Mai Mohamed Eida
Computational Linguistics PhD candidate @ UIUC
I am currently on the job market for positions starting January 2027!
I am a computational linguist and researcher at University of Illinois Urbana Champaign advised by Dr. Jonathan Dunn. I am interested in how large language models (LLMs) represent and process linguistic variation, especially in low-resource languages, dialects, and speech communities.
My research focus is computational sociolinguistics. I work on investigating how linguistic variation is represented in language technologies, specifically training data (upstream), and how these representations influence model behavior and performance (downstream). I also explore how existing evaluation practices and benchmarks account for linguistic diversity, and what this means for our generalizations on LLM performance and capabilities. I am particularly interested in (1) how linguistic diversity, or lack thereof, in training data influence the representations learned by language models, (2) how these representations affect downstream model performance and language generation, and (3) how we can develop more representative resources and evaluation methods for underrepresented languages and dialects as efficiently as possible. I am interested in how LLMs encode language and how these representations are reflected not only in LLM language production, but also in their broader impact on the speech communities they are intended to serve overall!
Forthcoming
-
Driven Dialectology and Regional Variation
Jonathan Dunn,
Research Handbook on Sociolinguistics. Edward Elgar Publishing. Forthcoming.
2026
2025
2024
2026
-
I'm looking forward to sharing our work at ArabicNLP this October 2026! I will be presenting our paper, “Some Dialects are More Equal than Others: Egyptian Arabic Non-Prestigious Dialectal Bias in LLMs". Stop by and say hello if you would like to talk about dialectal bias in LLMs! You can find the paper on arXiv and the conference program here: Conference program.
-
How do LLMs use quantifiers? I’ll be presenting our poster, “From Tables to Quantified Statements: LLM Inference Generation through Executable Verification,” at INLG 2026 in Utrecht, the Netherlands. Stop by to find out and say hello! You can read the paper on arXiv or take a look at the poster.