Research

"Anuj designed the 12 prompt evaluation framework that is the basis of our SemEval Workshop work and also led the cross-lingual analysis explaining performance degradation across subtasks and languages. I'd highlight his instinct to look for reviewer concerns in particular, as he picked up on the majority of those concerns prior to submission, and the few that did come up were minor fixes."

"I have worked with Anuj on various projects but here I want to highlight our EACL workshop paper in particular. Initially we were trying to expand AfriXNLI, but ran into problems getting the data from the native language speakers. Anuj took the lead by turning this constraint into a more interesting research question by asking whether more data even helps in low-resource evaluation, and if yes, how much? He did the experiments and found that the clearest evidence for a language specific vs model driven scaling behavior is observed between Yoruba and Kinyarwanda language. He set the direction for the project and then together we worked on it, which made the paper possible. That kind of pivot from a dead end to a more useful question is what makes Anuj worth collaborating with."