous perception) or abrupt and qualitative (categorical perception). However, the clarification that categories (within categorical perception) do not have to be all-or-nothing classifications or a matter of degree as sensorimotor categories are a mixture of the two. This reading also outlined ideas such as the Whorf hypothesis, learned categorical perception and language-induced categorical perception where factors such as our ability to produce sounds and our grounding other words through direct sensorimotor experience affect our categorization. My question relates to Sepand’s point above about the ability of language to morph our perceptions of reality, which is if there is another subsystem, we (as humans) have relating to beliefs. For example, I can be told that if I pet a dog it will certainly bite me, but that will not automatically switch my perception of my category for dog, (which typically involves the trait ‘friendly’).
ReplyDeleteHi Karina, I think your question is very interesting. When I think about specific beliefs in language, the first thing that comes to my mind is that "biting" or “being friendly” is not a distinguishing feature of a dog to place it in a particular category, since some may bite, and some may be friendly. I believe in this case, all the other distinguishing features of a dog are so dominant in our heads that even if a dog bites when we usually think of them as "friendly", this would not change our categorical perception of a dog.
Delete
DeleteHi Karina,
This brings up a very good point. I just wanted to add to Alara's response. Specifically, relating to cultures' ability of language to morph into perceptions of reality - how are other cultures able to morph their own perceptions? For example, you used the idea of still switching the perception of the category dog with the trait ‘friendly’. Most likely, in places such as Canada, we can have the same perception aka believe dogs to be friendly. But then in other countries dogs are not perceived to be friendly and rather dangerous. For instance, I’m from the Philippines and often people will tell you to not approach dogs roaming around cause dogs are “dangerous and not friendly”. In such cases, if we were to take someone from the Philippines would their ability of language morph differently if they were to come to Canada?
Hi all! A few points sparked my interest here. Karina, while I agree with Alara that I wouldn't immediately consider friendliness in the categorization of animals, I think it raises an interesting point. Your question makes me wonder if sensorimotor experience could have a greater effect on CP than than verbal learning alone. As you mentioned, if someone simply told you dogs bite and are unfriendly, it wouldn't alter your perception of dogs. However, some people are categorically afraid of all dogs after being attacked by one, creating a new category feature for dogs that is tied to fear responses. Perhaps this is off topic because it is more tied to emotion tagging than categorization, however emotions play a role in categorization as well (faces, for example, are tied to emotions).
DeleteTo my knowledge, categorical perception (CP) used to refer to the perceptual change at the boundary between two categories (i.e., the perceptual change that occurs when a perceived quality shifts from one category to another). However, CP is better defined as what occurs when we perceive within-category differences as compressed and/or between-category differences as separated, relative to some baseline of comparison. The latter definition is more flexible because it accounts for how learning affects our categorical perception.
ReplyDeleteContradictory to the Motor Theory of Speech Perception, CP is both innate and learnable. We learn CP when stimuli which we respond to differently become more perceptually distinct, and when stimuli which we respond to the same way become more perceptually similar. This learning accounts for the compression of intra-category differences and separation of inter-category differences mentioned above (relative to a baseline of how much similarity-within and difference-between the categories was perceived pre-learning).
I have two questions:
1. I'm a little confused by continuous perception... Does continuous perception refer to the adjectives that we apply to a category in light of a context? e.g., elephants are "big" in the context of animals, but "small" in the context of bodies. "big" and "small" being two adjectives that exist along a continuum of size and are used in these contexts to describe the category of "elephant".
Also, I'm curious what studies could be conducted to demonstrate that our CP of the world is influenced by what we are told about things? To me, it seems like it would be hard to design an empirical experiment that could shed light on this because how can we determine and explicitly state the "baseline of comparison"?
Polly, CP can occur as a boundary along a continuum (as in colors and phonemes) but it can also occur with distinguishing features abstracted and made more salient, from a many-featured space. (Neural nets do that.)
DeleteIn category learning, the CP separation/compression effect is based on before vs. after learning the distinguishing features.
The motor theory of speech perception may not be the whole story, but it’s part of it: speech has both a perception and a production aspect (think of mirror-capacities), so features and affordances can be sensorimotor (and can influence one another, as in the Miwor example in a commentary above).
Good question about the difference between (some) adjectives and nouns. They are a bit hybrid between content-words and function words. “Big” means more-bigness (size) and “small” means less-bigness. (“Big” is the “unmarked,” generic pole of the “bigness” continuum and “small” is the unmarked pole: Google “markedness.”) That also becomes a relative rather than an absolute judgment given a local context of alternatives. No such thing with “square” and “circlular” (unless you artificially morph them, which gives you an artificial CP in that morphing context: more circle-ish and more square-ish…)
About verbal effects on perception, see the comments and replies about social categories and stereotypes – all the way back to Tajfel’s studies (q.v.). Verbal definitions and descriptions affect perception too, if they tell you the distinguishing features of a category. Maybe they could even generate a bit of CP (not tested yet, but testable).
Hi Polly, I think continuous perception refers to the category that their differences can be perceived as gradual and quantitative, which can be described by adjectives sometimes (like big/small). Still, since it's quantitative, you can also use numbers to explain the difference, so it's not just an adjective. It depends on the context and just itself (I don't know how to word it). Prof wrote in one of the comments that "Equal-sized physical differences in wave-length are perceived as being bigger when they cross a category and smaller when they are within a category." In my opinion, it means that we are better able to detect a physical difference when the stimuli come from different categories than when they come from the same category.
DeleteSorry i just saw prof commented!
DeleteJust to add-on to Polly’s first question regarding adjectives and their dependence on context…
DeleteIn class we talked about how categorization is approximate, and features we associate to categories are context-dependent. We then discussed the potato chip example to show how context can increase confusion if we base features and approximate a category in one context, but then in a new context the previous features don't apply or don't include a new category member properly.
The example we discussed went like-so: let's say we have associated the feature of 'edible' to the category potato chips. But then, we have one scenario where we have potato chips on a deserted island and in another scenario we have potato chips while climbing a mountain– the right answer would be that we shouldn’t have the chips while climbing but should have them on the island. However, if we haven’t already learned this, our previous approximation that potato chips should be eaten/ are 'edible' would be wrong in the context of the island.
Are adjectives the same? If we approximate that elephants are big in a specific context, but in another context they end up being the smallest animal, learning will allow us to re-approximate and associate a different feature with the category…?
According to the motor theory, our mouth and nasal anatomy limit what we can produce, which in turn limits what we can perceive. As mentioned in a commentary above, this theory would suggest that kids hear “library” as “libwawy” because that’s the only way they can pronounce it (due to anatomical limitations). Although according to the reading, the motor theory has been abandoned (?), Dr Harnad mentioned above that it is not completely wrong. Based on my understanding, the reason for this is because there is a relationship between our sensorimotor interactions with the external world, and our ability to categorize. Does this imply that the way we distinguish categories is shaped by limitations of our bodily anatomy, or does it suggest that our bodily anatomy has evolved as a result of category separations?
DeleteAt the end of the article it is suggesting that language induced CP can demonstrate the weak Whorf hypothesis right?
ReplyDeleteAdditionally, can we use categorical perception(CP) and perception interchangeably? Presumably every perception is categorical?
I find the discussion of innate vs learned categorical perception to be really interesting especially in terms of the critical language learning periods as mentioned a few times above. I have a friend who is a native English speaker, but whose parents are not. When I talk to her parents, I (whose parents are both native English speakers) can detect the accent of their native language, while my friend says she cannot hear it. I wonder if this difference in our CP can be attributed to the exercising of use-it-or-lose-it feature detectors in our critical periods in the same way it can be for two children in separate environments learning different languages.
ReplyDeleteI found the section on motor theory of speech perception to be really interesting. In particular how Harnad had described how we perceive items based on the production of speech, “wherever production is categorical, perception will be categorical; where production is continuous, perception will be continuous” to be really interesting because it describes how we perceive things following the pattern we are presented with. I never really thought about how much our perception could be manipulated based on the information presented to us, and he does further elaborate how this is not a binary concept either, and there is a grey area.
ReplyDeleteThis reading was quite clear and helped me consolidate the concepts introduced in 6a, notably the significance of within-category compression and between-category differentiation in the abstract selection process of an input’s category. CP uses the former to draw similarities between inputs it wants to group together, and the latter to discriminate separate-category inputs. Learned CP allows us to form new connections and distinctions that build on our innate ones, and allow us to allocate categories through verbal associations alone of grounded categories. The other skies helped me understand the between-difference is much more significant when learning new categories because we need to understand the general features of a category before we start to specialize.
ReplyDeleteAfter reading the article, I am still left wondering with the question whether CP of human emotions is something that is learned or innate. The paper mentioned a study by Eimas et al. (1973), who found that infants already have some speech CP in place even before they begin to utter coherent words. Is this the same case for emotion recognition? (i.e., seeing an angry face makes the infant react a certain way vs. seeing a smiling face)
ReplyDeleteWhen rationalizing colour perception in terms of AI development, I thought of the concept of colourblindness. Many colorblind individuals are able to make correct color categorizations despite physical inabilities to sense color of a particular wavelength.
ReplyDeleteIn this way, there exists much variation in the ways of achieving the behavioural equivalence of correct color categorization.
To me, this strengthens the idea that we could create a behaviourally equivalent AI consciousness. Just as humans can come to create similar categories of color perception given different means of sensorimotor information, it seems increasingly likely that a computer could also do the same for even more complex concepts.
I agree that your colour blindness example strengthens the argument that we should focus more on AI instead of studying the brain directly if we want to learn how humans make categorizations (which Harnad has argued is what cognition is anyways). As a slight wording change I would say we need to create a functionally (not behaviorally) equivalent AI conscious though if we want to reverse engineer capacities.
DeleteThe line that divides categorical perception and continuous perception is quite blurry. For instance, in many of our sensory systems, we rely on lateral inhibition to maximize and emphasize contrasting elements in any given modality. This allows us to efficiently organize the informational input to produce the appropriate output. This can be used to explain why certain colours appear to be more distinct, while others fall on a spectrum forming some sort of range. However, this becomes more complex when we consider the variances in which differing cultures interact with colours and how this can be reflected in language. For instance, certain languages have more words to describe varying shades of green than others due to their ancestors' increased exposure to more natural scenes. What I would like to ask is whether users of languages with different colour boundaries actually perceive these colours differently or if this difference goes only as far as language? In other words, does one’s increased exposure to a certain colour visually sensitize them to the differences across shades, or do they merely only refer to these shades more often?
ReplyDelete"What I would like to ask is whether users of languages with different colour boundaries actually perceive these colours differently or if this difference goes only as far as language?"
DeleteThis is an interesting question. I think according to the Whorf hypothesis the answer is yes, different languages do affect the speakers cognition in different ways (like the ability to categorize different shades of green better than people who speak another language). Your question is more specific though, referring to perception vs language (ie categorization). I think that the Whorf hypothesis would lead us to say they are "visually sensitized to differences accross shades" and not that they merely refer to them more
The idea of language-induced CP effects is interesting. It is fascinating to think that, through language alone, we can acquire new categories that we have not had sensorimotor experience with. As long as we possess previously-grounded categories, they can be combined to produce higher-order combinations. In this way, humans alone have the ability to abstract categories that no other species can. For example, we can create a category for peek-a-boo unicorns, although we have no sensorimotor interactions with such a thing because it does not exist. By acquiring this category through language, however, “peek-a-boo unicorn” inherits compression/separation of horse + horn + disappearing, while generating additional CP effects of its own.
ReplyDeleteIn the reading, it is admitted that we know very little about the mechanisms of category perception and learning in the brain, however I think that the computational modeling of CP with the back-propagation nets actually makes a lot of sense as a hypothesis as to what the brain might be doing, especially if we regard it in the context of how we use categories. Imagine we are put in a situation where we are shown objects coloured with different shades of blue: if we are asked to name each object’s shade, we would need to separate the inputs in order to generate the different names (output) of each shade, but if we were instead just asked what colour these objects are, one would just answer blue, so generate the same output. In this context, one can see how a process like this may be occurring in the brain.
ReplyDeleteI just wanted to post this comment as a follow up to my question in class regarding “crimson” and “scarlet.”
ReplyDeleteI know another example that was brought up in class was that of “black” and “white” and the shades of grey in between. From my understanding, the pigeon experiment showed that there is no learned CP with shades of grey because you have the categories of black and white… and the innate CP would just categorize the shades under black or white… However, this made me reflect on other languages and I thought of my mother-tongue, Farsi. In Farsi, we have specific names for shades of grey: “خاکستری” (pronounced ‘khakestari’) and “طوسی ” (pronounced ‘toossi’). If I were asked to categorize different shades of grey in Farsi, instead of labelling them as ‘black’ or ‘white’, I would very easily be able to categorize them under one of these two specific labels… so is this not evidence that even though colour perception is innate, when it comes to subcategories, they can be learned and furthermore, can this also be evidence for the claims of the Whorf hypothesis?
There is a colour wheel on this website (https://arga-mag.com/574913/تفاوت-رنگ-طوسی-و-خاکستری) that perfectly depicts the two categories of grey… the ones on the left would be ‘khakestari’ and on the right ‘toossi.’
Categorical perception (CP) functions like a threshold that once reached signifies membership or non membership in a given category. This ability is inborn in humans and other animals to some extent and can also be induced by learning. CP is also responsible for within category discrimination, however I did not understand this point as well and could use some clarification. The interplay of continuous and categorical perception in human language is one reason why we are able to use previously grounded words to scale up and introduce complex language from those elementary grounded building blocks. Our ability to do this so well is something that makes use uniquely human and the most social and dominant species on the planet.
ReplyDelete