Pronunciation: /ʌnˈdʒɪʃnuː/ (un-JISH-noo)
I am currently on the 2027 job market for postdoctoral and tenure-track positions.
I am a fifth-year Ph.D. candidate in Computer Science at George Mason University, advised by Antonis Anastasopoulos and Ziwei Zhu in the GMU NLP Lab.
Research directions
AI increasingly mediates how people access information and understand one another across linguistic and cultural boundaries. My research asks how we can build AI systems that work equitably beyond dominant languages and cultures. I pursue this goal through three connected directions:
Selected peer-reviewed and under-review research. Full list on Google Scholar
, Ziwei Zhu, Antonios Anastasopoulos.
We show that bilingual pretraining changes deeper English representations across eight language pairs, even when the models’ word embeddings align.
, Chutong Meng, Antonios Anastasopoulos.
We expose gaps between the languages LLMs claim to support and their actual behavior, arguing for multilingualism as a deliberate design goal.
, Chahat Raj*, Ziwei Zhu, Antonios Anastasopoulos.
We expand social-bias evaluation to 24 languages with culturally grounded data and examine regional biases across six Indian languages.
, Ziwei Zhu, Antonios Anastasopoulos.
We introduce MAPLE and LocalNewsQA to show how geographic metadata during pretraining helps language models choose answers that fit a question’s locale.
, Ziwei Zhu, Antonios Anastasopoulos.
We introduce DalleStreet and CultureAdapt to study cultural associations in multimodal models and use extracted artifacts to adapt images to local cultural contexts.
, Aylin Caliskan, Ziwei Zhu, Antonios Anastasopoulos.
We study how language-specific instruction tuning and bilingual pretraining affect cultural knowledge across six languages, revealing uneven gains and persistent cultural biases.
Jinhao Pan, Chahat Raj, , Sina Mansouri, Bowen Wei, Shloka Yada, Ziwei Zhu.
We introduce KnowBias, which reduces social bias by strengthening bias-knowledge neurons using a few yes/no questions, without retraining and with minimal loss of general capability.
Mamnuya Rinki, Chahat Raj, , Ziwei Zhu.
We measure culturally specific, intersectional biases across ten South Asian languages and test whether prompting reduces them in open-ended generation.
Chahat Raj, , Aylin Caliskan, Antonios Anastasopoulos, Ziwei Zhu.
We uncover hidden social associations across nine bias dimensions in vision-language models, revealing subtle and extreme biases that familiar stereotype-based evaluations can miss.
Chahat Raj*, , Aylin Caliskan, Antonios Anastasopoulos, Ziwei Zhu.
We adapt the contact hypothesis from social psychology to reduce language-model bias through simulated intergroup interactions and instruction tuning.
Syosset High School
Multilingual and multi-attribute cultural steering for LLMs
Thomas Jefferson (TJHSST)
Accent and dialect disparities in voice assistants, focusing on South and Southeast Asian English
IIT Roorkee Qualcomm
IIT Roorkee InMobi
IIT Roorkee Glance
Cultural adaptation for Indian food images (B.Tech thesis) · project
GMU IntraFi
South Asian bias in multilingual, open-ended LLM generations (M.S. thesis) · paper
GMU SyncoreLabs
Auditing cultural representation in text-to-image models (AI, Power & Society course project) · report
GMU
GMU Aireon
Intersectional bias across Indo-Aryan languages (Advanced NLP course project) · slides
GMU DoorDash
Assessing cuisine biases and detecting hallucinations in chart understanding in LLMs
Teaching Assistant · George Mason University
Spring 2023
Fall 2022
Teaching Assistant · George Mason University
