An Invitation to Cognitive Science, 2nd Ed. Volume 1: Language. Cambridge, MA: MIT Press.
Alternative sites: 1, 2.
The topic of language acquisition implicate the most profound questions about our understanding of the human mind, and its subject matter, the speech of children, is endlessly fascinating. But the attempt to understand it scientifically is guaranteed to bring on a certain degree of frustration. Languages are complex combinations of elegant principles and historical accidents. We cannot design new ones with independent properties; we are stuck with the confounded ones entrenched in communities. Children, too, were not designed for the benefit of psychologists: their cognitive, social, perceptual, and motor skills are all developing at the same time as their linguistic systems are maturing and their knowledge of a particular language is increasing, and none of their behavior reflects one of these components acting in isolation.
Given these problems, it may be surprising that we have learned anything about language acquisition at all, but we have. When we have, I believe, it is only because a diverse set of conceptual and methodological tools has been used to trap the elusive answers to our questions: neurobiology, ethology, linguistic theory, naturalistic and experimental child psychology, cognitive psychology, philosophy of induction, theoretical and applied computer science. Language acquisition, then, is one of the best examples of the indispensability of the multidisciplinary approach called cognitive science.
Harnad, S. (2008) Why and How the Problem of the Evolution of Universal Grammar (UG) is Hard. Behavioral and Brain Sciences 31: 524-525
Harnad, S (2014) Chomsky's Universe. -- L'Univers de Chomsky. À babord: Revue sociale es politique 52.
I agree with Chomsky’s intuition as to “the interface between meaning and syntax is that language is what makes it possible for us to think at all”: without language, one can still feel, but one would not “think”. I would argue that by putting feelings into language, one becomes an active agent in the passive experience of living, and thus becomes “conscious”. If anything, language makes a person more conscious about what oneself is experiencing, and at certain times, one needs language to guide oneself to make the right decisions, and more.
ReplyDeleteI find this part quite interesting, too. “The possibility of thinking itself comes with certain Platonic constraints, which, in turn, result in the rules of UG.” If it is the constraints on thinking/thoughts that result in the rules of UG, then thinking/thoughts must have existed before UG. Where do the constraints on thinking/thoughts come from, then? (If not “logic”... I think this may have potential evolutionary explanations.) However, I am not sure if it is meaningful to further investigate this “idea about constraints on the nature of thought” since it is “neither an empirically testable theory nor a mathematically provable theorem”, but I do not see it as a rival theory of UG – again, maybe I am just being like Chomsky, who perhaps, “is closer to being a Platonist, in that the innateness and universality of the ideas is not a result of evolutionary selection but of the natural laws of the physical universe and perhaps even the universal laws of formal logic and mathematics.” And, it is never “*just* logic”. I, of course, do not think that I have enough technical knowledge of UG yet to tell if every UG rules could come into existence because of logic, but I do care enough to want to learn more about it so that I could see if there could be something more abstract beyond UG.
Chomsky does not think that organisms without language don't think. He just thinks that only thoughts that can be expressed UG-compliantly are thinkable. (And there are no "unthinkable thoughts.")
DeleteLanguage makes it possible to think more thoughts -- and if feels like something to think. In that sense language does increase "consciousness" (a weasel word), meaning that we can feel more things, because each new thought feels different.
Every (coherent) sentence expresses a thought. Every sentence also describes a category. (The subject category is "in" the predicate category: "The cat" is "in" the set of things that are "on the mat.") So all propositions describe or define possible or potential categories. It is in this infinite recombinatory potential of category names and subject/predicate propositions that potential thoughts are already latent in language. (This is not a profound thought. All potential integers are latent in counting.)
With the example you provided in class of "who did he think that went out" versus "who did he thank that went out", I see how as a native English speaker we wouldn't say the first phrase. But is it fully unthinkable? I feel that if someone said this to me I would understand. Is that just me rearranging their non-UG compliant phrase into a UG compliant phrase in my head in order to make sense of it?
Delete
ReplyDelete“For one thing, the most general learning algorithm one might conceive of -- one that is capable of hypothesizing any grammar, or any computer program capable of generating a language -- is in trouble. Without negative evidence (and even in many cases with it), there is no general-purpose, all-powerful learning machine; a machine must in some sense "know" something about the constraints in the domain in which it is learning.”
I am a little confused as to what Pinker is implying here. Is he saying that because of the way language acquisition is acquired by children, that previously held perceptions about the development of language are no longer relevant? Or is he saying that language acquisition by a computer will never be possible due to the limited impact negative evidence has on actual language acquisition?
I don’t believe Pinker is arguing either of those points. Shortly before this passage, he explains four possibilities of children forming hypotheses when learning a language. The fourth possibility states that the child’s hypothesis language is a superset of the language attempting to be learned. This means that the child utters multiple sentences, including grammatically correct and incorrect phrases. Without negative evidence, the child would not realize that their language hypothesis is incorrect, and consequently they would continue saying grammatically incorrect sentences. Pinker argues that there could not exist a machine that could learn a language without negative evidence, because it could never learn the boundaries. The boundaries of what is correct and incorrect in language must already be known. While some, such as B. F. Skinner, have argued that children receive this negative evidence through parental reinforcement and correction, this hypothesis has been largely unsubstantiated. Parents rarely correct children’s grammar, and even if they did, the corrections would have to be continually reinforced to eliminate the error. On this point Pinker says, “If the world isn't telling children to stop, something in their brains is, and we have to find out who or what is causing the change.”
DeleteIt's much simpler than that. To be able to learn UG, the child would have to hear and speak UG-compliant utterances (positive evidence) and UG-non-compliant utterances (negative evidence) in order to be able to detect the features or rules that distinguish the two. That would be ordinary supervised learning: trial and error with corrective feedback. And that is how OG is learned. (the child makes plenty of OG errors, and gets corrected, or figures it out. So the child can learn OG.)
DeleteBut the child does not hear or speak UG-non-compliant utterances. No negative evidence. No correction. No way to learn the rules.
So how is it that the child only makes UG-compliant utterances?
Chomsky's conclusion? The rules of UG must already be innate.
How they evolved is another matter, and a hard one (though not the hard one).
UG is definitely throws a wrench in multiple levels of language. On one hand, it disrupts the evolution of language. It also brings to question the learnability of language and if there is an innate grammar that is within humans or potentially other organisms. As Baldwinian theory states that animals may not have had the redisposition or motivation to develop language, they could also have UG and just never needed to use it? I don’t know how much I believe that is true, but since we don’t even know much about UG in humans, I couldn’t see how we can prove that it is not in other organisms as well. With this said, we do that that UG exists and one aspect of UG is that it requires no negative evidence. We also know that for anything else, we need both positive and negative evidence to acquire a category. Computation needs this too in order to categorize. The computation, like you said, needs to know what the boundaries are. However, I don’t believe that this is the only we way to get at language and the development of language. It may take much longer and involve a lot more trial and error, but I think you can get a machine to learn language with solely sensori-motor and take out the UG all together. This could be a case of underdetermination where you don’t necessarily need to get at the language the same way that evolution did so. Because at the end of the day, evolution is lazy. There could be another, more optimal way.
DeleteI question whether it would be possible for a machine to learn natural language purely on sensorimotor input without considering UG at all. While I agree that evolution is lazy and other ways of communication can and have been shown in other species, there is nothing as nuclear as language. Even though the lack of negative evidence we experience with UG conflicts with how we understand categorizes are acquired, UG-compliant utterances are nonetheless still universal across humans and to me, it seems simplistic to dismiss incorporating UG when building a T3-passing robot. If this is necessary to modeling language for input-output equivalence, the issue to me now is how then we can even begin to formalize UG? It cannot be via unsupervised learning with correlations. Supervised learning is also ruled out as it does not reflect natural language acquisition. In the case of the adult students who taught themselves UG rules, they already possessed UG and thus were able to judge positive and negative evidence (the latter at which we never receive in real life). While I disagree that UG can be ignored then building a machine that acquires language the same way as we do, how it can be incorporated at all is a big mystery to me at the moment.
Delete“It is a general danger that pops up in cognitive psychology whenever anyone proposes a model that depends on correlations among features: there is always a temptation to glibly endow the features with the complex, abstract representations whose acquisition one is trying to explain.”
ReplyDeleteThis passage is trying to expose why children could not learn the grammatical categories in language using correlations features. Pinker argues that in order to recognize the grammatical category (noun, verb, etc.) of a given word, the child must be able to recognize the structure of the sentence, because the linear order of the words is not always a good indicator of the role of a given word in this sentence. But it implies an already existing mental representation of the structure of the sentence, and since the child could not learn a structure without knowing the role of the words, the underlying structure must be innate.
But I am not sure if the example of grammatical roles given in the paragraph 8.1 relies on ordinary grammar (learnable grammar) or in universal grammar (unlearnable). Is the fact that **there are** nouns, verbs, adjectives, etc. a fact of universal grammar? And on the other hand, knowing that a given word is part of this given category would be an example of ordinary grammar?
Pinker is completely conflating OG and UG, and using (learnable and learned) OG examples to illustrate UG and its unlearnability.
DeleteUG is far too complicated to be learnable through unsupervised learning (correlations). And the absence of negative evidence makes UG unlearnable through supervised learning.
This is more of a comment to clarify connections (if any) between concepts.
ReplyDeleteFrom my interpretation of the article, negative evidence seems to be similar to (or the same as) corrective feedback. It also seems that children do not really need negative evidence to learn what is grammatical versus ungrammatical. Parents tend to correct the content of the utterances (i.e., if the utterance is true or not) instead of the grammar of the utterances. Moreover, parents may not even react to ungrammatical utterances and can still understand them. So, it seems like corrective feedback on ungrammatical utterances aren’t that important for learning a target language. (Please correct me if I am wrong.)
If this is the case, then does that mean that learning a language is a form of unsupervised learning? Children are essentially learning to do the right thing with the right kind of thing, but without being directly informed of what is right. Or is learning a language considered supervised, since the right kind of thing (i.e., correct grammar) can be learned through positive evidence? In other words, the grammatical sentences of the target language are “demonstrated” by their parents; is that enough for it to be similar to supervised learning?
“If this is the case, then does that mean that learning a language is a form of unsupervised learning?”
Delete“Or is learning a language considered supervised, since the right kind of thing (i.e., correct grammar) can be learned through positive evidence?”
Ordinary grammar (OG) can be learned by unsupervised observation (I think an example would be a child learning to associate the sound sequence ‘dog’ with an actual dog, after having been repeatedly exposed to contexts in which the two co-occur) as well as supervised learning (positive/negative evidence). Positive evidence refers to the grammatical sentences available from parental input, while negative evidence is the correction of a child’s ungrammatical sentences by a parent/caregiver.
Universal grammar (UG) is innate (not learned). Children never hear examples of violations of UG rules, and they also don’t produce UG mistakes that they could potentially get corrective feedback for (no negative evidence, i.e. the poverty of the stimulus). Since positive evidence alone is not enough to learn UG (I believe a child would produce both grammatical and UG non-compliant sentences if that were the case), it must be innate.
“So, it seems like corrective feedback on ungrammatical utterances aren’t that important for learning a target language.”
I’m a bit confused as to how important negative evidence is for learning OG, given that the literature suggests that parents rarely supply negative evidence and children often don’t notice/ignore corrections (see McNeill 1966 for an example of the latter). Could it be that negative evidence plays a minor role in OG learning, while positive evidence is really what drives it? But isn’t negative evidence necessary to learn what sentences are ungrammatical (since only knowing what sentences are grammatical doesn’t prevent me from producing ungrammatical utterances)?
OG is learnable, and learned, through a combination of unsupervised and supervised learning as well as some explicit instruction. But it is probably completely learnable through unsupervised learning alone. All these possibilities for OG are the reason that OG is very misleading if used to try to illustrate the unlearnability of UG.
DeleteThe child cannot learn UG because of the lack of negative evidence (error-corrective feedback: no UG errors), also known as the "poverty of the stimulus."
You cannot understand why UG is unlearnable if you do not distinguish UG from OG.
“A key factor is the role of negative evidence, or information about which strings of words are not sentences in the language to be acquired. Human children might get such information by being corrected every time they speak ungrammatically. If they aren't -- and as we shall see, they probably aren't -- the acquisition problem is all the harder.”
ReplyDeleteHere Pinker is outlining the poverty of stimulus argument, which posits that a child is not exposed to examples of ungrammatical sentences. This proves that supervised learning and instruction are not enough for a child to learn a language. Therefore, there must be some internal, innate mechanism that allows a child to integrate external information with the underlying linguistic scaffolding to allow for language acquisition.
UG ends up being the preferred explanation for this internal template, but I think Pinker confounds ordinary grammar with universal grammar when he gives examples of grammatical roles. This ends up inevitably weakening his stance, as he lacks the technical details of the concept he’s advocating for.
I also found it incredibly interesting when Pinker brings up language and thought. He does not, however, mention Fodor's Language of Thought hypothesis, which although probably false, is an important part of this debate.
What is Fodor's Language of Thought Hypothesis and its relation to this debate? What is the debate?
Delete“Is language simply grafted on top of cognition as a way of sticking communicable labels onto thoughts?”
DeleteBriefly mentioned, the Language of Thought hypothesis by Fodor states our thoughts are strung together in syntaxial ways, similar, but not necessarily the same as language. Emmanuelle talks about the importance of this hypothesis in relation to the debate of how language began. Clearly, supervised/unsupervised learning is not enough to learn UG, so it must be innate. Maybe this cognitive mechanism Fodor hypothesizes could be a basis for UG? Is this why you think it is an important part of the debate?
“Children most definitely do need to hear an existing language to learn that language, of course”
ReplyDeletePinker is stating that in order for language to come about, even with the vocal cords that have been evolved for that purpose, spontaneous speech will not occur. Speech always has to be heard before it can be an output. This really brings about the question of what triggered speech that first instance that can no longer trigger it independently in children today. From Pinker, I would determine that something due to social circumstance served as the instigator for language as it now continues to do so. As to what was the initial evolutionary benefit of these first utterings, I find that Pinker does not provide any additional explanations. Particularly his examples in sections 8 and 9 do not provide any evolutionary grounding for the development of universal grammar. However, Pinker’s paper also determines that a language cannot be learned from exposure, so there must be some intrinsic factor that leads to the learning of language. This was already determined during last week’s class so it doesn’t feel as though Pinker provides much more insight than his previous paper.
“This really brings about the question of what triggered speech that first instance that can no longer trigger it independently in children today.”
DeleteI’m not sure that this question does follow, or that there’s any contradiction happening here.
I don’t think in Pinker means that children don’t use their speech faculties to produce sounds without hearing language first – I think by “speech”, he means that they don’t use ‘language’ without hearing it first, (of course not – how could one use arbitrary symbols to produce meaning if you don’t know the symbol system – and since the symbol system is arbitrary to the meaning (‘just squiggles and squaggles’) – one can’t ‘invent’ it from introspection).
And surely since language was developed by a group in a social context, (and as proposed in class, this was probably developed from abstracted gestures that had originally been pantomime), it’s not inconsistent that children would learn language in a social context as well - so I think this question comes from a bit of a false equivalence.
Also – I think Pinker’s paper doesn’t determine that “a language cannot be learned from exposure”, but rather that certain grammatical constructions (UG) aren’t learned from exposure, and instead are innate, while other constructions (OG) are learned from exposure.
Yes, language has to be distinguished from speech, UG has to be distinguished from OG, and language learning has to be distinguished from language evolution to make sense of any of this. The question is whether UG can be learned from unsupervised learning, supervised learning, or instruction.
DeleteUG cannot be learned from unsupervised learning, by detecting correlations.
UG cannot be learned from supervised learning, because there are no UG errors and corrections.
And children are no instructed in UG rules because almost no one (other than Chomskian linguists and their students) even know what the rules of UG are.
OG, on the other hand, can be (and is) learned in all three of these ways.
The question of how language or UG evolved in the ancestral environment is not the same as the question of how children today learn language. And UG is not learnable from what children say or hear.
None of this has anything to do whether language today is vocal, gestural or written.
To learn vocabulary is not to learn grammar, either UG or OG.
After reading this string of comments, I would like to say that I think it is very important to distinguish language learning today from language evolution. I do not think we should compare language learning of children today with how language first evolved because we will not get anything valuable from comparing the two. If anything, we will cause confusion. The context and environments in which children learn language today was very much different than that of the human beings that were present when the evolution of language was first taking place. The question of how language evolved is a difficult one because we do not have direct access to the ancestral environment and the environmental pressures! We cannot simply observe how children develop language today to make inferences about how humans developed language in the first place.
DeleteAs Pinker points out, Chomsky's famous argument was that "children learn languages that are governed by highly subtle and abstract principles, and they do so without explicit instruction or any other environmental clues to the nature of such principles. Hence language acquisition depends on an innate, species-specific module that is distinct from general intelligence"
ReplyDeleteThe innate, species-specific module is essentially UG. Pinker later discusses some of the components necessary for learning a language. One of these is positive evidence, which is essentially the input. Having linguistic input is necessary for a child to acquire that language. Pinker argues that negative evidence is not necessary for a child to acquire language. "Negative evidence refers to information about which strings of words are not grammatical sentences in the language, such as corrections or other forms of feedback from a parent that tell the child that one of his or her utterances is ungrammatical". This supports the poverty of stimulus argument that Chomsky proposed; a child does not receive all of the feedback on their speech, yet is capable of fully learning a language. Supervised learning is not necessary for language acquisition. Thus, there must be some internal mechanism humans have regarding the ability to learn a language.
I'm a bit confused by some of the examples Pinker brings up regarding how negative evidence is unnecessary for a child to acquire language. "For example, children can learn a language without the special indulgent speech from their mothers; they make few errors; and they get no feedback for the errors they do make." I could not seem to imagine an environment where negative evidence would be completely absent in a child's language acquisition. I'm referring to a child's entire language learning environment (not just some of the controlled experiments Pinker outlines in the article).
To make sense of this you have to distinguish (1) learning language from (2) learning grammar; and learning (3) OG from (4) "learning" UG.
DeleteFrom what I understood from last class, Universal Grammar is a set of unconscious structural constraints innate to all humans, upon which language is acquired. To kid-sib this, universal grammar differs from ordinary grammar in that the former applies to all languages and cannot be learned, whereas the latter is specific to each language and can be learned. I understand that UG cannot be explicitly defined and Professor Harnad mentioned that Pinker actually presents examples of OG; however, Pinker proposes "language learning algorithms" and methods by which children are able to acquire language (section 9, "Acquisition in Action") that seem to be universally applicable to all languages. For instance, Pinker posits that "children assume that semantic and syntactic categories are related in restricted ways in the early input" and thus by "[using] semantic properties of words and phrases", can then learn syntactic categories. Why is this not a mechanism of UG?
ReplyDeleteI learned in a child development class about a study in which 8-month-old infants were presented with a two-minute stream of three-syllable nonsense words: tupiro, bidaku, padoti, and golabu. Based on transitional probabilities between syllables, Shapiro et al. found that babies were able to distinguish new "words" from familiar ones and distinguish between word boundaries. This demonstrated a pre-linguistic child's ability to learn words with impoverished information. Is this statistical learning paradigm of language acquisition also an example of OG and not UG?
Hi, I think the first part of your commentary is pretty accurate, but I think specifically because your example of "statistical learning paradigm" as you say is about *learning*, it rules out the possibility of having anything to do with UG. Of course, on one hand, learning through algorithms might be a misleading title, but if I am correct, the algorithms in this example are basically what UG is. It’s hard to define it in specific terms, but we could imagine that UG is what gives the children the ability to learn those words, and the way to achieve that is to have algorithms that help them generalize and make sense of the stimuli they encounter. On the other hand, the words that are processed in those algorithms are indeed learned. It's possible that the ability to distinguish between words that don't make sense yet is a part of what UG offers to a child, but learning the words seems to be exactly what OG is. Isn't that where we draw the distinction between both?
Delete(1) The rules of UG can be learned -- by adult students, from positive and negative data gathered across decades by teams of linguists testing their hypotheses about what is and is not a rule of UG by trial and error: Their guide as to what is right and wrong is the fact that they all already "have" UG, so they can tell whether something is right or wrong.
DeleteThe language learning child does not have any of that negative data (or time!).
(2) UG itself is not learned, but the child does need to be exposed to language in order to "trigger" their innate UG. They also learn what "flavor" of UG their trigger language happens to have: is it Subject-Verb-Object order or Verb-Subject-Object? Does it have separate Pronouns, or are they inflected on the ends of words (Pro-Drop)? These are "parameter settings" on UG; they are learned through exposure, use, correction or instruction, just as OG is.
(3) Any innate "assumption" or "algorithm" about UG rules is "part" of UG -- except of course the capacity for unsupervised and supervised learning itself, which is just general learning ability.
(4) The relation of syntax to semantics remains a fuzzy matter when it comes to UG syntax. There is no problem with learning OG syntax, which, like categories and category-names (words) -- and hence their meanings -- are all learnable and learned, some of it via unsupervised learning (as in the old/new "word" exposure experiment with babies [tupiro, bidaku, padoti, and golabu], which misses the point completely, by not distingusing (i) UG, (ii) OG, and (iii) just the learning of old and new sounds and their boundaries).
This comment has been removed by the author.
ReplyDeleteOf course nonhuman animal communication capacities are precursors of human language capacity. The question is whether they include precursors to UG itself, its rules and structures. The current answer is somewhere between "don't know" and "no." UG is a huge complicated, abstract structure: Is there something like it that preceded language and its UG? If so, what was it, and how and why did it evolve?
DeleteThe "babble-luck" hypothesis is Skinnerian (supervised learning), and was refuted by Chomsky's famous critique of which the punchline is the poverty of the stimulus and hence the unlearnability of UG:
Chomsky, N. (1980). A review of BF Skinner's Verbal Behavior. Language, 35, No. 1 (1959), 26-58.
All categories and all descriptions/definitions of words are approximate (except in formal maths). Hence they all "fall short." The power of (grounded) re-combinatory propositions is that it can make the approximation as close as we like, by extending the descriptions/definitions of new categories. (You don't need to coin a word to name every possible category: it's enough to have a verbal description.
The difference between being unable vs able to produce and understand propositions is indeed qualitative (all-or-none), not quantitative, and that's the difference between other species' communication capacities and language. (But, on the face of it, that has nothing to do with UG in particular.)
I am also still confused about the ability of language to describe everything, and I think it is probably because I am still not on the right scope of your argument. My understanding of language cannot describe anything that comes from the intuition that many abstract things in the world exist on a spectrum that cannot be pinned down belong to a particular category and it irreducible to being captured by language. It is hard to illustrate as once I started trying to explain, I am still trying to capture this thing under the system of existing language. For example, the human emotion that one has for someone that can varies widely from love and hate, like and dislike and possibly many other spectrums and perhaps it is not linear with regard to the metaphor of “spectrum” and it is not possible to give this emotion a name. Other examples might be the feeling of seeing red and the taste of strawberry ice cream. Is in these cases, the proposition of “the feeling A has for B”, “the taste of strawberry ice cream” an adequate language to describe?
ReplyDeleteConcerning Pinker’s theory of language acquisition, I think the main idea of how children could learn the first language is straight-forwardly explained in this paper. However, I wonder how, related to his theory of the evolutionary account of language, he would explain the difficulty associated with second-language learning. I think the hypothesis we discussed in class says the degeneration of language acquisition device once the kids have reached a certain age. But if we have the device already that is hardwired with us when we are born, why would we bother to de-ware it after a certain age, as it does not do any harm to us. In other words, what is the thing that Pinker thinks that accounts for the loss of the language acquisition device? Does that still conform to the big picture of evolutionary theory? Is it possible that with the human course of socialization becomes more and more globalized, we observe a trend that children of each generation have a longer period of possessing this device?
“many abstract things in the world exist on a spectrum that cannot be pinned down belong to a particular category and it irreducible to being captured by language.”
DeleteFor the language, you do not necessarily need a category to explain each thing, even if they are abstract. The language is powerful because you do not need a name for each category like Funes does in Borges’ short story. Approximate explanations are enough for the language. The fact an emotion doesn’t have a name doesn’t change the fact that you can describe it by language. Instead of saying “the feeling A has for B,” you can explain what you feel when you see red: “seeing red is like tasting a hot pepper… etc.” to demonstrate the power of language.
I think your question regarding the language acquisition device is interesting. I am going to speculate about it a tiny bit. As we discussed in the class, UG has different parameter settings. Even though all languages have UG, different languages can have different parameter settings for UG. Maybe the parameter settings for the first language become so rigid that later you have a difficulty of learning a second language. So perhaps it is not the dewiring of language acquisition device but rather the inability to use it that leads to the problem regarding the learning of a second language.
On the topic of Funes, it makes sense that the inability to categorize would be a massive detriment to anyone as it completely impedes one’s language ability. In the story Funes can express his feeling in language whereas in real life, he would not be able to. I understand that Funes was not born with this deficit, but I wonder if we could draw any parallels between someone born with the inability to categorize would be synonymous to someone lacking UG? Here the patient wouldn't be able to learn language even if they were provided with ample stimulation? Therefore, is categorization an essential part of UG’s innate algorithm that promotes language learning?
DeleteIn relationship to your second points about bilingualism, I had a similar question but about simultaneous bilinguals. Simultaneous bilinguals are those who learn two languages from birth because they receive almost equal stimulation for each language. An example of this would be an infant growing up in a home with one English speaking parent and one French speaking parent. Here, the child will receive equal input from both the languages, so is the child’s UG being stimulated by 2 sets of parameters at the same time? I guess in your words Omar, this would imply that the parameters become “rigid” for both different languages at the simultaneously.
This leads me to my next question, simultaneous bilinguals are relatively common, but could there be a “cap” to the number of languages a child could learn from birth? My intuition tells me theoretically no, because if UG is the base for any language and it is stimulated during the critical period then it should be able to learn any new language in the way one learns their mother tongue.
Mingjuan, categories (whether sensory, emotional or "abstract") are all approximate (except in formal maths). Sensory feature-detectors (learned and innate) are approximate. So are verbal definitions and descriptions. A picture or obect is always "worth more" than 1000 words.
DeleteWhy does lazy-evolution produce "critical periods" (whether for duck-imprinting or 1st-language learning) in development rather than continue with full sensitivity lifelong? Probably expensive to maintain: a mole need retain the ability to grow again just in case he may change burrows later in life.
Onur, yes, 1st parameter-settings may produce conflicts with later languages.
Akhila, UG is syntax, not semantics, but it is the syntax of a natural language (whose words also have meaning), not just a formal language, like mathematics, which does not "use" meaning, and is hence pure syntax. And for words to have meaning, they must name (enough) grounded categories.
Remarkably, while the window of the "critical period" for 1st-language-learning is open, there seems to be no "cap"(except the number of hours in a day) to how many 1st languages can be learned in parallel. I don't know if there are any subtle deficits that come with learning more than one, the way there are in later languages. (Also, there is some evidence that learning many first languages can keep the language-learning window open much longer -- for some polyglots for a whole lifetime: there are some examples on youtube, and they are remarkable, but for the (few) languages I know, it is clear that those polyglots, though amazing, do not have 1st-language competence in all their languages.)
*"Children clearly need some kind of linguistic input to acquire a language. There have been occasional cases in history where abandoned children have somehow survived in forests, such as Victor, the Wild Boy of Aveyron (subject of a film by Francois Truffaut). Occasionally other modern children have grown up wild because depraved parents have raised them silently in dark rooms and attics; the chapter by Newport and Gleitman discuss some of those cases."*
ReplyDeleteFrom this, I am curious if we can use this as a way to test how important being a child is to learning language. Obviously there are numerous ethical issues with experimenting on children, but should a child survive such an experience, would they be able to learn a language later? How would their rate of learning differ from a more typical child learning their first language or an adult of similar age learning an additional language? I'm certain this question has undoubtably been asked so I suppose I am looking to be enlightened by my peers with more knowledge in this area.
Presumably, the trauma of a childhood resulting in being mute would muddy the data significantly, but I believe there is still some interest in this scenario.
I know that in many of the cases of feral children, they were eventually taught language. The problem though is that they were never able to fully learn the language. If I remember correctly, their vocabulary was severely limited as well as their grammar and they were only able to reach the linguistic skills of a 5 or 6 year old. The thing is, if I were to attempt to learn a new language, I may never learn to speak it like a native speaker would, but I certainly could eventually learn the language well enough to be considered fluent, however the feral children were never able to learn any language. So it seems as if there was a critical period to learn language, but the problem here is we can't be sure whether their inability to learn the language is related to UG because they also had severe cognitive deficits. So, similarly to how an adult with a severe mental disability may never be able to speak, it wouldn't be because of a deficit in UG, but rather in a more global cognitive deficit which would limit just about every intellectual capacity
DeleteRe: Johnny Nahas: “however the feral children were never able to learn any language.”
DeleteI’m actually curious to know what process are responsible for that. Kyle and Johny’s skywritings imply that trauma and general cognitive deficits would be to blame for these limited linguistic capacities, but I wonder what role the critical period for language acquisition plays in all of this. Kyle talks about he trauma of being mute but some children are born congenitally deaf and mute and are still able to learn sign language and communicate with others.
I was wondering, if we consider the critical period to be a Baldwinian evolution process, where children get a surge of motivation to learn language at that specific age, could we argue that this inability to learn language stems from a lack of motivation to do so, rather than trauma? I just find it interesting that feral children never develop a language beyond the level of a 5 or 6 year old, which also happens to be the age at which children experience the critical period Is there be more to the critical period than motivation that explains why that age is so important?
If Language Acquisition is learning a language and how to produce it in order to communicate any idea that can be communicated, then I feel there is a flaw in the example given of the child who was unable to speak due to a neurological reason, but understood everything said to him. This gets into the territory of being able to understand versus produce language. In the case of many children raised in a bilingual household, one language ends up being dominant, the one that's often used outside the home enabling more practice. The other language may not be spoken as often by the child, so they end up being perfectly able to understand it, but not able to produce the language, thus not able to say anything they would like to. This to me seems similar to the case of the young boy who understood but wasn't able to speak. He, and those raised in a bilingual environment, acquired the language, but aren't able to produce it to the extent of expressing any idea. How does this nuance relate to the idea of Language Acquisition? Is this not a fair parallel because the "bilingual" children then fall into a non-native speaker category so it doesn't apply to them? This would put them in a similar class, though perhaps with a greater understanding of the parameter of the language filtering the UG, as someone who learned the language as an adult.
ReplyDeleteI disagree with your assertion about the child who is unable to speak but can understand everything said to him. The communication of any idea that can be communicated is not constrained to just verbal contexts. Just because he cannot produce propositions verbally doesn't mean that he cannot at all - what about through the medium of writing? Along with this, I don't completely understand why you distinguish understanding categories and producing language as being mutually exclusive with the former being unrelated to the acquisition of language. As proposed by the test for natural languages discussed last class / previous reading, around a minimum of 1500 category names (along with functors and syntactic rules - UG) are necessary for language. Due to this, I am speculating that it is integral to language acquisition and thus, the example does apply. In your second case of the bilingual child, I do not think that just because he cannot produce full propositions or can only say single words in his second language that he has not acquired ANY of the language. As discussed, the parameters for UG become less flexible as we get older. If the bilingual child fails to use his second language often, producing correct grammatical sentences in the second language may be significantly more difficult. This is especially if the second language has a completely different grammatical structure the first; e.g., English is Subject-Verb-Object vs Chinese is Verb-Subject-Object. However, this does not mean one could not produce UG well-formed sentences in the second language at all. For example, with many English-speakers first learning Chinese, they tend to structure sentences in Chinese with the English Subject-Verb-Object grammar structure. Yet, the native speaker can still understand the idea they are trying to communicate. I do not think that this means they have failed to acquire the Chinese language at all.
DeleteIn response to both Gabriel and Madeleine, second language acquisition and the grammatical mistakes made in the process are not part of UG. The failure to express certain thoughts in another language than our original one is not an example of non UG compliance. Indeed, the superficial mistakes made by second language users are only part of OG. Second language users are basically just learning a new flavor of the same UG governed set, and what they are mixing is only the respective spices of these two flavors.
DeleteThe problem is that we need a meta-language to express the rules of UG which makes it easy to confuse UG mistakes and OG mistakes. Pinker mixes the two and I think that is why you guys are confused about this.
The article discusses, at the beginning, the question of language’s relationship with cognition, and sort of goes into an explanation of how this might be a moot point and that virtually all linguists think that the Sapir-Whorf Hypothesis is false.
ReplyDeleteThe main question asked here is the temporal order of the cognition and language link: is language just a series of labels for cognitions that already exist or does language in itself involve learning how to think differently. To me, what that really boils down to is whether we think in propositions before we acquire language or not, but the article uses the failures of the Whorf-Hypothesis to disregard this concern.
The way I’ve understood this hypothesis is that language influences cognition in the sense that our perception and view of the world changes as the result of language (and especially the specific language within which we are immersed). It’s true: we don’t legitimately perceive things differently, but I think that the question about when propositional thinking emerged might be a flaw in the criticism.
Isn’t it a fair question to ask? I have never heard (including in this article) an argument for when propositional thought (and not propositional communication as a step in the acquisition/development of language for humans) would have emerged, and I think it might be worth asking whether it could have emerged parallel to language.
I think your point is interesting -- what the Sapir-Whorf Hypothesis attempts to explore is the possible influence of language on idea of the world. What I have always wondered is whether or not this means altering our cognition or our *perception*. Reports of perception are, of course, extremely suggestible. Perhaps the Sapir-Whorf Hypothesis can tell us more about the malleability of sensory evaluation (as opposed to actual perception).
DeleteAs for propositional thinking, I think it should be examined separately from the Sapir-Whorf Hypothesis. The development of propositional thinking must be central to UG -- it is precisely this type of language-based thinking that leads us to speak, write, express, etc. using propositions. But I don't think that propositional thinking necessarily changes the way that we see/experience the world. It is interesting to consider (although it is impossible, given our inherent UG-reliant communication) what it would be like to experience the world without thinking in propositions. Is this our closest shot at experiencing the world through the perspective of a non-human?
But as we've discussed in class, if propositional language emerged at *some point* what was the benefit that allowed it to be selected for? Does it yield more precise communication? Is it a more successful vehicle for expression (that is, it allows us to say whatever there is to say)? There are few convincing ideas as to what this evolutionary benefit may be, though it clearly separates us from our closest ancestors. Whatever the benefit, I am not sure propositional language changes the way we perceptually experience the world. Propositions allow us to *declare* members for non-members through our sensorimotor feature extraction. But propositional thinking is not necessary for categorization (lizards categorize through their sensorimotor experience with the world, too). So, if propositional thinking is not necessary for categorization, nor can we explain propose its evolutionary benefit, perhaps this is why we turn to language acquisition (rather than language evolution) to explore this unique human trait. Evolution, in the mean time, has not proven a good place to start.
I agree with Gili that propositional thinking should be examined separately from the Sapir-Whorf Hypothesis. Moreover, I do believe that propositional statements are just labels for thoughts/cognition; if we look at the tip-of-the-tongue phenomenon, we can clearly see that there is a thought in the absence of any proposition, so it must be that cognition, at least in this case, precedes language. Although, as Veronique asked "is language just a series of labels for cognitions that already exist or does language in itself involve learning how to think differently." I don't think that language is directly causally involved in learning how to think differently but I do think that the two go hand in hand. For the simple reason that apes cannot truly learn language in the same way the humans do because they lack the cognitive capacities, I believe. However, I don't think that language would allow them to suddenly think differently, I believe that language and a certain kind of cognition would have to emerge together, with the cognition likely preceding the language abilities, although I'm just thinking out loud at this point
DeleteI take the same issue with this paper of Pinker's that I did with the first reading for week 8: he merges the ideas of OG and UG into the same one. Further, Pinker argues that this blended OG-UG grammar is learned by using examples *from OG*!
ReplyDeleteSo, while Pinker is right that OG can be taught, by giving corrective feedback to a child who doesn't understand, for example to pluralize foot as feet and not foots, he says nothing about the unlearnability of UG.
The reason UG cannot be learned, as we've discussed in class, is that there are no examples of UG-noncompliance. This is the poverty of the stimulus argument. In order to learn some category (in this case, what the rules of UG are), the child must be exposed to examples of things that are members of the set of things that are UG-rule-compliant *and* members of the set of things that are UG-rule-noncompliant. Because the child is not exposed to anything in the latter category, the child cannot learn UG.
This is the essence of what Chomsky says (and what Pinker specifically misunderstands): that UG must be innate in the child because of poverty of non-UG stimulus.
I want to quickly back up your point with a quotation from the paper that shows Pinker's misunderstanding of UG...
Delete"Innate knowledge of grammar itself is not sufficient. It does no good for the child to have written down in his brain "There exists nouns"; children need some way of finding them in parent's speech, so that they can determine, among other things, whether the nouns come before the verb, as in English, or after, as in Irish." (p20)
Pinker argues well in this paper for the complexities involved in language learning. He also deconstructs the gaps in various possible mechanisms of language acquisition in children, and argues as to why language is a different cognitive capacity with special features that are distinct from other cognitive processes. Although, as others have pointed out, Pinker does not use UG to argue for the uniqueness of language acquisition, an argument that would have strengthened his stance.
ReplyDeleteIn the first few pages, when Pinker describe the learning strategy that children use, and how they form hypotheses which they check against information from the environment, I’m assuming he is referring to OG and not UG or implicit hypotheses. Since UG is not learnable with felt or implicit hypotheses, I assume he is referring to OG.
To clarify the relationship between language acquisition device (LAD) and grammar, I’m wondering if we need LAD to learn OG. Or can OG be learnt in a similar manner as learning other categories, with negative feedback and examples of non-members?
There is also an inherent contradiction in Pinker advocating Chomsky’s view on language, including UG and LAD, and simultaneously arguing for the evolutionary origin of language. Chomsky seems to have suggested (with no claim for correctness), that perhaps the ability for language was an emerging property of a complex nervous system (https://www.newyorker.com/magazine/2003/03/31/the-devils-accountant). It could also be a property of a dynamic system like the infant brain, undergoing considerable change across time in the first few years (Pinker describes the latter point in this paper). Interestingly, there are certain researchers working on consciousness (what we call feeling), who hypothesize that the ability to feel is also an emerging property of a complex system. Perhaps, the features/ability/force that cause us to feel something and learn a language overlap with each other.
"Stromswold (1994) has an even more dramatic demonstration that parental feedback cannot be crucial. ... The boy's abilities show that children certainly do not need negative evidence to learn grammatical rules properly, even in the unlikely event that their parents provided it."
ReplyDeleteThis confuses me. According to the distinctions between UG and OG, UG is what we do not have access to its negative feedback and thus cannot be learned, whereas OG is what's learnable since both positive and negative data exists during language acquisition process. However, in this experiment, it seems the child does not even have access to negative feedback for OG: he was congenitally unable to talk and yet he was able to distinguish grammatical from ungrammatical, purely by listening to others. Though the content of listening has not been specified in this paper, it still could be the case that he had acquired some negative evidences from others' conversations, chances are that that couldn't be too much and most of the data are still positive. I believe that's also why the author has arrived at the conclusion that "children certainly do not need negative evidence to learn grammatical rules properly". Then it actually shows that even for OG (grammatical rules specific to a language), there's still occasional "the poverty of stimulus" problem, at least for some children, which means probably OG is also innate to them?
Or am I making a mistake here? Is that the specific grammar rules of the language is still UG, which is innate to the child, and what he has learned from listening is just setting the parameters for that language? which means his ability to correctly set parameters without negative feedback is also innate?
I must admit UG is really a confusing topic to me, especially because I don't really know what it really is (its rules, settings, inputs/outputs, etc.)...
I am not sure I understand your question correctly but I think, here, it is difficult to say that if that shows if even OG can make a “poverty of stimulus” argument. Just as you said in the third paragraph, here we are not sure if the negative evidence and ungrammatical sentences are picked out because of the violation of OG or if it is just a broad violation of UG, which is also innate, that does not OG to play a role.
DeleteAlso, I think your response reminds me of the discussion of “uncomplemented learning” in lecture. Does this example suggests a counter-example of “uncomplemented learning”? As the child does not speak and are rarely exposed to ungrammatical sentences that are pointed out from others. So little should the children know about what categorizes the grammatical sentences from the ungrammatical ones. Yet, the fact is that, without such an experience of being instructed about the negative category, the children can still perfectly picks out what belongs to the category and what does not.
In regards to you saying that child would not have access to negative feedback... I think that the child would actually have negative feedback, but in a different way. When they think in their head, they may say something a certain way and if it's ungrammatical they would notice through listening when it comes up in a conversation another time. Also, it's clear that he's learned grammar correctly from being able to perform perfectly when tested to see if he could understand complicated sentences.
DeleteHello!
DeleteI will try to see if I can clarify what I understand so far.
UG is innate and cannot be learned via supervised learning, unsupervised learning or instruction. Only Chompskian linguists learn UG by using empirical methods and only they could provide negative examples for it. UG is innate and is not learned by children because we have no negative examples of this provided to us when we are young. It is not possible for us to have non compliant UG thoughts, this never occurs! There are specific parameters for each language that tell you which version of UG applies to your language (example pronoun dropping). The evolution of UG is difficult to solve because it is innate.
Children certainly do not need negative examples of UG to learn correct grammar and to learn how to speak “correctly”. There are not negative examples provided because it is not learnable, it is innate.
OG is learnable! OG is learnable for particular languages. OG can be learned by using supervised learning, unsupervised learning and explicit instruction. I think it’s important to mention that OG is learnable because we have access to negative examples. I know this is contrary to what Pinker is saying, but when you think about it, kids are frequently provided with feedback about the incorrect utterances that they form. Children are constantly being corrected and given feedback on how and how not to say things by their parents. I believe that Pinker is incorrectly using UG and OG interchangeably in his paper… I think he is getting them mixed up (I believe that this point has been confirmed by Professor Harnad in a previous comment). I believe that in the Stromswold example that Pinker provided to prove that children do not need negative feedback to learn, he is mixing up UG and OG. The fact that Pinker is conflating UG and OG might be the stem of the confusion.
In the section about Motherese, it comes up that children in some communities learn language by being instructed on the proper sentences to say in a situation by their mothers. In a case like this, is the argument of whether language is innate or learned solved in a way? I of course don't think it's that simple, but I think that this example is one that should be looked further into when considering all the other hypotheses concerning this debate that originate from individuals considering languages like English in places such as North America.
ReplyDeleteIn the paper, Pinker explains one problem he has with Maratsos' and Chalkley's 1981 proposal. Pinker says "How is a prelinguistic child supposed to find the "subject" of the sentence in order to correlate it with the ending on the words he or she is focusing on?"
ReplyDeleteI want to argue that I actually think it's possible for a prelinguistic child to determine the subject of a sentence. I think that it's somewhat intuitive to hear a name and understand that it is the topic of concern in the sentence. I think that children have specialized hearing for names they recognize or have heard before because they've connected it to a physical object or person through inductive learning, and also there's an intonation style associated with pronouncing something/someone's name. So, I don't think that children necessarily have to know the definition of "subject", or even know how to form words themselves, to pick out the important part of a sentence (the subject). All this to say that I think Pinker's rejection of Maratsos' and Chalkley's proposal is flawed.
Hi Naz,
DeleteI agree with you that a prelinguistic child can determine the subject of a sentence. It might also be something that some other mammals have like dogs. Whenever we call them, they can understand they we are referencing the them. Therefore, maybe there is something innate that humans and dogs share.
"On this view, the child would set parameters on the basis of a few examples from the parental input, and the full complexity of a language will ensue when those parameterized rules interact with one another and with universal principles. The parameter-setting view can help explain the universality and rapidity of the acquisition of language, despite the arcane complexity of what is and is not grammatical (e.g., the ungrammaticality of Who do you think that left?). When children learn one fact about a language, they can deduce that other facts are also true of it without having to learn them one by one."
ReplyDeleteIs Pinker, with his parameter setting hypothesis referring to UG? He seems to be mixing OG and UG here, by saying that children , would set parameters, on the basis of some examples from parental input, and with which parametrized rules will interact with one another and with universal principles. Are the universal principles he is referring to UG? How can these parametrized rules interact with one another and with universal principles if the universal principles and parameters innate? I don’t know if I am confusing the set of parameters with UG or with OG parameters?
Within Pinker's paper he conjoined UG and OG in a way as he is pointing to the idea that UG is learned. Honestly this paper confused me a bit because I was unsure at times what Pinker was arguing for. I was confused if he was arguing for UG or not. There are many aspects of Pinker's paper which provides evidence for UG - and supports many aspects of UG that we have been discussing throughout this course. "Every speaker of English judges sentences such as I dribbled the floor with paint and Ten pounds was weighed by the boy and Who do you believe the claim that John saw? and John asked Mary to look at himself to be ungrammatical. But it is unlikely that every such speaker has at some point uttered these sentences and benefited from negative feedback." This passage is talking about negative feedback and states that people never express/say things which are not compliant to UG rules. Therefore they would never receive negative feedback for these incorrect statements. This idea of negative feedback has also been referred to as poverty of the stimulus and is one of the main arguments for UG. This idea suggests that if if people never really violate the rules/laws of UG then it must be innate.
ReplyDelete"It surely must help children when their parents speak slowly, clearly, and succinctly to them, but their success at learning can't be explained by any special grammar-unveiling properties of parental babytalk. " This is another instance which supports UG and its unlearnability. This quote addresses the fact that it doesn't really matter how parents/care-takers speak to children because the children will nonetheless be successful at learning language. This is seen even in cases where children are neglected and no one really speaks to them at all. Even if children are not explicitly exposed to statements/speaking that is UG compliant, they will still be able to speak in a UG compliant way.
Pinker supports UG and its unlearnability, yet he incorrectly uses examples of OG (which are learnable) to support his claims.
DeleteWhile motherese is widely used, it has not been shown to be the optimal way for children to learn language. Research has shown that children learn more vocabulary and learn quicker when parents speak normally to their children rather than adjusting their register and language when talking to them. This information is supported by other studies which show that parents vocabulary (and education level) directly impacts children’s language development, showing knowledge of up to double the words at the same age when exposed to more words. It is true that the coaching that some parents engage in is not entirely beneficial, let alone is the approach of those who believe they must not speak to their children. As shown by cases such as the Romanian orphans (overcrowded orphanages lead to children growing up with minimal attention or interaction with other, in their cots), children need human interaction in order to learn. While direct attempts at teaching vary and may not always be beneficial, exposing children to more complex language has been shown to at least increase their vocabulary much quicker.
ReplyDeleteI feel like much of what I wanted to say has been said already, rather than just enter the echo chamber, I’ll just give something that I personally (someone who has zero linguistic background) had found an interesting parallel. As an avid student of Anthropology, the notion that language influences the way people think it one that I had always a lot of trouble with, as despite its pervasiveness, it had been debunked long ago. Not that it was all too complicated -- a common example used was the notion of ‘sound blindness’ first registered by famous German-American Anthropologist Franz Boas in his field work with various American Indigenous groups in the early 20th century. One of his main purposes was to debunk the ascribed ‘primitivity’ of indigenous peoples and minorities that was becoming common due to the emphasis on eugenics at the time. In doing so, he catalogued the languages of the indigenous groups, which were often labeled to have a simplistic vocabulary entirely ill-equipped for what was termed ‘civilized thought’. Conversely, Boas registered that the indigenous languages had in fact developed an extensive vocabulary which paralleled the complexity of vocabulary of the common western languages like French, English and German. In fact, much of the earlier studies which tried to decipher the still foreign languages (particularly of the indigenous arctic peoples) had actually missed most of the words spoken by those they interviewed over missing the nuanced and complex intonations which were altogether unintelligible to the western-trained ear. While that fact alone gives credence to Pinker’s discussion of how language learning capacities cement themselves over time (languages are learned), the intelligibility of the indigenous languages after they had been learned and understood likewise proved that these languages also shared an innate syntax which governed the use of the language, despite how foreign sounding the words were to the early anthropologists. (UG)
ReplyDelete