Andrea de Varda

Human-like fleeting memory improves language learning but impairs reading time prediction in transformer language models Human memory is fleeting. As words are processed, the exact wordforms that make up incoming sentences are rapidly lost. Cognitive scientists have long believed that this limitation of memory may, para...

New preprint! w/@drhanjones.bsky.social

Adding human-like memory limitations to transformers improves language learning, but impairs reading time prediction

This supports ideas from cognitive science but complicates the link between architecture and behavioural prediction
arxiv.org/abs/2508.05803

Can't wait for #CCN2025! Drop by to say hi to me / collaborators!

Evidence from Formal Logical Reasoning Reveals that the Language of Thought is not Natural Language Humans are endowed with a powerful capacity for both inductive and deductive logical thought: we easily form generalizations based on a few examples and draw conclusions from known premises. Humans al...

Is the Language of Thought == Language? A Thread 🧵
New Preprint (link: tinyurl.com/LangLOT) with @alexanderfung.bsky.social, Paris Jaggers, Jason Chen, Josh Rule, Yael Benn, @joshtenenbaum.bsky.social, ‪@spiantado.bsky.social‬, Rosemary Varley, @evfedorenko.bsky.social
1/8

The BLiMP-NL dataset consists of 84 Dutch minimal pair paradigms covering 22 syntactic phenomena, and comes with graded human acceptability ratings & self-paced reading times. An example minimal pair: A. Ik bekijk de foto van mezelf in de kamer (I watch the photograph of myself in the room; grammatical) B. Wij bekijken de foto van mezelf in de kamer (We watch the photograph of myself in the room; ungrammatical) Differences in human acceptability ratings between sentences correlate with differences in model syntactic log-odds ratio scores.

Next week I’ll be in Vienna for my first *ACL conference! 🇦🇹✨

I will present our new BLiMP-NL dataset for evaluating language models on Dutch syntactic minimal pairs and human acceptability judgments ⬇️

🗓️ Tuesday, July 29th, 16:00-17:30, Hall X4 / X5 (Austria Center Vienna)

I'm sharing a Colab notebook on using large language models for cognitive science! GitHub repo: github.com/MarcoCiappar...

It's geared toward psychologists & linguists and covers extracting embeddings, predictability measures, comparing models across languages & modalities (vision). see examples 🧵

Cracking arbitrariness: A data-driven study of auditory iconicity in spoken English - Psychonomic Bulletin & Review Auditory iconic words display a phonological profile that imitates their referents’ sounds. Traditionally, those words are thought to constitute a minor portion of the auditory lexicon. In this articl...

📢 New paper out! We show that auditory iconicity is not marginal in English: word sounds often resemble real-world sounds. Using neural networks and sound similarity measures, we crack the myth of arbitrariness.
Read more: link.springer.com/article/10.3...

@andreadevarda.bsky.social

Many LM applications may be formulated as text generation conditional on some (Boolean) constraint.

Generate a…
- Python program that passes a test suite.
- PDDL plan that satisfies a goal.
- CoT trajectory that yields a positive reward.
The list goes on…

How can we efficiently satisfy these? 🧵👇

The cerebellar components of the human language network The cerebellum's capacity for neural computation is arguably unmatched. Yet despite evidence of cerebellar contributions to cognition, including language, its precise role remains debated. Here, we sy...

New paper! 🧠 **The cerebellar components of the human language network**

with: @hsmall.bsky.social @moshepoliak.bsky.social @gretatuckute.bsky.social @benlipkin.bsky.social @awolna.bsky.social @aniladmello.bsky.social and @evfedorenko.bsky.social

www.biorxiv.org/content/10.1...

1/n 🧵

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

The extended language network: Language selective brain areas whose contributions to language remain to be discovered Although language neuroscience has largely focused on core left frontal and temporal brain areas and their right-hemisphere homotopes, numerous other areas - cortical, subcortical, and cerebellar - ha...

Excited to share new work on the language system!

Using a large fMRI dataset (n=772) we comprehensively search for language-selective regions across the brain. w/
Aaron Wright, @benlipkin.bsky.social, and @evfedorenko.bsky.social

Link to the preprint: biorxiv.org/content/10.1...
Thread below!👇🧵

A language network in the individualized functional connectomes of over 1,000 human brains doing arbitrary tasks A century and a half of neuroscience has yielded many divergent theories of the neurobiology of language. Two factors that likely contribute to this situation include (a) conceptual disagreement…

New brain/language study w/ @evfedorenko.bsky.social! We applied task-agnostic individualized functional connectomics (iFC) to the entire history of fMRI scanning in the Fedorenko lab, parcellating nearly 1200 brains into networks based on activity fluctuations alone. doi.org/10.1101/2025... . 🧵

Conceptual Combination in Large Language Models: Uncovering Implicit Relational Interpretations in Compound Words With Contextualized Word Embeddings Large language models (LLMs) have been proposed as candidate models of human semantics, and as such, they must be able to account for conceptual combination. This work explores the ability of two LLM...

1/n Happy to share a new paper with Calogero Zarbo & Marco Marelli! How well do LLMs represent the implicit meaning of familiar and novel compounds? How do they compare with simpler distributional semantics models (DSMs; i.e., word embeddings)?
doi.org/10.1111/cogs...

To the brain, Esperanto and Klingon appear the same as English or Mandarin MIT research finds the brain’s language-processing network also responds to artificial languages such as Esperanto and languages made for TV, such as Klingon on “Star Trek” and High Valyrian and Dothr...

So excited to have our work on conlangs out in PNAS: www.pnas.org/doi/10.1073/... Congrats, Saima, Maya, and the rest of the crew -- well done!
Here is the MIT news story:
news.mit.edu/2025/esperan...

New preprint w/ @jennhu.bsky.social @kmahowald.bsky.social : Can LLMs introspect about their knowledge of language?
Across models and domains, we did not find evidence that LLMs have privileged access to their own predictions. 🧵(1/8)

In conclusion, our results show that (1) LMs are broadly applicable models of the human language system across languages, and (2) there is a shared component in the processing of different languages (14/14)

Languages greatly vary in form, but there is a massive overlap in the concepts they can express. We speculate that this shared meaning space is responsible for successful encoding transfer, but we’ll look more into this in future work (13/)

What supports the transfer of the encoding models? Form and meaning are two promising candidates. However, form-based (phonological, phonetic, syntactic) language similarity does not predict transfer performance (12/)

Encoding models trained on existing fMRI datasets successfully predicted responses in new languages, generalizing across stimuli types and modalities (11/)

In Study II, we tested transfer in a more stringent condition: training encoding models on existing fMRI datasets (sentence reading: Pereira2018, Tuckute2024; passage listening: Study I data, NatStories) and testing them on newly collected fMRI data in 9 new languages (10/)

In the “across” condition, performance improves for models with stronger cross-lingual semantic alignment (where translations cluster together in the embedding space) (9/)

But what kind of model properties influence LM-to-brain alignment across languages?

In the “within” condition, encoding performance is highest for models with good next-word prediction abilities (8/)

We also replicated in a cross-lingual setting the finding that the best fit to brain responses is obtained in intermediate-to-deep layers (for each subplot pair, the left one is “within”, the right one “across”) (7/)

We evaluated 20 multilingual LMs with different architectures and training objectives, and all of them were able to predict brain responses in the various languages (“within”) and critically, generalized zero-shot to unseen languages (“across”) (6/)

Critically, we fit two kinds of encoding models:
1️⃣ “within” encoding models, training and testing on data from a single language with cross-validation
2️⃣ “across” encoding models, training in N-1 languages and testing in the left-out language (5/)

In Study I, we:
1️⃣ Present participants with auditory passages and record their brain responses in the language network
2️⃣ Extract contextualized word embeddings from multilingual LMs
3️⃣ Fit encoding models predicting brain activity from the embeddings (4/)

We address these questions through two studies, combining existing (12 languages, 24 participants) and newly collected fMRI data (9 languages, 27 participants). (3/)

We ask two core questions:
1️⃣ Does the LM-brain alignment generalize to typologically diverse languages?
2️⃣ Are brain representations similar across languages? (2/)

Multilingual Computational Models Reveal Shared Brain Responses to 21 Languages At the heart of language neuroscience lies a fundamental question: How does the human brain process the rich variety of languages? Recent developments in Natural Language Processing, particularly in m...

New preprint! 🧠🤖

Brain encoding in 21 languages!

www.biorxiv.org/content/10.1...

w/ Saima Malik-Moraleda, @gretatuckute.bsky.social , and @evfedorenko.bsky.social (1/)