Formalism/Theory distinction citation

Who was the first to talk about the distinction between the formalism and the language theory in HPSG? I did not find anything direct in Pollard in Sag (they do talk about modeling the empirical domain but that seems to be about theory; they do say that they “eschew extreme formalization” when talking about theory, so, that’s related but I am not sure I can infer the distinction from it as we understand it?). Nor in Sag, Wasow, Bender. The 2020 chapter in the handbook (by @ebender and @guyemerson) talks about the distinction directly and clearly but does not cite anything.

I think this comes up here:

Bender, Emily M. 2008. Grammar Engineering for Linguistic Hypothesis Testing. In Nicholas Gaylord, Alexis Palmer and Elias Ponvert, eds., Proceedings of the Texas Linguistics Society X Conference: Computational Linguistics for Less-Studied Languages. Stanford: CSLI Publications ONLINE. pp.16-36.

Can’t check just now if I cite anyone else for it, but if I don’t I probably should have…


Thanks, yes, that one also talks about it clearly (and certainly predates the 2020 handbook) but does not cite anything. Could it be that you were the first to talk about it in these terms?

It could be, but it equally could be that I wasn’t! Maybe a good query for the HPSG list?

1 Like

OK guess what:

theory will thus be formulated in a metalanguage to the language in
which grammars are written - a metametalanguage to any language for
which a grammar is constructed.

This is from Chomsky’s Syntactic Structures, a footnote on page 54 :). Suggested to me by Antonio Machicao y Premier as a response to my query on the HPSG list :).

Attributing this to Syntactic Structures would be buying into a creation myth (cf. Pullum, 2010).

1 Like


Well, maybe. Except I am not after a mathematically rigorous definition here, I want to know who was the first to say that formalism and theory are distinct. (So far, many have told me that it is sort of an axiom which does not necessarily require a citation; simply something which was always understood in sciences.) But at any rate, what Chomsky says in SS seems like that idea, and so far this is the earliest expression of that idea that I have found with respect to linguistics. The terms in which it is expressed are not exactly the clearest though and one has to sort of guess what the exact meaning of the quote is. I think the earliest quote I found which talks about the distinction in clear terms is Givón 1979. But Givón 1979 is written so angrily and emotionally (at least in that part where he talks about formalism/theory, page 7) that it makes me not want to rely on it as the main citation.

My impression is that a lot of the assertions of this from within HPSG at least are in response to later work in the Chomskyan tradition explicitly conflating theory & formalism, so an early Chomsky quote for the point would need some contextualization.

1 Like

Not if I am not focusing on the history of the argument though… I suppose I could say: Chomsky may have noted this very thing in a footnote and later, as argued especially strongly in Givon, it became apparent that this isn’t always adhered to in practice. In HPSG, it is.

If you’re looking for the history of the idea that formalism should be stable even while the theory changes, then Chomsky’s quote doesn’t support that. I’ve just looked up the original passage, and there’s nothing there to suggest that we should try to keep the “metametalanguage” stable. He’s arguing for a distinction between a particular grammar (formulated in a metalanguage) and a general theory (formulated in a metametalanguage). But that says nothing about formalism (in his terms, this would be formulated in a metametametalanguage, which is something he doesn’t mention).

If you really want to cast your net wide enough that the Chomsky quote is relevant, then I would say that the idea of a metalanguage (and metametalanguage, etc.) was well-known in formal logic before Chomsky, and was also applied to natural language before Chomsky, for example in Tarski’s work. The early work on Categorial Grammar (e.g. by Ajdukiewicz and Bar-Hillel) was already carefully formalised, and also pre-dates Chomsky. The sentence which has the footnote (“Completion of the latter two tasks will enable us to formulate a general theory of linguistic structure in which such notions as “phoneme in L”, “phrase in L”, “transformation in L” are defined for an arbitrary language L in terms of physical and distributional properties of utterances of L and formal properties of grammars of L.4”) clearly builds on Harris’s work (e.g. “Distributional Structure”), which in turn builds on Bloomfield’s work. If the footnote is relevant, so is a lot of previous work.

Either way, I’d be interested to see the collection of citations that you find.


Right. I think I am looking for the idea that the formalism and the theory are distinct (as obvious as it may sound). I think Chomsky’s logic can be extended to suggest that, and the reason I want to include the quote is that apparently it does come to mind, to some people, when the topic is mentioned. I get the “creationist myth” issue, however if the goal is to communicate something to people who do refer to Syntactic Structures a lot, then it could be counterproductive not to mention it. For example, this quote was suggested to me by Antonio—and he is an HPSG syntactician!

This is a very interesting discussion!

In which case, Chomsky’s quote isn’t directly relevant. He doesn’t talk about formalism, only theory.

Perhaps it deserves a footnote?

Okay, more seriously – in that case, it could make sense to mention the quote and explain how he fails to make the distinction, even when he comes close to discussing it.

From a logician point of view, yes. But examples could make this discussion more explicit. We have the language (e.g. English). We have many possible grammars for English written in many different languages. We have formalizations of such grammars (e.g. ERG) in different formalisms (e.g. HPSG). Am I am taking it right so far?

But of course it is a footnote! You underestimate me ;). The entire thing is a footnote, for that matter (should I make a footnote in a footnote?).

Here’s what I have so far:

What do you think?

Well, yes, but then we also have a theory of human language, and that’s where it becomes a little more muddy. There is the HPSG theory and the HPSG formalism, for instance; those are distinct. An example would be perhaps something like: “Unification is the only operation needed to describe human language.” That would be a theoretical claim, I think. But the same formalism can be used to suggest theories which assume other operations as necessary and say something like: “Using shuffle operators as well as unification allows for an elegant account of this set of data”. Both would be HPSG theories.

Haha, nice.

I’m not sure this is true. I meant that they understood the difference between an object language and a metalanguage – if Chomsky’s quote is relevant, then their work is also relevant. I don’t think any of these people make a distinction between formalism and theory. I don’t understand the reference to Haspelmath, as he doesn’t really work on formalism.

I would be very wary of forcing the formalism/theory distinction onto work that doesn’t make that distinction. I mean, It’s not an obvious decision to say that you don’t just want a formalism that’s general enough for your current theory, but a formalism that’s also general enough for any future theory you might come up with. This is the HPSG approach, and I find the motivations for it compelling, but there are other well-formalised approaches that don’t do this – for instance, Steedman argues for the CCG formalism precisely because it is restricted, Joshi did the same for TAG, etc. This is a difference in perspective from HPSG. A vague reference to theory or formalisation could be compatible with either of these two perspectives. But the other perspective doesn’t make the formalism/theory distinction.

1 Like

I also found the following discussion (outside of HPSG) in Chater et al.'s 2015 book “Empiricism and Language Learnability”, p.77:

Haspelmath talks about metalanguage; things like the fact that the language in which you talk about any science is distinct from the concepts in that science. He doesn’t work on formalisms but he lays out a system of definitions which relates to it.

Perhaps I should be clearer about who talks about “metalanguage” and who talks about “formalism”. Looks like maybe Givon is then the oldest reference for “formalism”.

Thanks for Chater quote! I’ll try to find the book; I’d need to know what they mean by “grammar-writing linguistic theory”… As is, I am not sure I understand the quote. At any rate, in this case, I am interested more in who said what about science/metalanguage and theory/formalism, and less about Chomskyan fallacies (though the two things are connected).

Actually, here’s, I’d say, how Haspelmath’s quote is relevant: his paper is about a “theory-free” approach, and in that context, he finds it important to emphasize that metalanguate is distinct from theory.