So we don't have a defined process model that can guide is in coming up with a bespoke statistical model.

An exmp of where theory is not strong enough. In the sentence “Pritesh emailed Themba. I saw him at the gym.”, the subjecthood hypothesis states that him refers to Pritesh, while under the syntactic parallelism hypothesis him refers to Themba. In reality, language users do either at varying degrees.

It includes an optimisation process in one of the three components, which makes regression unhelpful, for example.

Very often, the same expectation could come from very different process models. We really need to work more on process models to be able to sensibly come up with bespoke models. One exception is the XT/3C model of sound processing, currently being developed.

We might be able to be better than "there will be a difference" in certain cases, but current linguistic theory is not generally equipped for more precise expectations. Since we don't have "contrasting" behaviour to look at, bespoke models end up being regression models.

We can't even agree on basic concepts: dlc.hypotheses.org/3975

Our problem is that our process models are underspecified or non-existent, so we cannot really have bespoke models. It is just not possible with the current state of knowledge. (Some subfields within linguistics might be better equipped, but in general, they are not).

I love the idea of bespoke models, but we're not there yet in linguistics (we are still stuck in the conceptual-theory phase of "there will (not) be a difference between group A and B". Very underwhelming). I hope for more mathematic-computational process modelling. Wish list 2026 :D

For my own field (linguistics) IV is very difficult just because we know so little about the underlying processes (because most of the research focussed on I), but there is something stirring in the right direction. I think II is a good entry into IV for the time being.

And it applies to all research, not just "scientific" research (whatever scientific means, misquoting King Charles)!

And I mention them in my stats class/textbook (intending do design an improvers quant meth course where we do it more thoroughly).

Yup! osf.io/preprints/os... I’m using them for everything I do now (when relevant). Currently working on a DAG with @jess-hampton.bsky.social on language endangerment and biodiversity intactness.

Ultimately, it is relevant to all of research! I think we don’t discuss it enough in University. People working with qual meth are more aware, possibly.

I’ll have to read it, but do you also argue that those three features are what sets Human Language apart from other animal communication systems?

That is just fabulous! I never thought about it that way, but it is worth doing so!

BTW, was showing your slide on the non-causal psych research yesterday in class! We discussed the multilingualism paper. I wanted to thank you for all the very interesting things you are doing!

Me neither until I realised I was one of them 🤣 (Part of my stance is anti-realist/idealist and anti-positivist so ultimately I believe there is no objective reality but then we straddle onto spiritualism which is another part of my stance). I guess it’s just more complicated, like most things! :)