Dipankar Srirag

I'm a researcher currently working on dialect-aware language models and their evaluation. My research focuses on making language models more equitable and robust across different varieties of English, particularly Australian, British and Indian English. My recent work introduces methods for evaluating dialect robustness of LLMs through conversation understanding tasks and proposes low-rank adapters for improving performance on dialectal variations.

My research interests span dialect adaptation, parameter-efficient fine-tuning, and evaluation of language models. I'm particularly passionate about creating more inclusive NLP systems that work well for speakers of different varieties of English. Some of my recent projects include building sentiment classification benchmarks for varieties of English, developing dialect adaptations for decoder models, and proposing evaluation frameworks to measure dialect robustness of language models.