Our paper "The State and Fate of Multilingual, Contextual Evaluation in the NLP World" is accepted at #COLM2026!
We audited 51 multilingual benchmarks (242 datasets, 219 languages) to ask: is multilingual eval measuring what we think it is?
Dashboard: https://t.co/xHZP25BZxE
I'll be at #COLM2026 in SF and I'm looking for PhD positions (multilingual NLP, interpretable and trustworthy AI and AI safety)! If your lab is hiring, DM me or come say hi!
Paper link: https://t.co/88zuwZcWFS
Why do multilingual LLMs drop in performance on low-resource languages? Leaderboards give us per-language accuracies, but they just map the gaps; they don't explain why they happen.
Excited to share our latest preprint: DEPART [A thread]
https://t.co/FRM1V7nEWg
π’ Presenting our paper at #ACL2026!
"UPDESH: Synthesizing Grounded Instruction Tuning Data for 13 Indic Languages"
Please come find us at π Grand Hall β Session 16, Poster Session F ποΈ Jul 7 | 09:00β10:30
Paper: https://t.co/HJRig5C86V
Details on the work below π§΅
Why do multilingual LLMs drop in performance on low-resource languages? Leaderboards give us per-language accuracies, but they just map the gaps; they don't explain why they happen.
Excited to share our latest preprint: DEPART [A thread]
https://t.co/FRM1V7nEWg
DEPART shifts multilingual eval from passive measurement to an actionable diagnostic framework. It is a step towards making evaluations more explainable and insightful!