“I am a junior undergraduate student majoring in linguistics and have recently started conducting brain imaging studies. . . .”

Posted on by

Isabella Lai writes:

I am a junior undergraduate student majoring in linguistics and have recently started conducting brain imaging studies.

Yesterday, I came across a paper published in Nature Human Behavior by Grand, Blank, Pereira, and Fedorenko that raised several concerns for me. The paper attempts to find word-embedding using Amazon Mechanical Turk, or, in their words, “investigates context-dependent knowledge using semantic projection in word embeddings”, but the methodology might have a few issues. Take the simplest example, regarding the use of Pearson’s correlation as an evaluation measure for concept acquisition, which might be potentially misleading. The acquisition process for linguistic concepts with multiple dimensions might involve phase transitions, and Pearson’s correlation, especially when outliers are smoothed out, measures the linear relationship between variables. It might not capture the nonlinear nuances of such phase transitions.

I thought you might be interested in this paper, as it is a paradigmatic study of Computational X or Computationalized X, where X is a humanities subject that has been taken over by machine learning models before a coherent, formalized theoretical foundation has been developed for this subject. Sometimes it is linguistics, other times it is psychology, and most likely X stands at the intersection of both subjects, for their shared status of being a modern science, and their shared lack of a well-defined mathematical basis.

I have been thinking about the neural bases of language since college, and after a few hands-on experiences in data preprocessing using MEG and EEG, I have come up with a tentative hypothesis that I hope to share with you. My proposal is that the lack of evidence for an exact neural correlate of language or any theoretical concepts in linguistics (e.g., Professor Chomsky’s favorite merge combinator and lexical access) might actually be evidence that language is an innate capacity. In other words, we might not need an additional consumption of blood glucose when using language, regardless of whether it is for computational purposes (talking to oneself via the “I-language”) or for communication purposes because language is the air of our thoughts. It might be the case that poverty of the stimulus is trivially true, but a median correlation of 0.47 and 52 category-feature pair are inadequate to disprove it.

I am curious about your opinion on this overly naive hypothesis (but it might be just parsimonious, if it turns out that it can be developed, and might one day go beyond being trivially true) from a statistical perspective.

My reply: You raise two issues: (1) the use of theory-free statistical analysis to obtain data summaries that are then used to make general scientific conclusions, and (2) evidence for neural correlates of language etc.

For item #1, I’ll just say there’s nothing wrong with looking at correlations and other theory-free data summaries as a way to turn up interesting patterns that then can be studied more carefully. The process goes like this: (a) root around in all sorts of data, run all sorts of experiments, calculate all sorts of things, look for interesting correlations; (b) consider “interesting correlations” as anomalies in existing informal theories about the world; (c) form specific hypotheses and test them. (Here I’m using the idea of “testing” hypothesis in the sense of science, not so-called “hypothesis testing” in statistics.)

For item #2, I have no idea. This one’s not just outside my areas of expertise, it’s also outside anything I’ve ever really thought about.

Lai adds:

On a separate note, I read your blog post yesterday discussing John Tierney’s opinion piece on school shootings and the potential negative impact of active-shooter drills. I hope to share my intuition, that school shootings are fat-tailed events, and the extreme cases involving psychopathic individuals are more likely to occur than the default priors. As the media reports on and exposes more school shootings, psychopaths who previously may not have considered such a possibility are now more likely to view school shootings as a newly discovered option. The increased visibility of these incidents might contribute to a compounding rise in the frequency of school shootings, which might not be mitigated, unless through interventions like implementing better gun control measures.

Yeah, that’s pretty much my take too. But I don’t have any evidence on this concern, one way or another.

The significance of a reliable crypto management tool like Ledger Live Desktop cannot be overstated. It ensures that your investment strategies are well-informed and executed efficiently.

23 thoughts on ““I am a junior undergraduate student majoring in linguistics and have recently started conducting brain imaging studies. . . .””

    • I could not penetrate the jargon. As far as I got towards figuring out what they actually measured was this:

      For instance, a typical question read, ‘How likely are you to describe animals as large/big/huge or small/little/tiny?’. The 153 category–feature pairs were rated on a scale from 1 (‘not likely at all’) to 5 (‘extremely likely’).

      I guess there were other questions like:

      ‘How likely are you to describe clothes as large/big/huge or small/little/tiny?’
      ‘How likely are you to describe animals as dangerous or safe?’
      ‘How likely are you to describe clothes as dangerous or safe?’

      Then they found that people said they were more likely to refer to animals as dangerous/safe than they were to refer to clothes that way. This was also the case in a corpus of english text. From that they conclude:

      “the geometry of word embeddings explicitly represents a wealth of context-dependent world knowledge”

      I don’t get it.

      Reply ↓

      • I have not even looked at the paper, but that sentence means that if you refer to animals vs. clothes as dangerous/safe, you have to be using world knowledge to make that choice for animals (and for clothes). From that it follows (for the authors) that world knowledge determines which words co-occur with which others.

        I am guessing that Isabella comes from a theoretical orthodoxy that takes the Chomskyan poverty of the stimulus argument for granted (it’s the idea that we don’t acquire language through probabilistic exposure to structures and words, we have pre-set informative priors, in Bayesian terms; language acquisition is then able to derive posteriors based on sparse data). Geoff Pullum went after this idea once: e.g., https://journals.linguisticsociety.org/proceedings/index.php/BLS/article/view/1336/1120

        I don’t know anything about acquisition, but watching my child grow up and acquiring language, there were moments when I felt that the generalization he made at a particular point was not possible without some kind of prior in place that was generating, so to speak, posterior predictive utterances based on the data seen so far (I should have made a note of that observation then, but as an exhausted dad I had other concerns at the time). At the same time, probabilistic acquisition of knowledge seems to play a huge role. But it’s hard to just come up with a yes/no answer to questions like “is language innate” or “is language acquired through exposure”? The answer is probably a complicated mix of those two extreme positions.

        From what I can understand from the quote above, the Fedorenko-group paper seems to be highlighting the importance of world knowledge in developing word associations? Doesn’t seem that controversial to me, and I don’t feel that would contradict a moderate version of the chomskyan position about innateness and poverty of stimulus, etc, one which allows probabilistic exposure to inform the posteriors of the parameters, so to speak. I didn’t read the paper though, so maybe they make a stronger claim (I would be surprised if yes).

        Reply ↓

        • I have not even looked at the paper, but that sentence means that if you refer to animals vs. clothes as dangerous/safe, you have to be using world knowledge to make that choice for animals (and for clothes). From that it follows (for the authors) that world knowledge determines which words co-occur with which others.

          This seems to contradict what I have observed about the world, which is that people rely on consensus/authority (+habit) heuristics almost all the time.

          If you drop me in the jungle and everyone there is talking about how dangerous/safe various mushrooms are to eat, but no one is talking about the fish, then I would gather I should be concerned about my choice of mushroom but not fish.

          This does not seem to require any actual world knowledge on my part. Is do what “other people do” or “important people say to do” considered world knowledge?

  3. “The increased visibility of these incidents might contribute to a compounding rise in the frequency of school shootings,”

    Its interesting how this idea is recieved in the context of “regular” or traditional media vs. “social” media. If social media was deemed to be a cause of school shootings I can’t even imagine the outrage that would be expressed by its many fanatical critics. But when “traditional” media hypes the bloody hell out of everything – climate change, school shootings, wars, racism, homelessness – it’s not a problem. There are no Frances Haugens “whistle blowing” to committees of like-minded female Senators on the threat traditional media poses to safety of our children.

    Many historians believe the media hype in the press of all the involved European powers directly contributed to the explosion of WWI and the resulting deaths of millions.

    Shrug.

    I guess traditional media hype is a natural uncontrollable force, unlike the climate, which will readily submit to our will if only we pass the appropriate legislation. But I’m not sure which – climate change or tradtional media hype – will cause more destruction in society. IMO traditional media hype has a big lead. Might be hard to catch.

    Reply ↓

  4. Hello,

    I’m wondering if Isabella Lai is not real but AI generated?

    She says: “I am a *junior undergraduate student* majoring in linguistics and have recently started conducting brain imaging studies” and then later says “I have been thinking about the neural bases of language *since college*” I found the text weirdly written as well. Also, if you google her there is a LinkedIn image that looks not quite like a photo as well.

    But maybe I’m just paranoid?

    Reply ↓

  5. how do you know the email was written a few months ago? and why does it matter when it was written?

    and how are so sure about the photo? what’s the basis for your claims?

    Reply ↓

    • Anon:

      The email indeed arrived on 29 Apr 2023. It’s fun to have this sort of discussion—I guess the end stage will be a blog run by a bot that does nothing but supply bot-generated responses to bot-generated questions and is then read and cited by other bots, keeping the human out of the loop entirely—; but for now I’ll just take these questions as they come in.

      Actually, come to think of it, just the other day I was grading a homework assignment and it looked computer-generated to me. I wonder if I should start requiring they be hand-written.

      Reply ↓

    • I’m Isabella. Sorry for the bot-like texts, but a bot would at least ensure that the grammar is correct. It was an initial cold approach email, and it would be even more unnatural if there’s no brief self introduction.

      Reply ↓

  7. Regarding the neural correlate part (“My proposal is that the lack of evidence for an exact neural correlate of language or any theoretical concepts in linguistics (e.g., Professor Chomsky’s favorite merge combinator and lexical access) might actually be evidence that language is an innate capacity. In other words, we might not need an additional consumption of blood glucose when using language, regardless of whether it is for computational purposes (talking to oneself via the “I-language”) or for communication purposes because language is the air of our thoughts”) – I’m not a language expert, but I am familiar with neuroimaging, and this brought up two thoughts for me:

    I’m not sure what she means by ‘lack of evidence for an exact neural correlate’, as there are a number of known language-related brain areas (Wernicke’s and Broca’s are classics). But it reminds me of something I read once, about how the most shocking result you could have in a neuroimaging study would be to run two different tasks, compare them, and find no differences in the brain at all. It could be taken as evidence of Cartesian dualism, assuming the study was conducted properly.

    The other thought is that there are plenty of other innate processes that have obvious neural correlates and use blood glucose. For example, humans can innately see things (not necessarily name or understand, but simply see), and you get very robust neural activity in visual cortex in many different experiments. Looking at something as opposed to having your eyes closed is related with more glucose use, and doing more visual processing compared to less is also related with more glucose use. I think Isabella’s language argument would have to be something along the lines of we are constantly using language at a constant level, or constant enough that any differences don’t show up in neuroimaging.

    Reply ↓

    • Alex, thank you for sharing your thoughts and experience in imaging.

      Chomsky is skeptical about neural correlates because “neurons compute too slow” (which is certainly not true as we can, and have learned so much from eye-tracking alone; I guess his point was about fMRI’s temporal resolution), and he is in favor of biolinguistics, which is still linguistics as a rocket science, but from a perspective of constraint – an innate tendency to acquire language due to the lack of a better alternative (physiological constraints and cognitive limitations).

      That very naive comment from Isabella (who has no first-hand experience in imaging studies) was based only on intuition, that:
      1. linguistics is not a rocket science, too many path-dependent factors confused with randomness (or the diversity of parameters and diversity of native speaker’s judgements);
      2. innateness is a benign truism, there is no need to find a neural correlate to confirm this self-evident truth empirically;
      3. neuroimaging is quite expensive in terms of what it can contribute to in language processing – many foundational, well-replicated studies were done using only eye-tracking and EEG;
      4. Broca’s area and Wernicke’s area shine up in tasks aren’t only “linguistic” as well, e.g. elevated activity at Broca’s area has been observed in processing disharmonious music syntax as well, and Wernicke’s area has been settled down in recent years to be a region responsible for speech planning (before the sentence is uttered).

      Reply ↓

    • MRI machines have a resolution of about 1.5 to 2.0 mm; that’s an enormous number of neurons. And temporal resolution is really bad, too, 20 to 30 ms. They only light up _after_ a region of the brain has done some work and starts consuming oxygen to catch up reserves. So it’s a really fuzzy mushy signal/image.

      So the idea that we haven’t seen something indicates or means anything is untenable.

      Besides, even if language processing is “innate”, it’s still processing that will require energy.

      It seems the philisophical and linguistic arguments over inate/learned are somewhat problematic, shall we say. It’s obvious that human brains come with enormous gobs of amazing innate stuff, since it obviously is doing things that nothing else around comes even close to. Whether that “something” is the sorts of things linguists and philosophers are arguing about or not is, well, an argument. It’s important to note here that linguists are grinding an axe, often implicitly, but they do state it/claim it explicitly. Namely, Chomskian linguists is strongly committed to the idea that it’s possible to have a “science of language” without any concern for the human ability to reason about the world, i.e. it’s possible to have a “science of language” without dealing with meaning. At a naive level, this is nuts; language is about meaning.

      As you might guess, I think the bad guys won the linguistics wars. Your oppinion may differ. And the generative semanticists can be accused of having had too much fun with their examples. But they were, IMHO, less wrong than the Chomskites.

      But I digress…

      Reply ↓

Leave a Reply

Your email address will not be published. Required fields are marked *