"NLP was a way to reconcile my technical background with my interest in languages"

by Bernice Chan

Photo of Mozhdeh Gheini
Photograph by Angel Itua

Mozhdeh Gheini received her Ph.D. in computer science in January 2025, she was supervised by Jonathan May. She was born and raised in Tehran, Iran. She did her undergraduate degree in computer software engineering in Iran and then came to the United States for graduate school. She joined ISI during the second year of her Master's degree, then began her Ph.D. here.

Why did you choose a computer science field?

We had a programming course back in high school and I really enjoyed it. I became interested in artificial intelligence and natural language processing (NLP). My first course during my Master's degree was an NLP course.

What were your primary research focuses at ISI?

I mainly focused on NLP and efficient transfer learning methods, which is about adapting large language models to do specific tasks, and asking high-level questions like: 'Can we do it more smoothly with less data?' and 'How should we change the processes?' I've always liked languages and NLP was one way for me to reconcile my technical background with my interest.

What's the most impactful project you've worked on at ISI?

My first official paper in my Ph.D. career was published in the 2021 Conference on Empirical Methods in Natural Language Processing (EMNLP). I was interested in transfer learning, which is about adapting an existing translation model to be able to translate other languages as well. I was curious if we needed to change all existing parameters while going from one language to another. My paper demonstrates that you can adapt models by only changing one part of the architecture called "cross-attention," which is the parameter that checks the source material during translation. The project showed that we don't need a lot of computational resources for translation, making it more accessible.

What's the biggest takeaway from your time at ISI?

Research is a patience game. It's like a domino effect: you can't reach the 100th domino without going through the first 99. You have to give everything the amount of time that it needs. Also, people often underestimate how far a simple hello can go. I had many discussions with people in the ISI corridors, which led to unexpected collaborations. The ISI community is strong and I will always feel like I am part of this family.

What are your plans after ISI?

I want to start a research-oriented position in industry, preferably allowing me to continue doing NLP research. I want to experience how things are outside of academia, and observe how the nature of problems change.

Want to write about this story?