Kyle Mahowald

Happy to share the published version of "Art, Understanding, and Mystery"! I often hear some version of the thought that it's bad to understand artworks; this paper attempts to make that claim precise and show one way to defend artistic understanding! journals.publishing.umich.edu/ergo/article...

Is there any work on conditionals like '[Larry David is making a new show with the Obamas.] If Susie Essman is involved, hooray!'? It seems really hard to make sense of it on most of the approaches to conditionals/expressives that I'm aware of, but I'm curious if people know stuff or have thoughts.

Check it out if you're at SCiL! Fun tool for easily getting semantic projections from word embeddings into interpretable space.

Experimentology cover: title and curves for distributions.

Experimentology is out today!!! A group of us wrote a free online textbook for experimental methods, available at experimentology.io - the idea was to integrate open science into all aspects of the experimental workflow from planning to design, analysis, and writing.

Some attitudes we usually do not have I present a new attitude puzzle involving disjunction. Specifically, though it can sound strange to ascribe the belief that ϕ$\phi$ or ψ$\psi$ when ⌜ϕ⌝$\ulcorner \phi \urcorner$ and ⌜ψ⌝$\ulcorner \ps...

My paper just came out in PPR today, here: onlinelibrary.wiley.com/doi/10.1111/.... It's about interesting relationships that our attitudes bear to the world. I argue that belief is very different from other attitudes, and this difference follows from its relationship to the truth of token contents.

So excited to welcome @kanishka.bsky.social (back) to UT!

Picture of the UT Tower taken by me on my first day at UT as a postdoc in 2023!

News🗞️

I will return to UT Austin as an Assistant Professor of Linguistics this fall, and join its vibrant community of Computational Linguists, NLPers, and Cognitive Scientists!🤘

Excited to develop ideas about linguistic and conceptual generalization (recruitment details soon!)

No. We are opting for simplicity. It doesn't help to pile targeted interventions, even if each makes a lot of sense on its own. The key to our process is the ACs, so we just want to help them

Causal Interventions Reveal Shared Structure Across English Filler-Gap Constructions Large Language Models (LLMs) have emerged as powerful sources of evidence for linguists seeking to develop theories of syntax. In this paper, we argue that causal interpretability methods, applied to ...

We believe this work shows how mechanistic analyses can provide novel insights into syntactic structures — making good on the promise that studying LLMs can help us better understand linguistics by helping us develop linguistically interesting hypotheses!

📄: arxiv.org/abs/2505.16002

Do the causal mechanism underlying filler-gap processing in models transfer across constructions? We find yes … but with some wrinkles. Check out the paper_t that Sasha wrote ___t!

A key hypothesis in the history of linguistics is that different constructions share underlying structure. We take advantage of recent advances in mechanistic interpretability to test this hypothesis in Language Models.

New work with @kmahowald.bsky.social and @cgpotts.bsky.social!

🧵👇!

Come by to our panel at APS to share your thoughts, and ask us all the hard stuff!

At APS in DC, thrilled to be moderating a panel on statistics and abstraction in cognition with 3 stars: Jay McClelland, Tom Griffiths @cocoscilab.bsky.social, and @adinawilliams.bsky.social at 1pm tmrw. Bayes. Neural nets. Linguistics…we’ll have it all! www.psychologicalscience.org/conventions/...

Robbie Kubala, Harvey Lederman & Adam Lovett, On the Value of Irreplaceable Objects - PhilPapers Bradford (2023) calls attention to the fact that the strength of our reasons to preserve distinctively valuable objects increases as the number of such objects decreases. Bradford develops an account ...

@rkubala.bsky.social, @adamlovett.bsky.social, and I have a new paper forthcoming in the Journal of Philosophy. It’s called “On the value of irreplaceable objects” – Here’s a thread!

philpapers.org/rec/KUBOTV 1/n

Delighted to have Elias joining the UT NLP community!

What do "Maui, Sicily, Thailand" have in common? Ok, "places". But I say "White Lotus locales": it would be quite a coincidence if I hit on all 3 by chance! We ask how LMs do at this kind of inference.

Also fun to do a study on "the number game", the first Bayesian cogsci I learned in grad school!

I might be able to hire a postdoc for this fall in computational linguistics at UT Austin. Topics in the general LLM + cognitive space (particularly reasoning, chain of thought, LLMs + code) and LLM + linguistic space. If this could be of interest, feel free to get in touch!

Writing my first post here to announce that I've accepted an assistant professor job at TTIC! I'll be starting in Fall 2026, and recruiting students this upcoming cycle.

Until then, I'll be wrapping up the PhD at Berkeley, and this summer I'll join NYU as a CDS Faculty Fellow 🏙️

PINEAPPLE, LIGHT, HAPPY, AVALANCHE, BURDEN

Some of these words are consistently remembered better than others. Why is that?
In our paper, just published in J. Exp. Psychol., we provide a simple Bayesian account and show that it explains >80% of variance in word memorability: tinyurl.com/yf3md5aj

@kmahowald.bsky.social with a beautiful high-tech illustration 🎨 while describing @qyao.bsky.social's latest paper at the HSP online seminar series!

Paper: arxiv.org/abs/2503.20850

Will be talking about this work (and more) at 2 ET/11 PT in the HSP talk series on Computational Language Models and Psycholinguistics! www.hspsociety.org

If you give a mouse a cookie....does an LM learn something different than if you "give a cookie to a mouse"? Or if you don't give anyone anything? Or if you do other weird stuff to the input? New paper on manipulating ling input and training small LMs to study direct vs indirect evidence.

examples from direct and prepositional object datives with short-first and long-first word orders: DO (long first): She gave the boy who signed up for class and was excited it. PO (short first): She gave it to the boy who signed up for class and was excited. DO (short first): She gave him the book that everyone was excited to read. PO (long-first): She gave the book that everyone was excited to read to him.

LMs learn argument-based preferences for dative constructions (preferring recipient first when it’s shorter), consistent with humans. Is this from memorizing preferences in training? New paper w/ @kanishka.bsky.social , @weissweiler.bsky.social , @kmahowald.bsky.social

arxiv.org/abs/2503.20850

Cool mix where dative alternation preferences in models come from: effect of global preference for "short before long" in English (which we test by scrambling sentences using dep parses). I'm excited about this paradigm, and 1st year UT ling student @qyao.bsky.social did amazing work on this!

If you give a mouse a cookie....does an LM learn something different than if you "give a cookie to a mouse"? Or if you don't give anyone anything? Or if you do other weird stuff to the input? New paper on manipulating ling input and training small LMs to study direct vs indirect evidence.

Check out our new work on introspection in LLMs! 🔍

TL;DR we find no evidence that LLMs have privileged access to their own knowledge.

Beyond the study of LLM introspection, our findings inform an ongoing debate in linguistics research: prompting (eg grammaticality judgments) =/= prob measurement!

Language Models Fail to Introspect About Their Knowledge of Language There has been recent interest in whether large language models (LLMs) can introspect about their own internal states. Such abilities would make LLMs more interpretable, and also validate the use of s...

Full Paper: arxiv.org/abs/2503.07513
Again, thanks to my amazing advisors @jennhu.bsky.social and @kmahowald.bsky.social for their guidance and support! (8/8)

And keep your eye out for @siyuansong.bsky.social, a star UT undergrad who I suspect you will see more from!

If I ask model A “is this sentence grammatical” and it says yes, does that mean model A is more likely to produce that sentence than model B? Check out our new paper on whether models introspect about knowledge of language.

I'm excited to announce two papers of ours which will be presented this summer at @naaclmeeting.bsky.social eting.bsky.social and @iclr-conf.bsky.social !
🧵