We do sort of have “grammar” in our brain, but Chomsky wasn’t more right or wrong before Ding’s paper.

In a recent paper, Nia Ding and colleagues in David Poeppel’s lab, made a big discovery: When understanding language, our brains are sensitive to the abstract hierarchical structure that governs it.

The reaction:

However, the press (http://www.nyu.edu/about/news-publications/news/2015/12/07/chomsky-was-right-nyu-researchers-find-we-do-have-a-grammar-in-our-head.html) and the Chomskian’s seem to have jumped on this and interpreted these findings as support for Chomsky’s theory of language.
This claim was then repeated by other scientific news outlets (e.g., here and here) and has even spurred some heated enthousiasm/discussion on (Reddit).

In the original NYU press release, James Devitt, of NYU press office, makes the bold, and completely unsupported claim that “Chomsky was right, […] We do have a “grammar” in our head.”

James (of NYU)… Really?
[EDIT: 9/12/15 – 7:46PM CET] Dana Dovey (of Medical Daily)… Really?[/EDIT]

The Study:

Here’s a simplified summary of what happens in the study:

Researchers presented sentences to speakers such as the ones below:
– “Dry fur rubs skin”
– “New plans gave hope”

If you’ve never studied language formally before (i.e., taken a linguistics course) you may be unaware that it is commonly assumed that behind these words, hides a certain hierarchy which governs how the words should be organised. It could look something like that:

Figure 1. Schema of the hierarchical structure of the sentence “New plans gave hope. Adapted from Ding et al., 2015.

By recording the activity of those speakers’ brains using a technique called magnetoencephalography (MEG) the researchers found that not only does the brain activates (i.e., does some processing work) when it is presented with the individual words, but it also activates when a phrase has just been presented to the brain. That is when “new plans” has just been presented but also when “gave hope” has been presented and once more when the whole sentence has been processed.

Schematically this is what happens in the brain:
Figure 2: Schematic representation of the activity patterns uncovered by Ding et al. (2015). In reality this is of course much more complicated but I invite you to read the paperfor that.

Now, this is really cool. It shows that the brain is able to derive structure from this stimuli. By this, I mean that the brain abstracts over the specific items it is processing.

In those specific examples it is probably now parsing the two sub groups as who (“new plans”) and “what” (“gave hope”) –phrase level– and then how what is caused by who –sentence level.

What’s even cooler is that the brain can do that even though it is likely that it has never heard these specific items in that specific order before, especially in the case of the “dry fur rubs skin” one.

Back to the reaction:

So, what do we conclude from that?
Well the authors in the paper are very very parsimonious and refrain from bold claims, it is important to note.

[EDIT: 10/12/15, 8.59AM] Although… It would seem that the article on Medical Daily has included quotes from the senior author of the research paper, David Poeppel, who may have steered these inaccuracies on the par of the journalist. Here’s a screen capture of Peoppel’s quote… Bad, Bad David!
Screen Shot 2015-12-10 at 8.55.00 am

The big take home message is that, the brain extracts/”imposes” some sort of hierarchy when processing language. **[EDIT 10/12/15, 11.21AM: David Adger raised a point about my word choice of “extracts”. According to him I should use “impose” because indeed there is no hierarchy to “extract” from the stimuli as they are. I added the alternate wording]

Now, is this support for Chomsky’s idea of language?

NO, big big no.

Sure, as a facebook user already pointed out to me, in response to my original reaction on a colleague’s post:
– What’s the problem? Chomsky’s idea of language is such that it has hierarchy so those data are compatible with his theory.”

But so are the data compatible with most theories of language which all acknowledge the fact that language is hierarchical in some way.
But so are the data compatible with most theories of learning, which all acknowledge that the brain abstracts over sensory experience to build some “hierarchy”
But so are the data compatible with the fact that the brain is itself wired to promote the emergence of hierarchy.

Chomsky isn’t more right (or wrong) now than he was before!

Indeed, that data does not say anything of its specificity to language, it does not preclude that this “grammar” could have very well emerged from any kind of learning and exposure with language and it has yet to be proven that this pattern of activation would not be found in other animals.
Finally, the researchers find widespread activations, covering large areas of both the left and right hemisphere of the brain. If such a distributed network is support for the “modularity” of language, then thinking –all thinking is just as modular as language!

So why choose Chomsky as the only theory that this data supports, I don’t know… Maybe because he is famous and that’s sensational?

What now?

Well, I’m on my way to get my broken jaw fixed. Really… I fell off my bike on Friday!


Ding, N., Melloni, L., Zhang, H., Tian, X., & Poeppel, D. (2015). Cortical tracking of hierarchical linguistic structures in connected speech Nature Neuroscience DOI: 10.1038/nn.4186


  1. Great job going through the data and claims systematically. Appalling science journalism on the part of Medical Daily and NYU’s press office, they should have gone back to the peer-reviewed source to get context. I must say that David Poeppel’s comment re Chomsky was totally unnecessary, though, and this was perhaps the root of all the issues that followed. I can understand that he wanted to hook his research into Chomsky’s public notoriety and gain a bit of publicity off of that, but it wasn’t totally accurate and – indeed – if it was hugely relevant to Chomsky’s research then that would have included in the peer-reviewed discussion, which it wasn’t (and for good reason).

    • bastienboutonnet (Author)

      Thanks for reading and showing your support. Glad to know I was not the only one in shock at how badly this was reported.

  2. David Adger

    Not sure your characterisation is right here. You say the big message is that the brain ‘extracts’ a hierarchy. But the paper shows that the hierarchy is not extractable from the acoustic or statistical properties of the stimulae. The paper shows that the brain rather ‘imposes’ hierarchy on data, suggesting strongly that tree is a mentally represented grammar that does this. And that’s exactly what Chomsky suggested a long time ago (that apparently abstract hierarchy is mentally and hence neurally real) and we now have nice physical evidence for this. Now it’s true that some non generative approaches admit hierarchical grammatical structures, but not all do by any means. Think of the Frank, Bod and Christiansen paper in proceedings of the Royal society a few years back. And it’s true that, for example, Construction Grammar, takes hierarchical structures to be learned properties that emerge from applying general processes of abstraction during language acquisition (though I’ve never understood how that’s meant to scale up to actual linguistic phenomena). You’re right that this paper doesn’t address issues the issues of domain specificity or canalisation which are associated with Chomsky, and is consistent with , say, a mentally represented Construction Grammar, but it’s been Chomsky that has argued over and over again, for many years, for the `psychological reality’ of the abstract structures proposed by linguists (see, for example, Rules and Representations) against quite vociferous opposition from philosophers, neuroscientists (e.g. the Churchlands) and psychologists (e.g. Christiansen/Chater). He was without doubt the first to advocate the mental reality of hierarchically structured syntactic mental representations, and so the headline, and David’s comments in the piece, are actually pretty accurate.

    • bastienboutonnet (Author)

      Hi David,

      Good points. The main issue is with how this paper was reported and sold to the press. Which we all agree is inaccurate. I will edit the summarisation I make to incorporate your point, although I think to a lay audiance the nuance in the language is not relevant. I think actually, that the data do not show that the brain necessarily imposes structure. Sure it does not “extract” because it’s not there in the stimuli. So insofar as it “imposes” structure, it could do so from an acquired experience with language, even in the absence of other cues that the authors went to great lengths to control.

      I think the authors are very very honest in their paper, but that the press is mislead and in turn misleading.

      My article, has no intention to question the data or to question Chomskyan linguistics (although I honestly am wholly in disfavour of it). It’s purpose: reacting to an extremely inaccurate reporting of research to the general public.

      • David Adger

        thanks for the reply Bastien – I’m sure we’ll disagree, but at least the NYU piece just says “A team of neuroscientists has found new support for MIT linguist Noam Chomsky’s decades-old theory that we possess an “internal grammar” that allows us to comprehend even nonsensical phrases.” I actually think that, at least is accurate. They did find support for the theory, it is a decades old point that Chomsky specifically has made, and the content of the theory is that there’s an internal grammar that’s used to assign hierarchical structures to linguistic input. Did the press say anywhere that it supported the view that the hierarchy does not come from acquired experience? I didn’t see that anywhere, and if they did, then that’s completely inaccurate, I agree. (I think the issue is maybe that when people read `chomsky was right’ they are reading that as a universal statement about everything Chomsky has said, or perhaps as a statement about what people think is the only thing that Chomsky has said – that there is an innate UG, whereas it is an existential statement about something Chomsky said – that grammars and the abstract structures they assign are psychologically and neurally real, to the extent that that matters. On that reading, I think, what the press link you gave reports on, quite accurately.

        I’m actually quite interested in what you wrote when you said “I think actually, that the data do not show that the brain necessarily imposes structure. Sure it does not “extract” because it’s not there in the stimuli. So insofar as it “imposes” structure, it could do so from an acquired experience with language”. I don’t think I understand this. I agree that imposed structure could come from acquired experience, in fact at least some of it has to, as speakers of different languages presumably impose different kinds of structure, but how do the data not show that “the brain necessarily imposes structure”? Do you mean that a grammar that has been learned doesn’t impose structure? Sorry to be obtuse, but I’m not really understanding what you wrote.

        • bastienboutonnet (Author)

          Just briefly. David, theoretical disagreements aside which are not the point of this post and in which I hate to engage –especially in public channels, we’ve got to understand the implication for the general audience. Chomsky’s idea of language **comes** with innateness. Saying that such data lends support to this claim is therefore misleading. It does support one tenet of the theory, but then at least the reporting should be fair and explain that Chomsky’s idea of language has many more implications, which the data cannot speak of. It is therefore not a parsimonious interpretation and that is what I think is misleading.

          The main issue, is a problem with the pressures of reporting scientific research as *sexy*. Sometimes it goes to far, and can mislead very quickly.

          Wrt the clarification, what I am saying is that a) yes the data does not speak of innateness, b) does not speak of specificity to language, c) I don’t like the term “impose”. I think it is fair to say that even after the extreme control of the stimuli used in this experiment, the brain can simply abstract over groups of words (in a structural manner) like it has learned to do and like it would do for other types of stimuli (cf., Tom’s comment). Moreover, in a predictive view of the brain, we can easily expect that the brain brings about all sorts of expectations of how to parse linguistic stimuli, these expectations could very well dictate how to group things based on the statistical regularities of the language in general. What the patterns of activity recorded show is that the brain registers, or “imposes” if you want these regularities. But that’s not more chomskyan than anything, at least in my opinion.

          Ultimately, I think this is a theoretical debate of the kind that is far more complicated to resolve and discuss on such channels. And completely besides the point of the post I made. I would not want to be quoted on such issues too much. And I completely agree that my vision would differ from yours (or the ones of many others) but that’s fine. It happens for all theoretical accounts of pretty much anything. I hope this helps.

          • David Adger

            Thanks for the clarification. So your basic worry is that people will misread the headline of the press release (because the actual content of that release is accurate). I’m pretty sure David had no control over the headline – you normally don’t. I remember the new scientist reporting that my PNAS paper with Jenny Culbertson had shown that there was a `language instinct’, which was particularly infuriating as I hate that term, but we had no control over the headline (or much of the content of the article!). I understand that annoyance, but I thought that the actual content of the NYU piece at least was reasonably accurate.

            I’ll not enter into the theoretical issues beyond just saying that my own view is that generative linguistics and neurolinguistics are both edging towards each other as syntactic theory gets more refined. Of course the point of generative linguistics is to explain linguistic phenomena, but it definitely comes along with claims about mental reality (which this paper supports) and the precise nature of the systems that construct/abstract a grammar from the data. The big issue is, as you say, whether such systems are keyed to specifically linguistic input (analogously to systems keyed to visual input, say), or whether they are domain general processes of statistical processing. Clearly the latter are required, but I’m skeptical about whether they are sufficient, mainly for poverty of stimulus type reasons. But the only way we’ll find that out is by investigating it, and I personally think it’s good to investigate that issue from as many directions as possible, as it’s pretty hard!

          • bastienboutonnet (Author)

            I fully agree with you. And I am glad to see that more generativists are now approaching the “psychological reality” of their theory. And that’s what science is about.

            My issue is indeed with the wording of the press release. It’s title but also the fact that a single theory (Chomsky’s) is mentioned as supported. Since this theory makes a bunch of other strong claims about language that other theories actually do not, I am annoyed that this was the one chosen to be highlighted by the journalist and also most likely pushed forward by the senior author. He’s not to blame personally since I think he’s victim of pressures of sensationalising science, for which the name of Chomsky would be a good pivot.

            This was an interesting exchange however! Very much appreciated. B

      • David: In the sense you are speaking of, the brain imposes structure on everything – shapes, faces, voices, ideas, math, etc. – so I don’t see why finding that this is also true for language in the abstract gives special help to Chomsky above all other linguists. After all, he gained many of his ideas from more foundational work like that of Zellig Harris, who also presented arguments for hierarchical structure. Plus, it’s hardly a modern controversy to say language has a hierarchical form. Even if I were to agree with your characterisation of the top-down results, i.e. the ‘imposing’ of hierarchy, what’s written in the press release and subsequent media articles is more than a little far-fetched – indeed, if these results undoubtedly had the sort of ramifications they say, why was this not in the peer-reviewed discussion section?

        • David Adger

          Completely agree that the brain imposes structure in this sense more widely. But I read the press release (haven’t read other stuff, so links would be highly appreciated) as just saying that here was neural evidence for abstract hierarchical structure, and I think that is still controversial to many, viz the Proc.Roy.Soc article I mentioned (http://rspb.royalsocietypublishing.org/content/early/2012/09/05/rspb.2012.1741). The claim that there is a mind/brain reality to abstract syntactic hierarchy was definitely something that Chomsky was the first to say (Harris would have been horrified at that idea!) and was a huge bone of contention in the 70s. The issue is not that none else but Chomsky thinks that there’s hierarchy in language, it’s the fact that that hierarchy is mentally and hence neurally represented as such that was the controversy. Perhaps that doesn’t now seem controversial in the neuro world, in which case, great. But I still think that the actual content of the press release is both factual, and is in essence correct: this is support for a view first propounded by Chomsky that abstract hierarchy is mentally represented. They also did say in the press release that this was a decades old theory.

          Its important to separate out the separate claims: the reality of mental representations of syntactic hierarchy vs the aetiology of the system that imposes that hierarchy (i.e. are the principles of grammar domain specific or not). The paper, and I think the press release, are unambiguously about the former. I’m sure we disagree about the latter!

  3. Olaf K.

    I really do not understand the beef with saying that the paper shows that Chomsky was right when it comes to hierarchies. A hierarchical treatment of sentences does not predate Chomsky much (as far as I know) and it was certainly Chomsky who turned it into a full-fledged theory. So when the objection is that other theories do hierarchies too, then that’s fine, but they got that idea from Chomsky. So it’s still Chomsky’s idea. That the headline makes this a Chomskyan thing is completely deserved. And that Chomskyan comes with other stuff, that’s at most unfortunate but does not make the press release wrong in any way. It is certainly more right than press releases I have seen trumpeting AGAINST Chomsky.

    • bastienboutonnet (Author)

      It’s a question of subsets…
      Here are three theories:
      Y=(B+ not C)

      If B = TRUE, Y & Z are true but not X.

      This is both a logic and parsimony question.

      This therefore makes the press release misleading and, in fact, *yes* “wrong”.

      Regarding other headlines “trumpeting AGAINST Chomsky” that’s irrelevant for my point.

  4. Jeremy

    Does this line from the abstract of the paper not suggest a division between syntax and semantics that is distinctively Chomskyan, as opposed to say the Construction Grammar David mentions above: “Notably, the neural tracking of hierarchical linguistic structures was dissociated from the encoding of acoustic cues and from the predictability of incoming words.”

    Considering how prominent dissociation is in argumentation for generative linguistics, I read that as “Notably, the neural tracking of hierarchical linguistic structures (syntax) was dissociated from the encoding of acoustic cues and from the predictability of incoming words (I’m assuming this would be taken by some to be based on semantic considerations).”

    I just wonder if the authors of the paper were quite so parsimonious. Or, if the syntax-semantics divide was also a part of what they took to be their “big take home message.”

    • bastienboutonnet (Author)

      I don’t really have any problem with that divide potentially existing. I have no problem with the fact that the brain could have after exposure worked out a level of abstraction “specific” to “syntax”. Of course in reality it almost never works like that, so it’s impressive that this data was obtained but not impossible by any means. This still isn’t evidence for Chomsky’s theory of language any more than a theory that would posit that structure is worked out, and that the brain can “represent” it.

  5. Ambrosio

    Thank dude, really nice job explaining the findings. Kudos!

  6. David Adger

    Bastien, I don’t understand your answer to Olaf. Can you fill in the variables for this particular case. As far as I can understand, the press release said

    1. Chomsky was right
    2. What he was right about is that we have an abstract grammar in our heads

    This presupposes that

    3. Chomsky claimed that we have an abstract grammar in our heads

    It does not presuppose or entail

    4. No one else has claimed that we have an abstract grammar in our heads

    But there is some kind of weak implication that

    5. The claim that we have an abstract grammar in our heads is particularly associated with Chomsky.

    5 is, I think true (as I mentioned, he was the first to claim this against the prevailing Harrissian/Bloomfieldian views of the 50/60s). It’s probably the case that those people in cognitive science who believe 5 do as a result (possibly indirect) of Chomsky’s arguments against behaviourism in the late 50s.

    So given 5, it’s reasonable to claim 3 and deny 4 and hence 1 and 2 are correct. Since 1 and 2 is what the press release says, the press release is not, contrary to what you write, wrong. The piece in Medical Daily is, however, wrong and quite stupid.

  7. Bea

    Greetings! Very helpful advice on this article! It
    is the little changes that make the largest changes.
    Thanks a lot for sharing!

Leave a Reply