Saturday, 2 January 2016

9b. Pullum, G.K. & Scholz BC (2002) Empirical assessment of stimulus poverty arguments

Pullum, G.K. & Scholz BC (2002) Empirical assessment of stimulus poverty arguments. Linguistic Review 19: 9-50 

This article examines a type of argument for linguistic nativism that takes the following form: (i) a fact about some natural language is exhibited that al- legedly could not be learned from experience without access to a certain kind of (positive) data; (ii) it is claimed that data of the type in question are not found in normal linguistic experience; hence (iii) it is concluded that people cannot be learning the language from mere exposure to language use. We ana- lyze the components of this sort of argument carefully, and examine four exem- plars, none of which hold up. We conclude that linguists have some additional work to do if they wish to sustain their claims about having provided support for linguistic nativism, and we offer some reasons for thinking that the relevant kind of future work on this issue is likely to further undermine the linguistic nativist position. 


  1. 9a. This article takes the time to find evidence against 4 of the most cited examples of the argument of the poverty of the stimulus. In section 4.2 auxiliary sequences, Kimball claims that children know which order to place various pieces of language (in this case finite/non-finite auxiliaries and verbs) despite being presented the “crucial evidence”. The article touches upon the notion of this “vanishingly rare” evidence as a metholodological issues. Not only do linguists need to identify how many examples with complex structure a child is exposed to, but they also mention it is important to know what the child pays attention to.
    Later in section 4.4 the authors explain the assumption that “children encounter (or pay attention to) simple sentences prior to those with subjects containing finite relative clauses.”

    My impression of this article is that its goal is to encourage proponents of APS to look for better evidence, and seek more sound methodology to support the hypothesis. Many of their arguments seem to be centred around their ability to find a couple of counter examples which oppose the some of the more drastic/strong claims of the APS- that say examples of certain linguistic features are completely absent in a child’s environment. It appears that the authors seem content to have poked these holes in the APS using these few counter examples. However, in my opinion they may be lacking a huge piece of the puzzle that they touched upon but haven’t elaborated. This piece of the puzzle is attention. What is meant by the term attention?
    In section 4.2 is seems that “encountering” something is different from “attending” to it
    “This involves knowing something about both input (what utterances occur in the childís vicinity) and uptake (what the child pays attention to). “

    But later in section 4.3 they seem to be treating these two terms – encounter and attention- as synonyms, “APS instance based on auxiliary-initial clauses ìis based on an empirical assumption that children encounter (or pay attention to)…”.
    What is attention and how does it play into the interpretation of this data from an APS perspective as compared with an experience-based learning perspective?

    1. Pullum, like Pinker, mixes apples and oranges (UG grammar and non-UG grammar). The problem is the lack of negative evidence in what the child hears and says. It's all UG-compliant. I tiny percentage of lapses in the written literature is certainly no solution to the child's lack of negative evidence.

      I can't see how attention helps one way or the other with the problem of the poverty of the stimulus. Attention is related more to whether a category is learned explicitly -- either by being told the rule, or by figuring it out through deliberate, conscious, explicit hypothesis-testing (the way the teams of MIT linguists do it, while trying to test UG violations).

      But a lot of learning is implicit -- our brains do it for us (the way they find the name of our 3rd grade school-teacher), delivering it on a silver platter: And that applies to ordinary grammar (which is learned, plenty of positive and negative instances and correction) as well as to UG. We know a lot of ordinary grammar rules implicitly, in the sense that we know how to speak in a way that complies with them, yet we don't know the rules (unless they have been explicitly taught to us): we can't state what they are.

      So attention might be relevant with learnable rules that we learn explicitly. But it doesn't resolve anything about UG.

  2. - What I got from this article was that our innate language capacity should be quantified rather than generalizing as all being innate, and that both data-driven learning and biological mechanisms (if more research is done to find what the specific mechanisms are even involved in) play a role in language processing. The paper suggests that we are somehow predisposed to learning, that we’re innately primed, but it’s unclear to me whether this means that the language specialized regions, if any, is there to specifically process language, or that they’re involved in language processing simply because they’re suited for it/good at it. I guess the question is, are those domain-specific innate mechanisms specialized for first language acquisition there because evolution had meant for us to be equipped with those mechanisms? Or is it similar to how mirror neurons get their function, with the areas involved in language acquisition learn from being exposed to the lacuna present to a child, and taking on the role of language acquisition as more experience is gained?
    - I think by pointing out more research is needed see what DDL procedures can do really emphasizes on his earlier point on how they doubt most speakers could exert enough conscious control over the syntax of their utterance. I think it emphasizes on how we take language for granted because we use it so much that we don’t realize its complexity, and perhaps computational modelling will bring to light, the mechanisms involved in our almost automatic production of speech.
    - In terms of unusualness and sentences children may never be exposed to, such as the example from “The Importance of Being Earnest,” I wonder if children are able to form unusual sentences, such as those found in Jabberwocky, where semantics don’t play a role, only structural relations.

    1. Pullum fudges the relevant issues: Are we talking about UG rules or ordinary grammar rules? If UG, the only thing that's relevant is whether the child gets only positive evidence (only hearing and saying UG-compliant utterances) or also gets the negative evidence needed to allow the brain (or any learning system) to learn the rules that generate all and only the UG-compliant utterances by trial and error. The child does not get that negative evidence. And none of the tiny statistical oddities Pullum cites changes that in any way.

  3. This article shows the lack of evidence supporting the lack of stimulus argument that is one of the main supporting claims of the existence of an universal grammar.
    While analyzing the evidence or the lack of evidence for the poverty of the stimulus theory, after establishing their Argument from Poverty of the Stimulus, the authors search for a grammatical sentence that children do not hear at an early age but are still able to understand. Finding such a sentence might help in finding more evidence on if children’s grammar is innate or not. If the children recognize the sentence as grammatical, despite never hearing it before, then this would provide evidence of an innate component of grammar. If children fail to recognize the sentence as grammatical, then this would be evidence against the Universal Grammar. I am unsure about the utility of finding such as sentence, however. This sentence, when found would be used to find the answer for the C part of the Argument from Poverty of the Stimulus (APS).
    I have trouble understanding why some aspects of actions that people do are not used to better understand language. Cooking, making good food, is not innate. The appearance of spices and taste enhancers developed way to quickly to make cooking with these components innate. However, there are some foodstuffs that people would never mix together (except for a prank). By learning which foodstuffs your parents and other caregivers put together or by reading recipes, you know which constituents are a good mix and which ones are not. There seems to be some universality in cooking styles. However, the non-innateness of cooking is never questioned. Why cannot there be the same learned classification about language grammar constituents happen in humans where people learn which constituents go together and which ones do not by observation?

    1. Learning to speak UG-compliantly is not like learning to cook (whether the recipes are learned from explicit instruction or by trial and error). There is no instruction and there is no error. (Learning conventional grammar is like learning to cook. But that's ordinary category learning. No poverty of the stimulus. Plenty of positive and negative evidence.)

  4. Several comments/questions about this reading.

    First, while I was disappointed that Pinker didn’t really address the issue of the poverty of stimulus, I was also somewhat disappointed with this paper in the sense that I thought Pullum & Scholz tried to down play or minimize the importance of the lack of negative evidence in all of this … I mean, I thought understanding the lack of negative evidence was one of the major components of this puzzle, yet many of their examples don’t even touch this issue. I think understanding the lack of negative evidence is at the heart of this problem – and I am a little bothered that they didn’t discuss this much …

    Second, relating this back to our bigger question of “explaining how it is that we do the things that we can do”; it seems that UG would have to be built in our T3 robot & it is probably innate based on all the observations as well as the argument of the poverty of stimulus (and some linguists have been doing great work of “reverse-engineering” our language capacity & it’s establishing it’s parameters). So now, if we can say that UG is innate, this lead me to think of other questions, namely how and why did it evolve?

    I mean, if we cannot learn UG, then our capacity has to be inborn – but where did it come from? I recently read a paper by Jackendoff and I thought he brought up an interesting point concerning UG, namely that if it is “non-learnable”, doesn’t that also make it “non-evolvable”? So, did it evolve? If so, how? But also, what is its adaptive value? But is ultimately knowing the “how it evolved” help us explain “how is it we can do everything we can do”? I need to do more research in this.

    Lastly, this had me thinking about UG and categories. One proposal for the evolutionary adaptive value of language is that it helps us learn categories through linguistic instruction. But how sure can we be of this? I am not dismissing it, but I just don’t find it all that clear why UG would be needed to learn categories through linguistic instructions. Any thoughts?

    1. 1. Yes, Pullum fudges negative evidence (mixing it up with other issues that are not relevant). Like Pinker he also mixes up UG issues and ordinary grammar issues.

      2. Yes, if UG isn't learnable by the child, hence innate, that still leaves the question of how it evolved.

      3. The hunch that the adaptive advantage of language is that it makes it possible to learn new categories by instruction has to be kept separate from the question of the adaptive advantage of UG itself (which is unclear).

  5. “We also caution the reader against reading this article as a defense of the claim that purely empiricist language learning, via domain-unspecialized algorithms for knowledge acquisition, can suffice for learning natural languages, given children’s experience. It may well be that purely empiricist learning of natural languages can indeed be shown to be impossible. That possibility is left open here. We are concerned not with whether empiricist claims about language learning are true, but we are concerned with whether linguists have established their falsity.”

    Regarding the falsity of empiricist learning of natural languages (according to linguists), the authors neglect to even mention the number of discrete language-based impairments that people can face. Pullum & Scholz acknowledge that most neurotypical humans learn (or even master) a language within a few years. But there are a non-negligible number of instances where otherwise cognitively and socially normal children are unable to learn a language. Similarly, Pinker presented the argument of children with William’s syndrome, who have very low IQs but extremely impressive language abilities. Taken together, this information provides some evidence for domain-specific processing of language in humans and not a domain-general view (which seems to be what the empirical language learning argument is based on).

    Strictly speaking, what I’m proposing isn’t supporting the poverty of the stimulus argument. I’m just saying that Pullum & Scholz have to consider a lot more than the poverty of the stimulus argument if they want to look at why linguists don’t believe that empiricist language learning is likely, or even possible.

    1. Poverty of the Stimulus: The Burden of Proof

      1. "Empiricism" is too fancy a word. The question is just whether children can learn the rules of UG from the data available to them when they are learning language. The answer is: no. They don't get "negative data" (non-members of the UG category) -- which means examples of utterances that violate the rules of UG. They not only never hear violations of UG; they never speak violations of UG: They don't make UG mistakes, so they don't get corrected for them. Everything they hear and say is "positive data" (members of the UG category: utterances that obey the rules of UG).

      It's not that it's impossible to violate UG. Linguists do it all the time, deliberately, in testing what are the rules of UG, seeing what utterances they would generate, and whether they sound grammatically right or wrong to their (UG-prepared) ears. But the child never hears such utterances (unless its parent is a UG linguist!)

      Pullum -- with all his examples from spoken and written language -- provides virtually no violations of UG. Most of his examples have nothing to do with either negative evidence or UG.

      So Pullum does not show that there is anything to doubt poverty of the stimulus (which, by the way, is a form of dramatic underdetermination!). UG rules are not learnable from the data available to the child. So they must be innate.

      2. There are language-specific impairments and abilities, and they can be inherited, hence genetic. But they are not specific UG impairments, hence not relevant to the question of the poverty of the stimulus. There's nothing problematic about language-learning ability (in general) being innate. It's clearly part of the human genome. The problem is UG in particular. (See reply to Cait, above.)

      3. Pullum has shifted the "null hypothesis." The burden of proof is not on those who say UG is unlearnable, because negative data are lacking. The burden is on those who argue that, despite the its underdetermination by the data available to the child, UG is nevertheless learnable, and learned. (They either need to show that there is negative and positive data, from which a learner can abstract the rules (implicitly or explicitly. Alternatively, they need to show that UG is wrong, hence not needed to explain language ability.)

    2. In regards to #3 of this question (going to use kid sib vocab)
      So now the people who say that despite the low input given to the child to learn all they learn of UG doesn't prove UG is unlearnable, and that it is not innate are the people who have to explain themselves. This makes sense.
      Would they be able to try and refute this burden of proof by using machine learning and AI that is able to learn languages and become a better speaker through practice, and eventually is able to recognize grammatical structures that are not input directly to it?
      Not a linguistics student so might be misunderstanding.

    3. "There's nothing problematic about language-learning ability (in general) being innate. It's clearly part of the human genome. The problem is UG in particular."

      Based on the distinction between general language learning ability (which is understandable because language implies meaning as well as syntax, and a myriad of other elements too i would imagine, not having taken linguistics im unsure of all the factors) and UG I am not sure i fully understand the particularity of UG and how exactly it is distinguishable from normal grammar (though I do understand the purpose of the distinction because quite clearly, 'normal grammar' is so flexible across languages). I am essentially unclear on what would the 'additional implications' of having a UG as opposed to a general encoded language ability (which as Harnad points out above is undoubted) would be!

  6. While this article was in no way entirely convincing, their closing thoughts sum it up well.

    “This article is far from exhaustive. But on the basis of the four cases reviewed here, which include what are unquestionably the most celebrated and widely cited instances of the APS, we believe it must be concluded that the defenders of the APS are still searching for a well-confirmed case of the type they need”

    “Although this article will no doubt be misread as a defense of empiricism, that is not an accurate characterization of it. We are agnostic at this point, perfectly prepared to accept that perhaps at some time in the future data-driven algorithmic techniques for language learning will hit a serious roadblock, or that credible evidence for innate task-specific language acquisition mechanisms will turn up.”

    This is the same effect of agnosticism that the article has left me with and I believe it is a good thing. In my undergraduate career, my linguistics minor and cognitive science as a whole has repeatedly told me what the poverty of the stimulus argument was and seemed to imply that it was widely accepted at this point. As far as I know this 2002 paper is the most recent one I have read on the topic and I am glad to see the “data driven algorithmic” techniques for language learning are not dead in the water. As someone who entered this course more or less a computationalist, I was in favour of believing such a data driven approach to language learning. Until this point I had assumed that the poverty of the stimulus was right simply because everyone kept saying “there’s no way children could get all the proper input required to learn language” and I believed them because I certainly wasn’t going to go through a corpus of language and check my own hypotheses against each sentence. This paper did a good job of (in my opinion) getting some of the tedious and boring work done of poking holes in the APS which certainly needed to be done with any argument that claims to have a sufficiently complete understanding of all types of sentences children hear. I may simply be too eager to reject APS though. APS has one of the same problems for me as UG and that is that it relies on some sort of innate mechanism that is not understood. I personally have trouble understanding innate mechanism in our brain that seem to give us certain mental capabilities without having to be learned. To just say that we are born with a predisposition to acquiring certain types of sentences just doesn’t sit right with me although I can accept it as true since the evidence points that way.

    1. I'm afraid Pullum has failed to refute the argument for the poverty of the stimulus.

      And there's nothing wrong or problematic about the ability to learn language being innate. (After all, we're the only species that can do it!)

      What's a problem is UG in particular. If UG is innate, how (and why) did it evolve? What is its adaptive advantage over non-UG-compliant languages? And did UG evolve all at once, or gradually, by bits? It seems too complex to just have been a point mutation. But it seems really far-fetched that (like fins or wings) it evolved bit-by-bit, each bit conferring more and more advantage...

  7. What I understood from this article is that linguistic nativism is the belief that some parts of language acquisition are innate, and the “argument from poverty of the stimulus” supports this. The argument of the poverty of the stimulus says that even without evidence of a stimulus introducing an aspect of language, children can still acquire that part of language. This article says that evidence for this argument is lacking and says that many people have mentioned the argument without stating it properly. Studies have tried to use arguments, involving the child’s ability to learn language and the child’s environment, to prove linguistic nativism. I may be misunderstanding here, but some of the properties of the child’s environment refer to ordinary grammar and not universal grammar. For example, one of the arguments is on Degeneracy, stating that “Children’s data-exposure histories include numerous errors”. For nativism to be supported, wouldn’t universal grammar have to be supported as opposed to any arguments on the development of ordinary grammar? I guess my main confusion that this article brings forward is, how do we use ordinary grammar to prove that aspects of language are innate? Can anyone clarify please?

  8. Sorry, I think I misunderstood the concept of degeneracy. Is it saying that although there are problems with the ordinary grammar in a child's environment, they still learn proper grammar? This would mean that the ability to learn the right "ordinary" grammar from the wrong "ordinary" grammar is innate.

    1. Hi Amanda,

      I am not sure I fully understand your question, but I will try my best to answer. Our language ability is constrained by Universal Grammar (which is an innate set of complex rules that appears in all languages). We have come to this conclusion because children only get positive evidence when exposed to language inputs i.e. we only hear/speak in UG-compliant utterances. So, children never hear/say non-compliant UG utterances, so there is no corrective feedback necessary for this type of grammar; thus it is innate. On the other hand, ordinary grammar is not innate. Children often say grammatically wrong utterances (“I breaked” instead of “I broke”), but are then provided with negative evidence for the brain to learn the correct ordinary grammar rules. So our ability to “learn the right ordinary grammar from the wrong ordinary grammar” is the same as the ability to learn anything as a child, we do it by trial and error, and instruction from others.

      I think degeneracy is just another support for the poverty of the stimulus argument stating that the utterances children are exposed to while learning language are often ungrammatical due to speech errors, false starts, slips of the tongue, and run-on sentences, which means that it is highly unlikely that the entire structure of language can be learned without some initial innate capacity and formal rules provided by UG.

  9. This comment has been removed by the author.

    1. This comment was meant for the 9a article and I accidentally posted it here. Sorry!

  10. When reading the article what the authors meant by 'input' was not entirely clear to me. What exactly classifies as 'input' is input? Is it just sounds that have been spoken? Or sounds that have been attended too? Could someone clarify please.

    1. Hey Lucy, I also struggled with the meaning of “input” for the first half of the article. However, there is one line that really helps illustrate exactly what Pullum and Scholz mean when they use this term:

      This involves knowing something about both input (what utterances occur in the child's vicinity) and uptake (what the child pays attention to). (page 29)

      The way I understand it, you're right, input is “just sounds that have been spoken.” If we were to attach digital recorders (with the same perceptive strength as a child's ear) to the sides of a child's head, then everything the digital recorders pick up would be considered input for the child. On the other hand, uptake is “sounds that have been attended too.” If the child is distracted by their dinosaur toy across the room, so much so that they are not paying any attention to the noise around them, then the grammatical phrases uttered during this time would go right over the child's head. They would only be considered input, not actual uptake.

      To put it in simpler terms, input is everything the child has the potential to hear, while uptake is everything the child actually hears. Hope this helps clear things up!

    2. So this is not really answering your exact question, but just something that bugged me in this article and your question made me think of it. So, the author defines input as “what utterances occur in the child’s vicinity”. Yet, the method used seems strange. For example to look sentences with two instances of copula (p.43), the text of the The Importance of Being Earnest is analyzed to find example. I simply don’t see how this data can be taken to be representative of what a child might be exposed two. I would expect a lot of discrepancies between the speech to which a child would be exposed, containing a lot of what are considered to be “errors” by formal grammar. Conversely, text might contain forms to which a child might rarely (if ever) be exposed to (e.g., Who is that young person whose hand my nephew Algernon is now holding in what seems to me a peculiarly unnecessary manner?).

  11. "All we will say here is that for those readers who are disappointed that we will not treat the negative data issue further, and feel that the APS we do discuss is not the one that really deserves the name, we can only suggest that they reinterpret our abbreviation "the APS" to stand for "the Argument selected by Pullum and Scholz” (Pullum and Sholz 17).

    I kept up with the first half of this reading, but got lost with the auxiliary related discussion at the end. I found their approach to addressing the “lack of positive evidence” to be interesting and to be valid. By showing me concrete examples of times when children can get positive evidence of certain components grammar, I express some doubt over the realm of UG. However, I still don’t understand what UG is since it’s being conflated with ordinary grammar and I am sure that this decreases the strength of their argument.

    Yet the most interesting thing to me about Universal grammar is how the lack of negative evidence (correcting of children) still results in successful language learning. Earlier, Pullum talks about piecemeal fashion learning to find new rules that prevent overgeneralization and I’m extremely skeptical of this argument just because we know that children do hit a stage of overgeneralization that dissipates quickly, without parents showing any sort of corrective feedback. For this article to truly have been effective, I think Pullum and Sholz should have tackled negative evidence. After reading 9a and 9b, I have one main question: what is the perceived balance of innate mechanisms vs experiential learning in grammar. In 9a, Pinker talked about the unimaginable size of permutations in grammar and language that would require some form of internal constraint. However, in this article we learn that some kids get exposure to over 30 million words before 3, which was an extremely surprising statistic. Regardless of the amount of exposure, I think there would have to be some sort of internal constraint for children to learn to extract patterns just because the volume of information is so large if we really do get 30 million words by 3. At the same time, maybe the environment does have a much larger role than previously assumed by Poverty of Stimulus arguments. I think to understand UG better, we do need to do more cross-cultural studies with language because it seems like people confuse UG with all the rules in English grammar, whereas in 9A they did briefly talk about Spanish and the cascade of effects its grammar has.

    I am very curious about people who reverse-engineer language learning. I believe Professor Shultz presented some algorithms that were trying to emulate language learning in children and I found these models to be fascinating. Would it be possible to conduct a similar test to the Turing Test in this case? If a cognitive model we construct (with dynamic components to satisfy the symbol grounding that might be necessary to acquire categories of grammar) showed the same behaviours children did in learning language, could we not conclude that is the correct explanation. So far research on this focuses on computation alone, but as we saw in 9a and 9b, some parts of syntax seem to be dependent on meaning, so a T3 system would be necessary. The only reason I suggest this method is because it would allow us to conclusively measure the input to the system. However, I’m not sure what those internal processes might be in such a model.

  12. We do acknowledge that genre differences represent a real issue. Too little is known about the direction and effect of differences in construction frequency across genres. (page 22)

    One concept presented in this article that I found particularly fascinating was the stark differences between speech and text in terms of their syntactic characteristics. Pullum and Scholz spend most of page 22 discussing how some grammatical “utterances” are found almost exclusively in text, while others are found almost exclusively in everyday speech. However, they immediately go on to say that we don't really know just how significant these differences are, so let's assume they're not that important for the sake of the next 30 pages of their argument:

    Here we tentatively assume that differences between styles of speech or contexts of utterances are not so enormous as to make tape recordings of spontaneous conversation in front of children the only relevant kind of material. (pages 22-23)

    I wonder how the technological advances of the last 16 years have changed the dynamics of this situation. First, I imagine that the rates of parents reading to their children have been steadily declining, which means analyzing vast arrays of these texts is probably not the best way to determine what kind of grammatical utterances children are actually exposed to as they acquire a language. Instead, we should be focusing on the new forms of media marketed directly for children, such as television shows, movies, and interactive learning programs. Even if some parents make a point to read to their children every night, exposure to new media would still be much greater than classic text, making them much more relevant to this study than stacks of old children's books. Pullum and Scholz even consider this possibility when they discuss lines from the television program Mork and Mindy on page 43. Furthermore, analyzing television shows and movies would eliminate most (if not all) of the problems caused by the differences between everyday speech and written prose, as the writing is primarily normal conversation between characters rather than lengthy paragraphs.

    Second, I don't think analyzing every instance of spontaneous conversation around a child acquiring a language is still as insurmountable of a task as Pullum and Scholz make it out to be. Thanks to major advancements in data storage and processing power, this experiment does not seem outrageous at all. Attaching a digital recorder to the child's back would not be a perfect way of “hearing” everything the child does, but it would probably be good enough to make the results of the study statistically valid. All these audio files can be sent over to a data processing company, who can divide them up between its many employees. As such, generating transcripts of everything the child hears goes from a ridiculous and unmanageable task to just another Tuesday morning at the office. Medical transcription has been a major industry ever since doctors started taking verbal notes about their patient interactions, so it is likely the infrastructure is already there and just waiting to be used. These transcripts can then be run through a computer algorithm and shortened to a list of relevant phrases that the researcher can analyze. This experiment now requires very little time, especially if the head researcher can get one of their lab students to sift through the list of relevant phrases for them. While the task may be expensive, it is definitely not impossible, and as Pullum and Scholz point out, it would be the definitive study either confirming or invalidating the APS.

    In summary, I think analyzing television programs provides a better way of determining what forms of grammar children are actually exposed to than analyzing children's books does, and the task of analyzing every phrase a child hears as they acquire language is much more reasonable than it was 16 years ago thanks to the major technological advancements we've had since then.

    1. • “Surely, how a grammar is constructed is not the same question as why that particular grammar was constructed, yet Wexler appears to conflate the two questions (12).”

      This quote reminds me a lot of some of the issues you bring up - significant differences between utterances and why they might have decided to ignore that issue.
      Your solution to the problem of parents not reading to their kids anymore and children mainly gleaning linguistic information from new media sources like television and movies and little toy computers that talk back does not really change the "how" of the question of looking at linguistic acquisition, it just provides a different input to examine.
      I also disagree that television and film are analogous to learning languages the way that books are, especially for children. At least the children I have cared for interact with books in a completely different way to the ways they interact with television.
      In psych 213 (cognition, funnily enough) we studied attention a lot and the effects of tv on visual attention. Because everything is moving constantly most young children who are at the age of acquiring language in the critical periods outlined in the Pinker article are pretty much absorbed by the flashing images etc. This is a very different way of engaging with a language than in a book, where children will often read back to you words or comment on them. I do not think these two ways kids acquire language are as easily comparable as you equate, though there is definitely a lot of positive and negative UG rules that children can learn from TV...

  13. So for the previous article I hadn’t watched the entire video-lecture that accompanied it before writing my sky-writing response. It turns out that at the end of the video Stevan calls what I brought up in my sky-writing to be a “pygmy escape route.” The argument I brought up is that it seems really tricky to separate syntax from semantics. This is because the structure of a proposition can have an effect on the meaning of the proposition. Generally, I do not think the goal of language is to communicate grammar alone. In fact, I can’t even think of a way to imagine syntax completely isolated from semantics.

    Anyways, I’m bringing this up because this whole goddamn article seems to make the assumption that you can separate the two (which is funny, because they’re trying to point out other peoples’ assumptions). They do this most clearly when they outline the form of the Argument for the Poverty of the Stimulus. The first claim (a) reads: Human infants learn their first languages EITHER by data-driven learning OR innately-primed learning.

    I think it absolutely has to be both, operating at the same time. Sartre thinks people are so sad all time because the individual must constantly assume the crushing responsibility of meaning-maker. In other, words you can’t turn off meaning.

    Ok, but what if I open up a dictionary for a language I don’t know? All I’m going to see are squiggles and lines, the shape of the language. But, it still FEELS like something to not know something. If people can’t stop feeling, how can we study grammar/syntax isolated from meaning?

    1. Hi Nicholas,
      I agree that language acquisition has to be a combination of data-driven and innately-primed. Regardless of the actual words used in the text, perhaps Pullum’s EITHER/OR is not exclusive and instead inclusive, which means it could either one, or the other, or both.
      With regard to syntax vs semantics, I would be of the opinion that it’s possible to study the two separately. That is, syntax really does specifically target the grammatical structure of a language. There is no need to know meaning in order to study syntax, as syntax strives to find universal rules to all languages with parameters that are simply changed depending on the specific language. I’ve taken a syntax class and drawn syntactic trees for Malagasy, Chinese, Gaelic, and many other languages I do not speak a word of. Furthermore, it’s possible to create a grammatically/syntactically correct sentence which has absolutely no semantic meaning (the token sentence for this in linguistics is “Colorless green idea sleep furiously.”). The reverse in semantics is not true; every sentence that has a semantic meaning (i.e. means something) have a syntactic realization. You can’t have an “ungrammatical sentence that means something”. In that sense, syntax and semantics are not dissociable. But the other way around, they are. You can easily study the syntax of a language without understanding meaning.

  14. This article by Pullum & Scholz (2002) nicely outlines that a child will hear UG-compliant phrases and learn this way through positive evidence, rather than using negative evidence. It put forth very compelling evidence in favor of the existence of UG, which I myself have not spent a lot of time digging into, so that was nice. What is interesting to note is that while we may not be sure that everyone has consciousness or even the same degree of consciousness, we are sure that us humans share similar enough brain structure that gives rise to Universal Grammar (UG). This biological similarity is a fundamental aspect of our human nature which makes it all the more interesting to study!

    APS, as outlined by Pullum and Scholz, is characterized as there being two possible options for learning a first language - 1) data-driven learning or 2) innately-primed learning - and of the two, children learn the language by using the latter. I really enjoyed this argument being represented in deductive form as I could relate it to FOL learned from my deductive logic class. Although we suppose here that there are only two possible alternatives, my mind kept thinking of finding a counterexample or anything that could undermine the premises. Firstly, why can't the first premise actually be a conjunctive one meaning that human infants are learning their languages as a result of a combination of data-driven and innately-primed learning? Why must it be one or the other? I am convinced (as many are) that we have innate language-specific structures that account for innate grammatical rules but that we also need to learn from our environment.

    1) So as a child who is born with innate UG rules continues to interact with its environment (hearing parents speak, strangers speak, TV, radio, toys, etc), could the richness of the environment (Sampson 1989) play a role in how quickly they are able to become proficient and tweak the UG parameters for successful language acquisition?

    2) Could our innate UG continue to evolve so that we are faster at picking up languages in the early years of our human lives?

    3) How much influence does our current environment have on our language acquisition relative to in the past when there were no TVs, radios, electronic toys, visitors? Has technology had an impact?

    1. To answer your question 3.
      3. Socialization of a child is a critical period during their childhood. The more things they are exposed to, the most they will acquire because there brain is plastic enough to acquire a lot of information. For example, someone may grow up in a unilangual house and so they will grow up only knowing that one question. Someone else may grow up surrounded by two parents and thus grow up to know two languages. So basically the more input a child has, like languages, the more they will acquire (socialization). TVs, radios, electronic toys etc. are just new ways are learning. Children watch shows on tv that teach them a lot of new things (children shows), so I think it's safe to say that when there were no tvs, children lacked that type of input but made up for it from there parents. Technology definitely impacts language acquisition. There are so many examples I can think of. For example, my cousin who lives in Israel is very good at Spanish and I asked her "why?", since that language is not so typical in Israel. Her answer was "because I used to watch Barney in Spanish when I was young, so I picked up on the language". So that's the answer right there. Without the TV input, she wouldn't have acquired Spanish language.


  15. I think the critical period hypothesis ties in nicely with the poverty of the stimulus or the absence of negative evidence. If we think of language acquisition like we do other critical periods, then usually there is not negative evidence needed. Vision for example – there are studies where cats who have their eyes sewn shut during the critical period do not develop proper visual circuitry in the brain. They need the visual input during this period to develop these systems. Maybe UG works the same in that during a critical period it simply needs positive input in order to develop the circuitry. Those who don’t learn language (ie don’t get linguistic input) during the critical period have been shown to never fully grasp even their mother tongue and especially have difficulties with grammar.

    It would be interesting to have children only hear non-UG sentences (that way there is positive input but it’s incorrect) during the critical period and see if they still develop normal language with UG. To do so might mean more conclusively that their UG circuitry is innate and it only needs positive input to develop. Obviously this would not be ethical though.

    I’ve been trying to think of why UG might have evolved. Maybe it was to “turn on” or guide language in the critical period and make it so that our language are all fairly similar. This would aid in communication which I’m assuming is the main adaptive benefit to having language in the first place.

  16. It sees like Pullum’s argument is that UG might be less innate and more environmental than Chomsky states. He says this is due to the fact that the linguist’s “poverty of stimulus” argument might not be as valid as they might think it is; that is, that there might be more stimulus available than accounted for in the linguists’ studies. He isn’t saying that poverty of stimulus is wrong, but simply that it maybe shouldn’t be given as much credibility as it currently is. He suggests that perhaps positive evidence is enough to acquire a language; i.e. data based acquisition is possible. However, even though he looks into arguments given in favor of poverty of stimulus and explains how they might not hold up under further scrutiny, there are other important arguments that he does not take into account. The main one is “absence of evidence is not evidence of absence”. That is, there are features of languages that children acquire when they have no reason to have done so. For example, topic dropping in a sentence. In Spanish, the subject in a sentence is optional and often dropped. Therefore in positive evidence for Spanish, both types of sentences will be present and the child will know that subjects are optional. In English, a sentence must always have a subject. Therefore all the positive evidence in English will have sentences headed by subjects. However, there is no way for a child to know for sure that sentences without subjects are impossible; the child simply might not have encountered such a sentence yet. Therefore there has to be some innate parameter which says that subjects are obligatory in sentences, until proven otherwise (this would be exposure to Spanish), in which case the parameter would have to be readjusted.

  17. First off, I want to say that I really respect what the authors are trying to do here. Whether or not they’re right about the strength of the poverty of the stimulus argument with its current support, I’m not sure (I would now say no after realizing their mix-up between UG and non-UG grammar). But the fact that they’re not content accepting a widely accepted assumption at face value is commendable.

    “People attain knowledge of the structure of their language for which no evidence is available in the data to which they are exposed as children
    What Hornstein and Lightfoot claim is that some of the sentences children never hear are crucial evidence for learning from ex- perience. That is, some aspects of languages are known to speakers despite the fact that the relevant positive evidence, although it does exist, is not accessible to learners during the acquisition process, because of its rarity: linguists can in principle discover it, but children will not.
    This claim, if true, would refute non-nativist theories of language learning. If learners come to have linguistic knowledge without being exposed to evidence
    that would be crucially necessary for experience-based learning, then learning is not experience-based.”

    I follow for the first point of the last paragraph: it is true that the claim made by Hornstein and Lightfoot would “refute non-nativist theories of language learning.”

    But I feel that Pullum and Scholz aren’t specific enough, and the entailment of the claim would be a refutation of 100% non-nativist theories of language learning. Is it not possible that there is a hybrid system at play in language acquisition? Could it not be that there are some grammatical facts that are innate, and others that are learned through domain-general or other non-nativist mechanisms?
    After reading some of the comments, perhaps this is simply Fullum comparing apples and oranges when it comes to UG versus non-UG grammar, but I didn’t pick this out when I was reading. Could anyone give some specific examples of where the two are confused in the article?


    “If people so rarely produce utterances that exhibit their grasp of the structure-dependent character of the auxiliary fronting generalization, then there could well be speakers around who have acquired an ìincorrectî structure-independent generalization instead, but who are never detected because of the rarity of the crucial situations in which they would give themselves away.”
    This seems like a pretty weak argument. Doesn’t the fact that the few utterances that are found are always correct strengthen the argument that it is always learned, no matter how little input a speaker is exposed to? Fullum’s argument here almost seems analogous to “there could well be a rare breed of apples that fall up which are never observed to fall up because of the rarity of observing this type of apple fall off of a branch.”

  18. I think the authors, Pullum and Scholtz are sort of missing the point of the APS. The authors suggest that the child either learns from the environment or has innate language abilities (one or the other). However, Chomsky’s argument is that the child needs to learn both from the environment and also have innate language abilities (not one or the other, but both). The child needs his innate abilities to hear language (or see signed language) that is being presented to them in their environment. Furthermore, within the brain, there exists specific regions that are devoted to language learning that are separate from other types learning of learning, to me, this is a very persuasive argument for UG. I don’t disagree with the authors when they say that a thorough investigation should be implemented, however I don’t know how such a thing could actually be done.

    1. Hey Lucy, I agree with you. I think that the main problem I have with people who are against the Chomskyan/UG theory is that they are taking it to mean that there is a complete absence of stimulus that occurs. What is more likely the argument for UG is that we are all equipped with the capacity for language, but that our environment unlocks it in the sense that we learn the language we are surrounded by. But the capacity for learning that language, for acquiring it, is somehow encoded in us because of our being born with UG. This way, if we were raised in Japan we would learn Japanese, but a child raised in Canada will learn English--not, obviously, Japanese. I think that though this might seem like a trivial thing to point out, I encountered the sort of argument when reading this paper at the point that the authors wrote "Our experience does not establish for us definitively that dressers with faucets are impossible, or that bedrooms do not contain cookers; yet we come to believe these things too ñ defeasibly (perhaps some people might have a cooker in their bedroom), and in a way that is depen- dent on our environment (we would not have learned the same generalizations growing up in a rural African village), but without explicit negative evidence. It would be rather radical to claim that learning facts of this sort from experience is impossible, so that there must be innate domain-specific knowledge about architecture, furnishings, and appliances. " I think that in response to your objection, or the problem you find that the authors don't want to accept the possibility for a language acquisition process that is both data driven and innate, I turn to a sentence at the end of the paper. It reads, "We are also will- ing to accept it as a possibility that once linguists and philosophers get straight on what is required to establish an instance of the APS, the distinction between data-driven and innately-primed learning will be less absolute." Maybe this is the solution. Maybe, instead of seeing language acquisition as one or the other, there will be an amalgamation of the two. That for now stands to be too vague to be satisfying but I think they're on to something with that. I can appreciate the problems that these authors have with APS but I also have a very hard time accepting the idea that there isn't some sort of UG, that we aren't born with an innate ability to be able to acquire the language we're surrounded by.

  19. I missed this class due to illness, so I apologize in advance if any of this has been answered already! These questions are a bit of an off-shoot based on these readings, but I just have some burning questions about the evolution of UG, which there likely aren't yet answers for.

    1) Judging from the comments above, we all acknowledge that Pullum and Co. centre their paper around everyday language/grammar, and fail to discuss the real issue at hand: evolution of UG. My question is: does Chomsky have a theory of how UG could have evolved? Don't get me wrong - Chomsky is a true giant and I'm not trying to suggest that this hole in the theory is someone he isn't aware of - but I find it incredibly frustrating to be presented with a theory (innateness of UG) that a) is not falsifiable and b) seems to suggest an innate (and thus, by necessity, evolved) mechanism that is inherently un-evolvable based on some of the same qualities that make it unlearnable... Without a doubt, the innateness of UG seems to me to be an irrefutable notion - the burden of proof clearly falls on those who aim to provide examples of negative evidence in the learning process, which has thus far proven impossible (another critical point glossed over by Pullum & friends). But what are we meant to do in the face of such glaring issues? How can one be both a Darwinian and a Chomskyan (as I would truly like to be)?

    2) Following from (1) - we've thrown the idea around many times that a putative adaptive advantage of language would be the ability to learn new categories by verbal instruction, instead of the far more risky induction. But if, as we/Chomsky are suggesting, UG is absolutely necessary for language, does that pose a problem for our theory? Not only because 'adaptive advantage' becomes a moot point in a potentially un-evolvable ability, but also becomes UG seems incredibly over-complex for something like acquiring categories by instruction.

    1. Hi Adrienne,

      What do you mean when you say that “the innateness of UG is not falsifiable?” If someone were to show both that there is sufficient environmental data to learn these rules when language is being acquired, and that failure to do so would lead to UG errors later on, it would show that UG is not innate. As far as we know, it will never be falsified, since our best evidence says that UG is innate. But “unfalsifiable” does not just mean “true.” It means that there is no conceivable sort of evidence which could ever show it not to be true. This is not the case for UG.

    2. I completely agree with your concerns about the evolution of Universal Grammar. The first problem seems to be that no-one can agree on what it actually is. Most of the linguistic papers I have read involve linguists conflating Universal Grammar with everyday grammar (like Pinker and Pullum) or advocating different and contradictory analyses (see blog post on 9a). If we cannot determine how exactly Universal Grammar is represented in the brain, then it seems impossible to begin working out how it got there.

      But even if we do manage to determine the rules contained in Universal Grammar, presenting a theory of how they became an innate part of the human brain still seems impossible. It becomes a like a ‘which came first, the chicken or the egg’ situation. Languages can’t have evolved before Universal Grammar; otherwise how would every language develop adhering to all the Universal Grammar rules? But it seems ridiculous to assume that Universal Grammar came before language; how can we have rules with no ‘content’? How can Universal Grammar have evolved over time if the whole reason for arguing for its existence is that children lack sufficient input to construct grammatical rules without it? How can there be any ‘in-between’ stages?

    3. I definitely agree with you, Rose and Adrienne. UG is such a complex structure, and while it makes sense to say that it must have evolved somehow, it’s hard to place delimiters on what is and is not truly innate. Related to what professor Harnad mentioned in the video, about finding negative evidence about a specific instance or arbitrarily allowing a non-UG rule to be legitimate, it doesn’t really change the structure of the UG – however, this still really doesn’t get us any closer to understanding why such specific and complex rules would be evolutionarily advantageous to possess innately, if anything, it makes it seem more puzzling.

  20. “Certainly, humans are endowed with some sort of predisposition toward language learning. The substantive issue is whether a full description of that predisposition incorporates anything that entails specific contingent facts about natural languages.
    We also caution the reader against reading this article as a defense of the claim that purely empiricist language learning, via domain-unspecialized algo- rithms for knowledge acquisition, can suffice for learning natural languages, given childrenís experience.”
    In the introduction of the paper the authors appear to not be refuting the APS but hinting at the necessity for linguists to pose the argument succinctly and attain “proper” empirical evidence (since they decide that there is no real solid evidence for it). However, as the article progresses they seems to attack the APS in a more aggressive manner claiming that, as it is now, it “still awaits even a single good supporting example.” Reading some of the comments I found very helpful in teasing out what exactly is going on, as, in reading the paper I got rather confused since it seems as though there are claiming elements of grammar to be overly nativistic (if that makes sense). To echo what has been said both in class and in the discussions, the authors miss the key point about what UG actually is (lack of negative evidence for non-UG compliant sentence structures).

  21. It’s always refreshing to have a reminder that the human sciences, are, in fact, conducted by humans. They are thus prone to fad and favouritism. This article convinced me, at least, that linguists have not been paying their dues in terms of establishing sound premises and ensuring the falsifiability of their hypotheses. The main complaints outlined in this article were about the false assumptions that the given construction was in fact unlikely to be encountered, and the lack of representative data given an unusual occurrence. This points to a situation where it becomes quite difficult for us to either confirm or deny the poverty of the stimulus argument. To falsify it we would need to examine all language data for a representative sample of the population and to confirm it we would need to conclusively determine that an acquisition was made without prior exposure (or to something similar), which will require a large subset of the population and a confident examination of all possible similar phrases.

    Thus, it seems to me to be most profitable to build computer models of how a child can learn, operating on a variety of assumptions and test these to real-world data, rather than attempting to parse and collect data from real world participants. We cannot yet articulate all the rules in speech; our grammar does not match what we speak. Therefore, we can test only through direct comparison rather than our own rule creation (something not mentioned in the paper).

    To connect this to the rest of the course, this paper identifies itself as being on the fence in terms of their being a non-empirical module that carries universal rules for acquiring language. Overall, it doesn’t provide an argument against a language-learning faculty, which is perhaps a weaker version of the universal grammar theory.

  22. “People attain knowledge of the structure of their language for which no evidence is available in the data to which they are exposed as children.” Hornstein and Lightfoot’s claim, discussed in Pullum’s paper and pointed out, is a very strong claim and really doesn’t make much sense. Limited evidence maybe, and finding out how limited that evidence can be would be very interesting and useful, but a child receiving no evidence of the structure of their language still having that knowledge doesn’t make sense. A baby born anywhere around the word could learn any language as their first language, for example a Japanese baby could learn English as their first language. The Japanese language and the English language have very different structures, so if a child was given no evidence how does it attain the structure of ‘their’ language when they don’t know upon birth which language is theirs?

    This paper also does make me think about Wharf’s theory again and how language may influence our society in the way we think. For example, how we gender the world and how that categorizes humans in our language that then effects their identity and how they behave and how they feel they are allowed to behave, affecting how they then speak and communicate. In some ways I think Wharf is right, I don’t think we would think in certain categories if we hadn’t have labeled them with our words. However, thought (in terms of language) and language is a cycle of influence and I don’t think you can pin point what started the cycle.

  23. I found this article exciting because ever since I first heard about UG and Chomsky 3 yrs ago, I have believed it immediately, so it was nice to read another view.

    “how a grammar is constructed is not the same question as why that particular grammar was constructed”
    I like this because it kind of answers my question from 9a about the evolution of language and differences between them. I was clearly conflating the how and the why.

    The authors describe Lightfoot’s weakened claim for linguistic nativisim as: “there could be positive evidence for the correctness of some rule or principle, evidence that could be found in language use by an adult linguist doing a really intensive search, but would not emerge in conversational data anywhere near often enough to guarantee that any particular child would ever encounter it”
    Although some people must think this, I guess this is not my reasoning. My reasoning is that children, especially young children, are not cognizing consciously at all. They are simply not linguists. Heck, I am a fully grown adult who speaks fluent English, and I could not tell you the (very complex) grammatical rules of English, let alone of UG. I don’t think children are thinking in terms of “rules” or “principles”. This is not specific to language, though language is probably the most complex case of rules. For instance, a small child learns the “rule” that if she goes near the stairs, her mum yells at her. But I don’t think she thinks of it as a “rule” per se – though enough data is present to her to know that her mum will make loud noises. Rather, she somehow doesn’t like yelling, or frowning, or whatever, and can “predict” without knowing she is predicting that going near stairs will cause that, though she does not understand cause. If you can “know” a rule without understanding prediction, causality, or indeed the concept of a rule itself, then sure, the child knows the rule. But I don’t think that is knowing at all.

    So I guess for me the question is: how does the child learn the rules of language, amongst other rules that she learns, without being capable of that sort of “Intensive research”? Especially since language lacks negative evidence. This is separate and broader than the question of UG, which applies more specifically to language and the rules that govern all languages. Because children can learn rules (and this capacity forms the basis of all learning), there is clearly innate something-or-the-other. And because language has rules, this innate something-or-the-other includes UG.

    I don’t think this is Chomsky’s argument of poverty of the stimulus, which I agree is what Lightfoot said.

    Now that I think about it more, maybe my question shows why poverty of the stimulus is not a good argument. The fact is, children are not hypothesizing. They are somehow rule-learning without hypothesizing, which is what we do at an older maturity. What they are doing, I don’t know. But that is why poverty of the stimulus doesn’t matter – because that applies to hypothesis-predictive-sort-of-explanations of language acquisition.

    Of course, the question then leads to: if other animals can learn rules, why can’t they learn language if UG is a type of rule-learning? And in that case, I guess UG is an important aspect of general rule-learning. Or perhaps this shows it is distinctive from even that.

    There are too many questions!! And the more I think about it, the more certain I am that I have missed the point of this article (and UG) altogether, which is disheartening.

    1. Hmmm I am thinking about this more, and I think where I differ with Lightfoot et al is that they think there are some sentences or even some total amount of sentences x which would make data-driven learning possible. I disagree because the point isn't that the children hear x sentences or not. The point is that they can infer *at all* from even x sentences.

  24. This article was quite lengthy so it took me a while to complete it. While I don’t agree with all that Pullam and Scholz have expressed in this review, I do admire the fact that they are trying to stick to the scientific method in order to figure out the validity of the poverty of the stimulus argument. This is great, and I like their criteria for empirical proof of UG. However, I feel like language and language development are so difficult to study that their recommendations are obviously way easier said than done. If this whole argument rests on whether or not we get enough positive feedback as children to develop language without the existence of UG, how can we quantify this? Moreover, children who do not learn language (e.g feral children) in the critical development period often cannot learn language later even with intensive instruction. These UG blueprints (if they do exist) are based on a use it or lose it principle, and therefore wouldn’t all of these studies have to be in children?

    I myself am on the fence with UG, but I can’t decide how much proof I would need to accept it completely. I feel somewhat hypocritical if I am so quick to dismiss evolutionary psychology as “just-so” stories but am way more ready to accept linguistic nativism and UG.

  25. It is nice that Pullum and Scholz acknowledged the sizable gap in their argument left by not addressing the negative evidence claims, but at the end of the day it is a large portion of the Poverty of the Stimulus Argument . However, I do like their general idea that this type of claim blew up in popularity so fast that it never quite got the criticism it needs to properly flesh itself out. Linguistics stands to benefit greatly from a sober reflection on (i) what it is that the Poverty of the Stimulus is actually claiming and (ii) what empirical evidence we have for substantiating that claim.
    If the schema for building a Poverty of the Stimulus Argument presented in this paper is the only logically sound construction of it, then I think that Pullum and Scholz have done a good job at demonstrating that at least four such arguments leave more to be desired. I am not sure if these four examples are representative of the literature on the whole, but I think that linguists seeking to claim a lack of positive evidence for some grammatical phenomenon should consider this five-part strategy.
    I also agree that there should be more empirical and data-driven work in the field. Now more than ever, we have access to nearly infinite recording data that we can analyze computationally (see, so there is no more reason to appeal to certain grammatical structures seeming “vanishingly rare.”

  26. Maybe I'm biased by the way that the 'argument for the innateness of UG from the poverty of the stimulus' (ASP) was taught to us, but I don't really see much sense in the way this article by Pullum argues against it.

    He goes on about how for ASP to be true, it means that there are not ENOUGH examples of non-UG-compliant sentences that a child hears during language acquisition for them to learn UG.

    I suppose then that it might be better, for the sake of argument to recharacterize this argument as the ABSENCE of the stimulus rather than the POVERTY of the stimulus.

    Children don't encounter ANY examples of non-UG compliant sentences, nor do they naturally produce any. In this way there is never any corrective feedback about UG, nor any incorrect examples of UG. If we think of UG as a category to be learned, children would never be able to learn it since they never encounter any non-members, nor do they ever incorrectly categorize sentences. Since they adhere to UG anyways, this would suggest that UG cannot be learned and must be innate.

    From these logical premises, it then seems nonsensical to read this article which demands more evidence that UG could not be acquired by data-driven means. "The advocates of ASP must shoulder the burden of participating in empirical work to support instance of the argument." Isn't the above argument conclusive? Unless I'm missing something I'm unsure why there should be any dispute over the ASP. What is Pullum getting at here?

  27. Pullum & Scholz are by no means arguing against the notion that there is some aspect of language that must be innate. They agree that humans must be endowed with some sort of predisposition to language - this must be so. Rather, as they put it: "the substantive issue is whether a full description of that predisposition incorporates anything that entails specific contingent facts about natural languages." Since there are many variations and arguments that contribute to the Poverty of the Stimulus theory (linguistic navistism), the authors choose to focus on just one, which they claim to be the strongest and most productive argument for lingusitic navistism: "People attain knowledge of the structure of their language for which no evidence is available in the data to which they are exposed as children."

    After reading all their assessments it is clear that unlike I initially assumed, Pullum & Scholz aren't at all opposed to the idea of linguistic nativism, they simply demand a quantified explanation for it, which in my opinion has been long overdue. In my language acquisition class, we were told about the concept of Poverty of the Stimulus but given no real evidence, just the assertion that children aren't exposed to A but they learn to produce A, so there must be some kind of innate machinery that allows them to do so. But this is a 'just-so' story, and not any kind of 'how/why' explanation.

  28. One thing before I get into my main idea:

    I think that for this article to resonate with me more strongly, Pullum and Sholz needed to look more at negative evidence. I wonder how, with no negative evidence, kids can still learn proper grammar? It’s interesting, but I think that for an effective article, they need to go more deeply into this.

    Now, what I found particularly interesting was when they got into the differences between speech and text. I had never really considered things such as:

    "For example, Biber (1988) finds that about 31 past participial reduced relatives are found per 10,000 words of journalistsí prose, but only one occurs
    in 10,000 words of typical face-to-face conversations."


    "On the other hand, we have noticed that some features of English have
    the opposite property, being much more frequent in spoken language than in
    printed prose. For example, it would appear that consecutive synthetic and analytic negations (e.g. We canít not go) are edited out of printed sources such
    as journalistic prose"

    I never considered this... and it makes me wonder about how the books that children read and the signs to which children are subliminally exposed during their developmental years affect their UG mechanism or IF it affects their UG mechanism.

    "And indeed, Weddell and Copeland (1997) found that even children between ages 2 and 5 pick up and understand much more (e.g., about news events) from the language they hear on television than their parents typically realize."

    This contradicts some of the information put forth in article 9a which states that "The child must have some mental mechanisms that rule out vast numbers of "reasonable" strings of words without any outside intervention". Picking things up from television is the same as picking things up from printed prose, as I'm sure the scripts are edited to the same degree as print would be. Due to the fact that some children learn primarily a different language at preschool/daycare than they speak at home (e.g. children who come from a Japanese family but who live in Canada and learn and are spoken to in English at daycare/preschool), they are exposed to formal texts in different languages. For example, a child in the aforementioned situation would be exposed to formal English prose but never really formal Japanese prose as their parents at home would be speaking colloquial Japanese and people don't have "enough conscious control over the syntactic properties of their utterances to enable them to switch their syntactic complexity level in any really significant way as they move from boardroom to street to kitchen to classroom to church to school." So this makes me wonder which have a greater effect. If their syntax in Japanese would be any worse/better than their syntax in English. The discrepancies between articles 9a and 9b should also be considered.

  29. I just wanted to point out the frequent use of innate in this article. Although it brings importance to the article, the authors fail to explicitly explain what ‘innate’ refers to. For example, do they mean ‘not from learning or does it mean from birth? According to the dictionary, innate is "inborn; natural.”. However, in this paper the author seems to use Innate in the context of “not acquired through sensorimotor learning” . In order for the full understanding of the word innate, the authors should have elaborated on this meaning.

  30. Pullum and Scholz’s article takes a good look at linguistic nativism and shows that there is work to be done to support the empirical premises of types of arguments for the subject. In their article, Pullum and Scholz claim that data of the type in question are not found in normal linguistic experience; conclude that language use exposure isn’t merely how people learn language. Also, they offer some reasons for thinking that the relevant kind of future work on this issue is likely to further undermine the linguistic nativist position.

    They challenge the view that a fact about some natural language is exhibited that allegedly could not be learned from experience. If it has to be learned from experience, it must be done with some (positive) data that is given to speaker/learner.

    Unfortunately, what Pullum and Scholz fail to address is the concept of negative evidence. As mentioned above in other comments, I found this a bit surprising. Furthermore, when we talk about UG, it interests me is how successful language learning is achieved without negative evidence.


    Additionally, I agree with Lucy’s comment above:

    “The authors suggest that the child either learns from the environment or has innate language abilities (one or the other). However, Chomsky’s argument is that the child needs to learn both from the environment and also have innate language abilities (not one or the other, but both).”

    I think that Chomsky’s argument makes more sense because our environment does play such an important role in learning. And having an innate ability would only go hand in hand with learning and gathering information/knowledge form the environment. Adding to that, wouldn’t an innate structure along with references in the environment help in learning a language?

  31. I found the technical nature of this article quite challenging, but this is perhaps due to the fact that I am quite unfamiliar with linguistics or the psychology of language. From what I understood, the arguments for innateness versus data-driven learning don't have to do with the language itself, or the lexicon, but rather the rules guiding syntax, or grammar.

    To prove the argument of the poverty of the stimulus to the extent that would a non-nativist, learned-language argument, some grammatical rule must be found that is registered and properly used by a native speaker, but it would not occur enough in conversational life for the child to pick it up themselves. The evidence from Kimball for APS was Aux T(M) (have+en)(be+ing), or “may have been”, where sentences where the “auxiliary is fully represented by a modal, perfect and progressive are vanishingly rare”. The evidence for its rarity was that an observer hadn’t noticed it being said very often in the past eight years. I think we can all agree by modern standards that hardly constitutes as evidence.
    Pullen and Sholz found that this phrase is found once every 3,000 or 4,000 words. Yet if children are hearing about 30 million words by their 3rd birthday, they would have been exposed to that phrase at least 7500 times. Again, I agree with the authors that if that amount of positive exposure to the phrase is not sufficient, the APS advocators must explain why, and what the cut off should be.

    To their point, I have often seen reference to Chomsky’s poverty of the stimulus argument where it is stated as something that has been proven. Yet, as the article notes, according to modern day standards of empirical evidence, the claim is somewhat unfounded.

    As a side note, I appreciated the example of the house, where you would be familiar with what is in a house through pattern recognition and repetition. The feeling that a faucet shouldn’t be attached to a dresser similarly to the sense of feeling that a “by and large” is correct but “of and large”, or “by and small” are not, is a very illustrative example of something we take for granted as knowing, but fits well with the argument Pullum and Sholz are making.

  32. This article was a bit frustrating because Pullum did not offer any solid grounds nor get to the real problem. His stance was that there is less effect from the innate UG and more from environmental learning, since the stimuli are not as impoverished as linguists assume. Rather than trying to disprove UG, he tries to make it less credible, or take away what can be explained without UG. As Pullum states that the burden is on the APS theorists to provide the necessary evidence on the poverty of stimulus, rather than on the UG skeptics to provide evidence of UG positive learning from early experiences of language, he suggests that until the linguists come up with further explanations, perhaps all UG components can be learned from early language experience. Further, he neglects the real problem of poverty of stimulus by overlooking the absence of negative evidence.

  33. I am very curious as to why we cannot have both learning that is data driven and innate? Are both necessary for communication?

  34. I appreciate the purpose of this article in clarifying the strength (or lack thereof) of the argument for the poverty of the stimulus. The evidence required to support a theory of innately-primed syntax must demonstrate some rule or principal that was absorbed by the speaker, and could not have been learned through explicit instruction or environmental exposure. A review of the literature was obviously quite important based on what Pullum and Scholz found. Forty years ago, Kimball used the evidence of having heard a particular utterance in conversation after “eight years of attention to the problem” less than a dozen times, thus concluding it was “vanishingly rare”. This may have been sufficient evidence to draw conclusions from in 1974, but with the ability of data searching that we are capable of today, Pullum and Scholz show that this is not the case. They found hundreds of examples of this utterance, including in children’s books. This example illustrates the need for review of the research for nativist’s argument.

    When considering the generalization of universal grammar to the capacity of cognition overall, rather than maintaining its specificity to language, it brought back the question of the Whorf Hypothesis with the extent to which our conscious experience is influenced by language. Our thoughts, memories and sense of self is largely attached to an inner narrative that develops along with our language capacity. Perhaps the evolutionary purpose of the framing of universal grammar existing in all humans is to ensure a similar state of a sense of self and inner narrative, such that we can relate to each other more easily in a social group.