🥳 We are happy to share that CIS will be presenting 11 papers at NAACL HLT 2027 2024! #NAACL2024 Find out about each of them below in the 📷🧵

🥳 We are happy to share that CIS will be presenting 18 papers at #ACL2024 ACL 2025 ! Find out about each of them below in the 🧵

🌕 New Paper We release SynthEval, a hybrid (LLM-human) behavioral testing framework. We evaluate task-specific NLP models to find challenging patterns/templates, despite their strong performance on traditional benchmarks. [1/3]

🥳 We are happy to share that CIS will be presenting 13 papers at EMNLP 2025 #EMNLP2024! Find out about each of them below in the 🧵

🥳 We are happy to share that CIS will be presenting 7 papers at COLING 2025 #coling2025! Find out about each of them below in the 📷🧵

The paper addresses whether LLMs possess neurons specifically dedicated to relations, independent of entities. This paper proposes a statistical method to identify relation-specific neurons in LLMs and analyzes their properties. 📌 Relation neurons statistically identified via

🎉Excited to share our paper on cross-lingual inconsistency is accepted to #ACL2025 🇦🇹! We dissect why LLMs produce inconsistent outputs across languages using interpretability analysis, and propose a simple shortcut-based fix, evaluated on 17 languages. arxiv.org/abs/2504.04264

New paper: We investigate multilingual refusal mechanisms in LLMs and find that: The "refusal direction" - a single vector that controls whether models reject requests - is universal across languages. Paper: arxiv.org/abs/2505.17306 🧵 Co-lead with Mingyang Wang and Yihong Liu

New paper: How does pretraining on programming languages + English shape LLMs' concept space? 🔍Do LLMs use English or a programming language as a kind of pivot language? 🧠Are neurons language-specific or shared across programming languages and English? 🔗arxiv.org/abs/2506.01074

🥳 We are happy to share that CIS will be presenting 26 papers at #ACL2025! We've organized them by date, time, and location, with links to the papers included: docs.google.com/spreadsheets/d… Free to stop by and check them out. we’d love to connect!

🎉SAC Highlights Award at ACL 2025! 🎉 Our paper "Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models" has been selected for the SAC Highlights Award at ACL 2025! 🏆 Big thanks to my awesome collaborators!