9b. Pullum, G.K. & Scholz BC (2002) Empirical assessment of stimulus poverty arguments
Reading: Pullum, G.K. & Scholz BC (2002) Empirical assessment of stimulus poverty arguments. Linguistic Review 19: 9-50
Instructions for commenting: Quote the passage on which you are commenting (use italics, indent). Comments can also be on the comments of others. Make sure you first edit your comment in another text processor, because if you do it directly in the blogger window you may lose it and have to write it all over again.
Pullum and Scholz’s discussion of compound plurals raises interesting questions about how much linguistic knowledge comes from experience versus innate structure. They revisit Gordon’s claim that children avoid regular plurals in compounds (saying rat-eater but not rats-eater) despite never hearing examples, which has been cited as evidence for innate grammar. However, Pullum and Scholz show that English actually includes many compounds with regular plurals, suggesting that the needed input might not be missing after all. Their findings invite reconsideration of how much of language learning truly depends on inborn constraints versus the richness of linguistic experience. It’s interesting think about this view after the last reading where Pinker argues that language is an evolved mental organ and that children’s rapid mastery of grammar, despite limited input, supports the idea of an innate language faculty.
ReplyDelete***EVERYBODY PLEASE NOTE: I REDUCED THE MINIMUM NUMBER OF SKYWRITINGS. BUT THE READINGS ARE **ALL** RELEVANT TO AN OVERALL UNDERSTANDING OF THE COURSE. SO, EVEN IF YOU DO NOT DO A SKYWRITING ON ALL OF THEM, AT LEAST FEED EACH READING YOU DO NOT READ TO CHATGPT AND ASK IT FOR A SUMMARY, SO YOU KNOW WHAT THE READING SAID — OTHERWISE YOU WILL NOT HAVE A COMPLETE GRASP OF THE COURSE TO INTEGRATE AND INTERCONNECT FOR THE FINAL EXAM.***
DeleteSannah, Pullum & Scholz, too, conflates OG and UG misunderstands POS. Can you explain how?
Isn’t the argument from poverty of the stimulus assuming we can’t learn from purely positive examples? Since children have way better language learning capacities and UG is shared across languages, couldn’t they infer these linguistic structures through exposure and mirror relations? While negative feedback can be more impactful (e.g. we feel stronger about losses than gains), positive conditioning also matters. Rather than innate linguistic knowledge, perhaps children are evolutionarily predisposed to recognize universal patterns and learn from weaker, positive stimuli. This could reflect a Baldwinian process, where evolution favors the ability to learn such shared structures efficiently (remember evolution is lazy, why would it "encode" UG).
ReplyDeleteYes, the poverty of the stimulus argument does assume that positive evidence alone is insufficient. Even if children hear thousands of grammatically correct sentences, the argument claims that this input doesn’t provide enough information to infer the underlying rules of their language. That’s why nativists argue that some grammatical knowledge must be innate. Pullum and Scholz challenge this assumption, pointing out that there’s little empirical evidence showing children don’t receive enough positive input. They argue the real issue isn’t that humans can’t learn from positive data, but that no one has convincingly demonstrated such data are actually lacking.
DeleteI really agree with your points, and reading Pullum and Scholz actually made me think about it even more. Children seem to be incredibly good at picking up patterns just from the input they hear, without needing built-in UG full of specific rules. And if you think about it, we actually have proof of this in real life.
DeleteEx: if a child grew up in a community where people consistently used “incorrect” English, like saying “cat the” or “house the” instead of “the cat” or “the house”, that child would end up speaking that way too. They wouldn’t default back to English word order or some UG template, they’d simply absorb the structure they’re exposed to. This suggests the “innate edge” might not be a specific grammar blueprint, just a very strong ability to detect and generalize patterns from positive input.
This also lines up with how LLM’s work, they don’t have innate grammar, but they still learn complex structure through repeated exposure.
Finally, going to your Baldwinian point, evolution doesn’t need to pre-code detailed grammatical rules if it can instead select for brains that are exceptionally good at spotting and compressing recurring sequences of language. The “innateness” may lie in our cognitive architecture, our attention biases, memory constraints, pattern sensitivity and not in specific linguistic rules.
What I found interesting in the reading, is that the authors point out that many of the poverty-of-stimulus arguments are not actually based on observations. They are often based on some other assumption about what children must not have heard in the input. They argue that the initial step of claiming that something is “unlearnable from experience” needs actual evidence about what input the children actually gets. I appreciated their example showing how researchers sometimes rely on intuition or “folk linguistics” about what parents say, rather than examining corpora. Pullum & Scholz essentially say: if you argue that children can’t learn X in the environment, you must be able to show that X never appeared in the environment.
ReplyDeleteAnother point that I thought was interesting: even though some structures may be rare in the input does not mean they are unlearnable. Children are exposed to massive amounts of language over time, and they may still be able to pick up details in the patterns.
I think that you bring up an interesting point, I actually think this ties back to what we’ve discussed previously in class about how we can go about proving something. In the case of argument for the nativists they need to prove that the totality of all language that children encounter is not enough to teach them language. And of course, as the paper states there is just not enough exploration into what encompasses that totality and that assuming data driven approaches have not shown to arrive at a contradiction—which is required for the proof. So I think that until we can understand the true corpus of text and grammatical structures children are exposed to its hard to definitively answer both for or against the nativist point of view.
DeleteFrom my understanding, the authors of this paper aren't saying that there is no possibility that some innate ability to learn language exists; rather, they are merely pointing out that the underlying assumption for the idea of Universal Grammar – that kids are able to learn and produce grammatically correct sentences without ever hearing those sentence structures before -- is inaccurate. Kids do have enough linguistics data/input to learn patterns from experience. Thus, the “poverty of the stimulus” argument is not convincing enough to prove that grammar is innate (the Chomskyan belief). But I was thinking, what if everybody is looking at this issue too narrowly. Could grammar just come from articulation – that the way our mouths and tongues move just happens to flow in the form of grammar? It’s just what's easiest for our anatomy? In other words, perhaps grammar is not fixed in the brain, but it comes from how we move our mouths or from the prosody and rhythms we hear. Could grammar just arise from practicing speech, refined by social feedback later; it's not necessarily neurologically pre-installed?(Also, this got me thinking about a form of symbolism-association mechanims, as in the ability to link a gesture or a sound, like a word, to meaning, which I believe is an innate ability. For example, when I was around 3 months old – too young to speak – I would hold my hands up in the shape of a triangle to signal I was hungry. I was able to connect a gesture with a desire, making the association between the hand symbol and receiving food. So perhaps grammar is learned, but the drive to symbolize + associate (and thus use words to communicate) is inborn?).
ReplyDeleteHey Elle, interesting take! I especially like the part about grammar maybe emerging from articulation and social feedback rather than being pre-coded in the brain. That fits so well with the whole point that we shouldn’t assume “innate grammar” just because we don’t fully understand how learning works yet.
DeleteI think your idea lines up with the “bottom-up” way cognition might build complexity, like in the symbol grounding framework. Maybe our basic innate capacity isn’t grammar itself, but the ability to categorize and associate symbols with things (like your baby triangle gesture example!). Then, as we grow and practice, those grounded categories and patterns naturally evolve into more structured “grammar-like” behavior, because social and sensory constraints shape what’s easiest and most efficient to produce and understand. So yeah, maybe the brain doesn’t come with grammar “installed,” but it comes with the hardware and motivation to discover it through use.
Elle and Shireen—I find both your takes on grammar, stemming from articulation and social feedback, fascinating! I think grammar is another example of Baldwinian evolution, as it is a learned behaviour that has become easier over time to acquire. Individuals who learn it faster or more efficiently are favoured, and as you mentioned, our anatomy probably evolved towards easier language and grammar articulation. Your takes remind me of another aspect of language, phonetics and accents, which are tuned based on the culture or region in which you grow up. For instance, some infants gradually lose the ability to distinguish “R” and “L,” suggesting that early experience shapes language learning. Pullum & Scholz’s critique of the “poverty of the stimulus,” which is mainly applicable to learning language through grammar rules, also fits here. What appears innate might actually be the accumulated result of experience-driven adaptation. In that sense, grammar isn’t an inborn template but an emergent structure shaped by feedback between biology and culture.
Delete9.b. Pullum and Scholz question whether children really face a “poverty of the stimulus” when learning language. They argue that input is richer than Chomskyans claim, which makes sense, but their response feels a bit too confident. Even if data are more available, kids still generalize from surprisingly few examples, and they don’t explain how that happens. Their paper is solid in exposing weaknesses in nativist arguments, but it leaves you wanting a clearer alternative for how exposure alone could lead to complex grammatical knowledge.
ReplyDeleteI really liked how Pullum and Scholz basically call out the “poverty of the stimulus” argument as unfinished business. They write that “a program for research that might one day be undertaken should not be confused with one already completed,” which is such a polite academic way of saying: stop pretending this has been proven. It honestly surprised me how much of the nativist argument seems to rest on faith rather than data. Theyre not saying kids don’t have any innate capacity, just that linguists haven’t shown why that has to include specific grammar rules. It made me think about the symbol grounding problem, how we can describe symbol manipulation or grammar rules forever, but that doesn’t explain how meaning actually gets learned or grounded. So if the poverty argument doesn’t hold up empirically, why do people still treat it like it’s settled fact? habit? prestige? or just lack of better evidence?
ReplyDeletePullum and Scholz question the idea that children do not get enough input to learn grammar. They say many poverty of the stimulus claims are based on assumptions, not real data about what children hear. Since children are exposed to a lot of language, it is possible they learn more from experience than researchers believed. They also show that English has more examples of certain patterns, like plural compounds, than people thought. This means children might already get the right examples to learn grammar without needing it to be built into their brains. The paper suggests that language learning may depend more on rich exposure and the ability to notice patterns, not just on an inborn grammar system.
ReplyDeletePullum and Scholz consider the possibility that children do in fact receive enough input during typical development to account for the grammatical rules they produce. With this approach, rather than having underlying grammatical structures that are universal across all languages, maybe inborn capacities for pattern recognition is what is universal, and the fast acquisition of grammar is just an emergent property from that. If this is the case, however, then why are some grammatical behaviours common across all languages? Is it that they address things that all humans experience regardless of cultural and linguistic differences?
ReplyDeleteIn Pullum and Scholz’s paper they assess the evidence for the Argument from Poverty of the Stimulus (APS), the key argument supporting linguistic nativism. They emphasize that proponents of the APS must empirically demonstrate that knowledge (the acquirendum) is attained despite the inaccessibility of crucial data (the lacuna). They find that linguists have consistently failed to undertake the necessary empirical work often relying on unsupported assertions of data rarity. Such as the celebrated APS case involving auxiliary-initial clauses (like Is the dog that is in the corner hungry?), which supposedly proves children learn a structure-dependent rule innately rests on the claim that crucial examples are "vanishingly rare". the authors demonstrate that these complex interrogatives, which falsify structure-independent hypotheses are easily found even in transcripts of language used with young children. They conclude that the evidence for data scarcity is an overstatement. By investigating specific cases like compound plurals and auxiliary sequences, Pullum and Scholz seek to shift the burden of proof concluding that claims for innate language mechanisms remain speculative until more comprehensive corpus linguistics and formal learning theory research is complete
ReplyDeleteTo add to your point Duru: what I found to be particularly interesting is that Pullum and Scholz highlight that there is a very vague assessment of what it means for a specific construction to be "rare" in a significant fashion. Never have proponents of the APS really established a threshold for the minimal necessary presence of a construction in the stimuli to which the child is exposed for empirical learning to be made possible.
DeleteI am very curious as to how this could be studied. I know frighteningly little about computational linguistics: would anyone have any ideas or reading recommendations?
What stood out to me is how the authors show that many linguists blur the distinction between OG (the specific rule patterns of a language like English) and UG (the hypothesized innate, species-wide linguistic structure). Their critique is that PoS arguments often take a fact about OG (ex. a constraint in English) and treat it as if it must come from UG, even though no one has shown that the child couldn’t have learned it from the input. To make a valid PoS argument, you first have to prove that the relevant OG rule cannot be learned from experience, but the authors argue that this crucial step has never been demonstrated. The flaw they point out is that Chomsky’s argument doesn’t prove the rule is innate, it assumes innateness on the basis of untested claims about what children supposedly never hear. You need evidence of true data unavaliabilty before attributing things to UG. But as others have pointed out, children are exposed to extremely large amounts of linguistic input. So how are researchers supposed to empirically study the full range of what children actually hear in a scientifically meaningful way?
ReplyDeleteWhat I understood from this article is that A) Much of the existing literature supporting the Poverty of the Stimulus argument (POS) has lacked a clear definition of the question. And B) That their evidence relies heavily on claims about the existence (or non-existence) of certain examples of UG in the corpus of language that has not yet occurred comprehensively.
ReplyDeleteOk, I find it interesting how Pullum defines the question of the POS explicitly as not a result of the lack of “negative” examples of UG, but rather as a result of no evidence at all. This seems to deviate from how we speak about the POS in class, as prof Harnad asserts that it is a complete lack of negative evidence. So are they saying the same thing just in two different ways? Or are they claiming two different mechanisms of the POS?
“We conclude that linguists have some additional work to do if they wish to sustain their claims about having provided support for linguistic nativism…”
ReplyDeleteI found this line very intersting because it is the opposite of the usual direction of the poverty-of-stimulus debate. Instead of accepting the standard narrative that children must rely on innate linguistic knowledge because the input is too poor, Pullum & Scholz argue that the evidence for that claim itself is poor. They show that many famous APS examples fall apart once you actually look for real support or clarify what counts as “crucial evidence.” What really interested me is the implication that the nativist argument succeeds because it treats absence-of-evidence as evidence-of-absence without ever rigorously checking either. If their critique is right, how much of our confidence in linguistic nativism is based on assumptions rather than empirical foundations? If the burden turns back to linguists, what would a solid, well-documented APS case even look like, and why has no one produced one after decades of claiming they exist?
While there is no clear evidence for the Argument for Poverty of the Stimulus this does not mean that there is not any chance that children have an innate ability to learn language, it just means that current evidence isn’t convincing enough. Language learning at its core is the ability to form rules from limited data, where that data is language that is typically seen or heard. This allows children to observe the use of language in many different contexts. For the poverty of stimulus to be proven it needs to be shown that people are able to know linguistic rules that they have never actually learned or experienced in any context. This is difficult because it is often hard to prove in an experiment that people have never experienced these rules or even that the rule chosen is actually grammatical. But this doesn’t mean the APS should be given up entirely. Linguists should instead focus on different data-driven learning to see what can and can’t be learned.
ReplyDeleteOne thing that speaks to me about the Pullum and Scholz paper is that it is careful about what it insinuates. They aren't saying the APS is totally false. Instead, they argue that linguists haven't done the required empirical work to prove that the claims about stimulus poverty are actually true. They lay out a precise schema requiring things like formal learnability theory and rigorous analysis of linguistic data for support, which they say is lacking in celebrated APS examples. Their main goal is to reduce skepticism about purely "data-driven learning" theories, arguing that disproving the APS cases weakens the belief that empiricist methods are impossible for language acquisition. They predict that searching for the "crucial evidence" the APS claims is missing might just result in finding it, undercutting nativist claims.
ReplyDelete“Advocates of the APS… must develop some explicit and quantitative answers to such questions if they wish to support their claim.”
ReplyDeleteWhat stood out to me in this passage is how direct the authors are: they basically say that the Poverty of the Stimulus argument can’t rely on intuition anymore. If someone wants to argue that children “cannot” learn a certain structure from experience, they need to show numbers, real data, and clear thresholds not guesses about what children probably hear or don’t hear.
I like this point because it makes the debate more scientific. Instead of imagining what the input looks like, Pullum & Scholz want researchers to actually measure it. But at the same time, I think this also shows why the problem is so hard: children hear thousands of sentences every day in unpredictable contexts, and no study can capture all of that perfectly.
So for me, the message isn’t that APS is wrong, but that it can’t be treated as “proven” until we understand the real size and variety of children’s input. The paper made me realize that the biggest challenge is not deciding between innate grammar or learning, it’s figuring out what the child’s environment truly contains in the first place.