Anna (Anya) Ivanova

Find Taha’s poster today at the @dataonbrainmind.bsky.social workshop #NeurIPS2025

I like this one proceedings.neurips.cc/paper_files/...

We leave the “how” to future work (by you and others and maybe us eventually)
I’ll keep ROSE in mind as a testable prediction of the ”how” claims

The main Q you deal with in your post - the how - is definitely an important one, but we didn’t engage with it much in this short piece; our focus is on the higher-level organizational principles. We did discuss it internally though, including the traveling waves idea.

…the claim in your post (lang areas as an executive processor) targets the same level of explanation as our proposal. And, as you say, it’s licensed by minimalism. So then it would make sense, to me, to compare this minimalist prediction with our proposal and find key differences (eg causal impact)

Haha a great use of Thanksgiving break 😂 thank you for engaging, but I still think my original question is answerable!

I agree, some areas of lang science make no predictions about the mechanisms of lang processing. And even those that do might not commit to a specific neural implementation. But…

Skyline of Madison, WI

🚨I am looking for a POSTDOC, LAB MANAGER/TECH and GRAD STUDENTS to join my new lab in beautiful Madison, WI.
We study how our brains perceive and represent the physical world around us using behavioral, computational, and neuroimaging methods.
paulunlab.psych.wisc.edu
#VisionScience #NeuroSkyence

Cool! What’s the prediction of the program about what would happen if the core language system was disabled? (I suspect this is where the predictions of the two theories would diverge)

What a privilege and a delight to work with @coltoncasto.bsky.social @ev_fedorenko and @neuranna
on this new speculative piece on What it means to understand language, nicely summarized in this
Tweeprint from @coltoncasto.bsky.social arxiv.org/abs/2511.19757

Neuroscience of language has a dilemma: how do we reconcile extensive patient and imaging evidence for **language-specific** processing with the fact that naturalistic language evokes extensive activity all over the brain? We propose a framework that accounts for both.

What enables human language? A biocultural framework Explaining the origins of language is a key challenge in understanding ourselves as a species. We present an empirical framework that draws on synergies across fields to facilitate robust studies of l...

Origins of language, one of humanity’s most distinctive traits, may be best explained as a unique convergence of multiple capacities each with its own evolutionary history, involving intertwined roles of biology & culture. This framing can expand research horizons. A 🧵 on our @science.org paper.🧪1/n

Genomic Investigations of Spoken and Written Language Abilities: A Guide to Advances in Approaches, Technologies, and Discovery Purpose: The aim of this tutorial is to show how the rise of molecular technologies and analytical methods in human genetics yields exciting new ...

Advances in genomics are giving exciting new perspectives on biology of speech, language & reading. My latest peer-reviewed paper is a tutorial, guiding readers from different backgrounds through the history of the field, current state-of-the-art, & where we’re heading. A taster in this thread.🧪
1/n

You get a few free runs if you register with an edu email. I tried once, it was ok - I think a truly successful approach would require some prompt iteration, so multiple runs. Also of course the result will very much depend on the problem at hand.

Top: A syntax tree for the sentence "the doctor by the lawyer saw the artist". Bottom: A continuous vector.

🤖🧠I'll be considering applications for PhD students & postdocs to start at Yale in Fall 2026!

If you are interested in the intersection of linguistics, cognitive science, & AI, I encourage you to apply!

PhD link: rtmccoy.com/prospective_...
Postdoc link: rtmccoy.com/prospective_...

Screenshot of a figure with two panels, labeled (a) and (b). The caption reads: "Figure 1: (a) Illustration of messages (left) and strings (right) in toy domain. Blue = grammatical strings. Red = ungrammatical strings. (b) Surprisal (negative log probability) assigned to toy strings by GPT-2."

New work to appear @ TACL!

Language models (LMs) are remarkably good at generating novel well-formed sentences, leading to claims that they have mastered grammar.

Yet they often assign higher probability to ungrammatical strings than to grammatical strings.

How can both things be true? 🧵👇

Delighted Sasha's (first year PhD!) work using mech interp to study complex syntax constructions won an Outstanding Paper Award at EMNLP!

Also delighted the ACL community continues to recognize unabashedly linguistic topics like filler-gaps... and the huge potential for LMs to inform such topics!

Excited to share our work on mechanisms of naturalistic audiovisual processing in the human brain 🧠🎬!!
www.biorxiv.org/content/10.1...

This document is scheduled to be published in the Federal Register on 10/30/2025 and available online at https://federalregister.gov/d/2025-19702, and on https://govinfo.gov DEPARTMENT OF HOMELAND SECURITY: Removal of the Automatic Extension of Employment Authorization Documents AGENCY: U.S. Citizenship and Immigration Services (USCIS), Department of Homeland Security (DHS). ACTION: Interim final rule (“IFR”) with request for comments. ______________________________________________________________________ SUMMARY: This IFR amends DHS regulations to end the practice of automatically extending the validity of employment authorization documents (Forms I-766 or EADs) for aliens who have timely filed an application to renew their EAD in certain employment authorization categories. The purpose of this change is to prioritize the proper vetting and screening of aliens before granting a new period of employment authorization and/or a new EAD. This IFR does not impact the validity of EADs that were automatically extended prior to [INSERT DATE OF PUBLICATION IN THE FEDERAL REGISTER] or which are otherwise automatically extended by law or Federal Register notice.

NEW: The Trump admin just ended the practice of automatically extending work permits when people file to renew them — meaning that if USCIS takes too long to process a renewal the applicant loses their authorization to work legally.

The interim final rule applies to renewals filed after tomorrow.

Much remains to be done on this front, ideas are welcome!

6/end

Finally, we wanted to ensure that what we're showing is LLM patterns not TunedLens patterns - see Appendix for the control analyses we do! (+a comparison with LogitLens) 5/

We then apply this approach to 3 case studies - prediction by part-of-speech, multi-token fact recall, and fixed-response question answering. Check the paper & Akshat's thread for details!

(look at this model predicting positive/negative sentiment - such a clear pattern!)

4/

Most early layer predictions get overturned! It appears that these are statistical guesses, made when the model has not yet processed enough contextual information.

(flip rate = how often the model's final prediction differs from the current layer's prediction)

3/

In early layers, most frequent tokens (e.g., "the") dominate predictions, whereas infrequent tokens gradually become more predicted later on.

2/

🤖📈 How do LLMs use their depth?

Akshat Gupta led a fun project to find out! We leverage TunedLens (~linear decoding of tokens) to explore how LLMs' internal representations change from layer to layer.

Preprint: arxiv.org/abs/2510.18871

1/

Perhaps! But language contains a lot of information about the physical world, so in principle these are learnable from language alone

Omg adorable! Thanks! :)

Finally out in TACL:
🌎EWoK (Elements of World Knowledge)🌎: A cognition-inspired framework for evaluating basic world knowledge in language models

tl;dr: LLMs learn basic social concepts way easier than physical&spatial concepts

Paper: direct.mit.edu/tacl/article...
Website: ewok-core.github.io

Dense Phenotyping of Human Brain Network Organization Using Precision fMRI The advent of noninvasive imaging methods like functional magnetic resonance imaging (fMRI) transformed cognitive neuroscience, providing insights into large-scale brain networks and their link to cog...

Why do brain networks vary? Do these differences shape behavior? If every 🧠 is unique, how can we detect common features of brain organization?
@rodbraga.bsky.social and I dig in, in @annualreviews.bsky.social (ahead of print):
go.illinois.edu/Gratton2025-...

#neuroskyence #psychscisky #MedSky
🧵👇

New paper in Imaging Neuroscience by Ammar I. Marvi, Nancy G. Kanwisher, et al:

An efficient multifunction fMRI localizer for high-level visual, auditory, and cognitive regions in humans

doi.org/10.1162/IMAG...

Our young lab greatly appreciates open datasets and code provided by pioneers in the field! @alexanderhuth.bsky.social
@amandalebel.bsky.social
@rjantonello.bsky.social
@mtoneva.bsky.social
@samnastase.bsky.social
@jixingli.bsky.social
@mschrimpf.bsky.social
@evfedorenko.bsky.social
etc
3/end