I am an Assistant Professor at The Ohio State University.

My research interests broadly include topics in Machine Learning for Natural Language Processing (NLP). In particular, I am interested in building language technologies that work for all people—language models that can uniformly support diverse languages, domains, populations, and individuals. Some research directions I am currently excited about are:

• Multilingual models that can equitably support written and spoken languages.

• Continual adaptation of NLP models to support new languages, varieties, domains, and capabilities.

• Personalization of language technologies to diverse user preferences.

• Evaluation of language models for scenarios that real users care about (especially interested in uses for experts like scientists and clinicians)

These directions touch various parts of a typical machine learning pipeline, including building new datasets, modeling paradigms and architectures, training and inference algorithms, and evaluation methodologies.

I was a postdoctoral researcher at the Allen Institute for AI (Ai2) advised by Noah Smith and Hannaneh Hajishirzi from August 2023 to 2024. I obtained my Ph.D. at the Language Technologies Institute at Carnegie Mellon University (CMU) in 2023 advised by Yulia Tsvetkov, with the final two years of my PhD spent visiting the University of Washington in Seattle.

News

• October 2026 — Presented a tutorial on tokenization in LLMs at EMNLP 2026. Attended by ~150 people!
• October 2026 — Our paper on accent bias in audio LMs won outstanding paper award 🏆 at EMNLP 2026.
• September 2026 — Our tokenization survey paper is out!
• September 2026 — HalluWorld is accepted to NeurIPS 2026 D&B Track.
• September 2026 — Our EMNLP 2026 paper on instruction hierarchy resolution in multimodal LLMs is now on arXiv.
• August 2026 — Our EMNLP 2026 paper on pretraining LMs with phonetic representations is now on arXiv.
• August 2026 — Three papers accepted to EMNLP 2026 (Main)! More details soon.
• August 2026 — New preprint on multi-byte prediction in latent tokenizer based language models.