While there are definitely environments where using a ghostwriter is acceptable, academic publishing typically isn’t one of them.
The reason is simple: Using a ghostwriter on an academic paper entails having an author do significant work on the paper without receiving credit or having their work disclosed. This is broadly seen as a breach of authorship and an act of research misconduct unto itself.
Forgive me for being dense – what is the connection to Wolfram here?
Go here and search on Matthew Cook.
Thanks. That whole review is amazing.
Pretty amazing story. In it, Shalizi name drops some of my favorite researchers of the 20th century, like Herb Simon and Per Martin-Löf (and of course, Kolmogorov). I find it amusing that when he strays into linguistics, we get
Lakoff started off as a Chomskyan, before straying from pure syntax into the syntax/semantics interface. Disagreeing even a little bit with Chomsky made him a target in the so-called Linguistics Wars, which consisted of Chomsky trying to purge the field of people who disagreed with him. Like Chomsky’s other targets, Lakoff didn’t stray very far, and generative semantics was essentially a Chomskyan-style theory, so the “evidence” consisted an armchair appraisal of cognitive plausibility motivated by the competence / performance distinction. The competence/performance distinction is what led linguists away from empirical science into the “as I claimed in my earlier books” approach to cognition. Or, as John Lyons puts it, with typical British understatement,
That’s a bit like saying the Titanic disaster was an unfortunate bit of boating. Instead, the field is driven by so-called “linguistic intuitions” rather than what other scientists might think of as data, like for instance, what people actually say.
Ha! I didn’t realize that the Linguistics Wars had a wiki page. Thanks!
But the disagreement was fundamental, not just between flavors of Chomskyan-style theories. Chomsky argues that it’s not just possible, but intellectually/academically necessary, for linguistic theory to be independent of meaning, understanding, and the real world as we experience/grok it. The generative semanticists thought linguistics needed to deal with meaning.
Back in the days of GOFAI, we thought the independent of meaning idea really silly. Language is about communicating meanings and undestandings and contentful ideas. The idea of language without meaning is nuts. People really do think, and they really do use language to express what those thoughts tell them their readers ought to be told about (e.g. “LLMs are a parlor trick”).
So as far as I’m concerned, the bad guys won the wars. And LLMs are still a parlor trick.
By the way, I may disagree with them and think them to be dead wrong, but they’re not idiots, they’re some of the smartest people around. It’s real hard to argue with them, since they really are smart and they’ve looked at a lot of language. When I was first studying Japanese, I thought it really kewl that adjectives _conjugate for tense_. Ha! This ought to trash those idiot transformationalists, I thought. And went after Ray Jackendoff (a ridiculously talented clarinettist and amateur viola player who had my father repair his viola) with this neato factoid. He wasn’t phased at all. Huh? That’s normal language stuff (or something to that effect) he said. But, seriously. They do look at a lot of data and come up with theories that cover it amazingly well. It’s not just “linguistic intuitions”.
I agree that language is about communication, part of which is meaning. I would say that LLMs are only a “parlor trick” inasmuch as human language is a parlor trick. I would say that LLMs have very poor memories, but are very good at language. To say they don’t understand is to embrace a very narrow notion of understanding . I also think that you can get so far with transformers says more about language than all the linguistic theory ever published, which includes my own book on semantics. I’m not saying there’s no role for linguistic theorizing, just that we need to think about different foundations for it.
I don’t know of any linguistic theories that cover the data even passably. Certainly not something from Chomsky or any of the spinoffs (LFG, HPSG, etc.). Grab a theoretical linguist and try to have them analyze a news story or a novel or an academic article. Oh, right—they won’t do that out of principle! To me, the principle seems mainly one of not being laughed out of the room.
Bob, I feel compelled to point out that what psycholinguists and the like present as data is in most cases just a fancy and expensive and time-consuming way to confirm the theory that they believed in all along. So even though linguists transitioned away from mere intuition to what we might recognize as hard data, the end-product is not much different.
One really surprising thing I see again and again in psycholinguistics is that if a difference in means is in the right direction and almost significant, the authors jut go ahead and claim it’s significant. OTOH if a null result is desired, a p-value of 0.06 will do just fine to make the case that there is no difference.
Also, you will find that pretty much every hard-data scientist in linguistics almost always finds in their paper that they were right all along. It is very rare for a linguist–empiricist or intuition-driven–to publish a result that goes against their own theory.
At the end of the day, there doesn’t seem to be a huge difference between intuition and research based on hard data.
That sounds like an indictment of the methodology in both linguistics and psycholinguistics! But then I don’t see how that’s possible. The so-called “research” based on intuition is about competence, whereas things like reading time studies are about performance. Linguistic “theory” as such, isn’t about what people say. So I don’t see the connection. Sure, we’re all talking about nouns and verbs, but we never make that notion particularly precise in anyone’s theory.
To be clear, the kind of psycholinguistics results I find compelling are things that go beyond syntax, like Steedman’s garden-path work or Tannenhaus’s eye-tracking experiments. These showed that semantic/discourse coherence and real-world context allow us to resolve ambiguities in syntax. With eye tracking, we saw we could resolve a reference to “yellow cup” after hearing half of the word “yellow”.
I am just looking forward to an epidemic of AI written journal article submissions that will put this obsession about plagiarism and ghost writing into context.
Rahul:
I strongly object to the claim that being interested in a topic, and writing about that topic, represents “obsession.”
And I strongly onject to the claim that my obsession with this blog is a mere statement of interest.
And I even more strongly *object to the lack of an edit button.
And as if on cue, Alex Tabarrok at Marginal Revolution has a post today ” In defense of Plagiarism”
https://marginalrevolution.com/marginalrevolution/2024/03/in-defense-of-plagiarism.html
Rahul:
I agree with Tabarrok’s statement, “The focus should be on whether readers have been harmed by a fraudulent use of other people’s ideas and words.” This is consistent with what Basbøll and I wrote in our article, To throw away data: Plagiarism as a statistical crime, where we say that plagiarism “involves the hiding of important information regarding the source and context of the copied work in its original form. Such information can dramatically alter the statistical inferences made about the work.” It is the reader who loses.
Relatedly, work with plagiarism is typically of low quality—no surprise, given that plagiarism is a sign that the writer has not fully digested the material that has been copied—so errors creep in, errors that can be hard for readers to recognize, once the original source has been obscured.
When ChatGPT writes an article would we consider it plagiarism?
Maybe we will see a reversal now of high quality plagiarism.
To Andrew actually…
I _strongly_ disagree with “The focus should be on whether readers have been harmed by a fraudulent use of other people’s ideas and words.”
The problem with that approach is that you have to recognize and prove the harm. So what, you are thinking. Well, no. The problem is that you can’t identify all the harms caused.
This is like lead paint on children’s toys: you’d rather there not be any than that you get to sue the manufacturer after you find your child has been poisoned. Lead paint and plagiarism should be squelched far in advance of anyone getting exposed to them.
I wouldn’t put it as dramatically as David does, but I agree that the issue of whether readers have been harmed is not where “the focus” should be. OK, I’ll agree that that’s the _single_ biggest issue, but it’s not the only one.
For example: is the plagiarist benefitting in a way that is detrimental to society? If your plagiarism gets you a reputation as a brilliant thinker or a great writer, earning you a coveted job, then you are taking a slot that should (and otherwise might) go to someone who is _actually_ a brilliant thinker or a great writer.
Or: what about the person whose work was plagiarized? The plagiarism may or may not hurt them. Coincidentally, just ten minutes ago I watched a video by a custom cabinetmaker who had posted a video a couple of years ago about a cabinet…and a company copied the design and started selling them. No harm to the originator, right? Except that he got an email from someone asking whether he had copied his design from the company, flipping around who gets the credit!