Training on a little 🤏 formal language BEFORE natural language can make pretraining more efficient! How and why does this work? The answer lies…Between Circuits and Chomsky. 🧵1/6👇

CDS is hiring a Clinical Professor of Data Science. Teach ML, programming, and specialized courses in our 60 5th Ave building. Renewable contracts with promotion opportunities. Apply by April 1, 2025. For details, see: apply.interfolio.com/155349 #MachineLearning #ML #AIjobs

We are looking for emergency reviewers for the CMCL workshop! Send me a message if you would like to volunteer! CMCL workshop NAACL HLT 2025

Abstract submissions for the 1st Computational Psycholinguistics Meeting is now open! cpl2025.sites.uu.nl

New brain/language study w/ Ev (like in 'evidence', not Eve) Fedorenko 🇺🇦! We applied task-agnostic individualized functional connectomics (iFC) to the entire history of fMRI in the Fedorenko lab, parcellating nearly 1200 brains into networks based on activity fluctuations alone. doi.org/10.1101/2025.0… . 🧵

✨New paper ✨ Introducing 🌍MultiBLiMP 1.0: A Massively Multilingual Benchmark of Minimal Pairs for Subject-Verb Agreement, covering 101 languages! We present over 125,000 minimal pairs and evaluate 17 LLMs, finding that support is still lacking for many languages. 🧵⬇️

Now I humbly request your participation May 3 in Santa Ana NAACL HLT 2025; Tatsuki Kuribayashi and I will be there

TFW when you've been implicitly predicting the future 😎

Two papers have been accepted to ACL 2025 (one as first author). – Developmentally-plausible Working Memory Shapes… arxiv.org/abs/2502.04795 – If Attention Serves as a Cognitive Model of Human Memory Retrieval… arxiv.org/abs/2502.11469 #ACL2025 #ACL2025NLP

A string may get 17 times less probability if tokenised as two symbols (e.g., ⟨he, llo⟩) than as one (e.g., ⟨hello⟩)—by an LM trained from scratch in each situation! Our #acl2025nlp paper proposes an observational method to estimate this causal effect! Longer thread soon!