Chomsky, according to Pinker, argued that kids learn the rules of the program from just seeing a bit of the computation. To get the algorithm in the normal way, you need the whole computation. Since kids could figure out the rules without explicit instruction of those rules, they must be tuned to look for the pattern in the first place. Kids have a leg up on deducing grammatical rules. Those constraints are Universal Grammar. I still don't get it.
Why should it be "species-specific?" Speechless bonobos and orangutans seem more like a reticence to talk, as though were they to open their mouths, they would share with us some horrible truth we're not yet ready to here (Ishmael promotion). Just kidding, but seriously. How do we explain the lack of motivation that S. Harnad mentioned?
"Presumably language evolved in the human lineage for two reasons: our ancestors developed technology and knowledge of the local environment in their lifetimes, and were involved in extensive reciprocal cooperation. This allowed them to benefit by sharing hard-won knowledge with their kin and exchanging it with their neighbors"
Yo, isn't this yo-he-ho?
"Williams Syndrome, an inherited condition involving physical abnormalities, significant retardation (the average IQ is about 50), incompetence at simple everyday tasks (tying shoelaces, finding one's way, adding two numbers, and retrieving items from a cupboard), social warmth and gregariousness, and fluent, articulate language abilities (Bellugi, et al., 1990)."
"Most adults never master a foreign language, especially the phonology, giving rise to what we call a "foreign accent."" The exception? Williams Superheroes. "William syndrome presents this remarkable separation of language from other domains." Music too.
Discontents/Misunderstandings regarding Learn-ability Theory
1. "the target - language is the one spoken in their community." Language allows you to say anything that can be said. Learning a target language suggests a child is somehow not learning the other languages. If you have one though, you have them all. Treating language as anything less misses the point from the get go (sorry about the cliches). 2. Environment: "This is the information in the world that the learner has to go on in trying to acquire the language." What, like variant data? Already in the womb the kid makes statistical inferences regarding phonology, so again, it seems like this definition talks passed Universal Grammar. 3. "The learner, using information in the environment, tries out "hypotheses..."" Roger Brown's original word game. Every new word itself is a hypothesis for the child that something about which she doesn't know exists in the world. Learn-ability theory calls the hypothesis generator the language organ. So, curiosity?
" The literature on "unsupervised learning" in computational linguistics is demonstrating that algorithms can be learned much more from text than most linguists would have thought possible. A substantive de- fense of the APS will have to address this sort of research."
The first thing that I thought of at the end of the article was that the APS was just considered true on pain of contradiction and this quote reminded me of that. I see the point of the authors to try to prove it wrong even if all the attempts to do so have failed at it. the fact that we don't know the mechanisms by which language is acquired makes the UG look like another just so story. The fact that people don't learn to formulate language after being left alone in a forest or in a basement means that there is need for some sort of stimulus in order for the child to develop his language but at the same time children with no acquired language left together, without the presence of any adult to convey an example of what is language, eventually creates a language of their own. This alone kind of convinced me of the theory of UG. Moreover this example made think that to answer the question of how is it that we acquire language as children might not be answered by focusing on the aspects of how acquire today's language but more on what is happening between those 2 children left alone who formulate a language of their own. There is clearly something going on here which is very important and might shed some light on the way acquire language. I don't know exactly what sort of language these 2 children might create but it would sure be interesting to understand what they created because it could give us some insight on how they really started their communication and how it later developed. It could give us a trace of their mental work (parkour) which could itself help us understand how do children start to cognitively grasp what is language and how to replicate it.
"APS still awaits even a single good supporting example."
Like i said in the beginning of this commentary, APS might just be true on pain of contradiction or be another just so story but I don't think that the arguments against APS will be found through the method they have been exploiting in this article which is to find a acquirendum and see if it can be learned only through accessibility of the lacuna because in every case it will be shown that there isn't enough examples for the children to learn the rule, maybe the child is is learning through another process that doesn't involve the specifics of the rule he eventually acquires.
''But each success is likely to eliminate or weaken some particular apparent APS instance, by revealing that more can be learned in a data-driven way than was previously though.''
In the article, each instance which delegitimizes proof of APS does nothing to legitimize the data-driven, or domain general way. Why he jumps to this assumption is unclear to me. Instead this article draws its strength on being critical of APS and by calling on APS proponents to go deeper and be unsatisfied with the progress they've made. I agree with Pullum that the limits of domain general acquisition need to be understood. That is, we need to know how what can be learned given a certain amount of stimulus. Unfortunately, I don't see there being a general rule of stimulus to learning ratio as children themselves and and the rule or object being understood will probably vary in its stimulus requirement. One method is to create a computer which begins as a tabula-rasa, with just one learning algorithm, and testing how much it can learn of language depending on the amount and type of stimulus. This gets complicated because we don't even know what the human general learning algorithm is, nor if there is one. Regardless, this might give us insight into what it would like look if humans had one, and to the possibilities of acquiring language in a domain general way. That being said, I think empirical proof of either domain general, or domain specific (i.e. UG) will be hard to come by, making it easy to undercut current research. The amount of control needed and the absurd lack of such control in studying language acquisition makes it an almost impossible empirical endeavour, Though Pullum's article is crucial in shining doubt on the uncritical certainty of the APS, it is an easy article to write given the inherent difficulties in finding empirical proof.
Chomsky, according to Pinker, argued that kids learn the rules of the program from just seeing a bit of the computation. To get the algorithm in the normal way, you need the whole computation. Since kids could figure out the rules without explicit instruction of those rules, they must be tuned to look for the pattern in the first place. Kids have a leg up on deducing grammatical rules. Those constraints are Universal Grammar. I still don't get it.
ReplyDeleteWhy should it be "species-specific?" Speechless bonobos and orangutans seem more like a reticence to talk, as though were they to open their mouths, they would share with us some horrible truth we're not yet ready to here (Ishmael promotion). Just kidding, but seriously. How do we explain the lack of motivation that S. Harnad mentioned?
"Presumably language evolved in the human lineage for two reasons: our ancestors developed technology and knowledge of the local environment in their lifetimes, and were involved in extensive reciprocal cooperation. This allowed them to benefit by sharing hard-won knowledge with their kin and exchanging it with their neighbors"
Yo, isn't this yo-he-ho?
"Williams Syndrome, an inherited condition involving physical abnormalities, significant retardation (the average IQ is about 50), incompetence at simple everyday tasks (tying shoelaces, finding one's way, adding two numbers, and retrieving items from a cupboard), social warmth and gregariousness, and fluent, articulate language abilities (Bellugi, et al., 1990)."
"Most adults never master a foreign language, especially the phonology, giving rise to what we call a "foreign accent."" The exception?
Williams Superheroes. "William syndrome presents this remarkable separation of language from other domains." Music too.
Discontents/Misunderstandings regarding Learn-ability Theory
1. "the target - language is the one spoken in their community." Language allows you to say anything that can be said. Learning a target language suggests a child is somehow not learning the other languages. If you have one though, you have them all. Treating language as anything less misses the point from the get go (sorry about the cliches).
2. Environment: "This is the information in the world that the learner has to go on in trying to acquire the language." What, like variant data? Already in the womb the kid makes statistical inferences regarding phonology, so again, it seems like this definition talks passed Universal Grammar.
3. "The learner, using information in the environment, tries out "hypotheses..."" Roger Brown's original word game. Every new word itself is a hypothesis for the child that something about which she doesn't know exists in the world. Learn-ability theory calls the hypothesis generator the language organ. So, curiosity?
" The literature on "unsupervised learning" in computational linguistics is demonstrating that algorithms can be learned much more from text than most linguists would have thought possible. A substantive de- fense of the APS will have to address this sort of research."
ReplyDeleteThe first thing that I thought of at the end of the article was that the APS was just considered true on pain of contradiction and this quote reminded me of that. I see the point of the authors to try to prove it wrong even if all the attempts to do so have failed at it. the fact that we don't know the mechanisms by which language is acquired makes the UG look like another just so story. The fact that people don't learn to formulate language after being left alone in a forest or in a basement means that there is need for some sort of stimulus in order for the child to develop his language but at the same time children with no acquired language left together, without the presence of any adult to convey an example of what is language, eventually creates a language of their own. This alone kind of convinced me of the theory of UG. Moreover this example made think that to answer the question of how is it that we acquire language as children might not be answered by focusing on the aspects of how acquire today's language but more on what is happening between those 2 children left alone who formulate a language of their own. There is clearly something going on here which is very important and might shed some light on the way acquire language. I don't know exactly what sort of language these 2 children might create but it would sure be interesting to understand what they created because it could give us some insight on how they really started their communication and how it later developed. It could give us a trace of their mental work (parkour) which could itself help us understand how do children start to cognitively grasp what is language and how to replicate it.
"APS still awaits even a single good supporting example."
Like i said in the beginning of this commentary, APS might just be true on pain of contradiction or be another just so story but I don't think that the arguments against APS will be found through the method they have been exploiting in this article which is to find a acquirendum and see if it can be learned only through accessibility of the lacuna because in every case it will be shown that there isn't enough examples for the children to learn the rule, maybe the child is is learning through another process that doesn't involve the specifics of the rule he eventually acquires.
''But each success is likely to eliminate or weaken some particular apparent APS instance, by revealing that more can be learned in a data-driven way than was previously though.''
ReplyDeleteIn the article, each instance which delegitimizes proof of APS does nothing to legitimize the data-driven, or domain general way. Why he jumps to this assumption is unclear to me. Instead this article draws its strength on being critical of APS and by calling on APS proponents to go deeper and be unsatisfied with the progress they've made. I agree with Pullum that the limits of domain general acquisition need to be understood. That is, we need to know how what can be learned given a certain amount of stimulus. Unfortunately, I don't see there being a general rule of stimulus to learning ratio as children themselves and and the rule or object being understood will probably vary in its stimulus requirement. One method is to create a computer which begins as a tabula-rasa, with just one learning algorithm, and testing how much it can learn of language depending on the amount and type of stimulus. This gets complicated because we don't even know what the human general learning algorithm is, nor if there is one. Regardless, this might give us insight into what it would like look if humans had one, and to the possibilities of acquiring language in a domain general way. That being said, I think empirical proof of either domain general, or domain specific (i.e. UG) will be hard to come by, making it easy to undercut current research. The amount of control needed and the absurd lack of such control in studying language acquisition makes it an almost impossible empirical endeavour, Though Pullum's article is crucial in shining doubt on the uncritical certainty of the APS, it is an easy article to write given the inherent difficulties in finding empirical proof.