Sarah Wiegreffe Joins UMD’s CS Faculty

Sarah Wiegreffe will join the University of Maryland’s Department of Computer Science as an assistant professor in fall 2025, bringing with her a research focus on the transparency and interpretability of language models. Her work investigates how these increasingly prominent systems function, why they produce the outputs they do and how users interact with them.

Language models are becoming embedded in everyday technologies, from chatbots to search engines. As their influence grows, Wiegreffe’s research addresses concerns about false outputs, anthropomorphic behavior and unattributed reproduction of text; issues that affect how much users trust the information these models generate. 

Wiegreffe earned her Ph.D. in computer science from Georgia Institute of Technology in 2022. Her experience includes internships at Google and the Allen Institute for AI (Ai2), where she was later named an outstanding intern.

“There’s a real need to help users better understand how these models work,” Wiegreffe said. “My work aims to increase the likelihood of positive outcomes by studying how to explain model behavior in ways that support informed decision-making.”

Wiegreffe’s interest in machine learning and data science began during her undergraduate studies, where she was drawn to the intersection of mathematical structure and real-world application. That foundation eventually led her to natural language processing, a field she sees as both intellectually rich and socially impactful.

“Text captures so much of human experience,” she said. “We preserve nearly everything we know and encounter through written and spoken language, which makes natural language processing both fascinating and highly relevant.”

At Maryland, Wiegreffe said she was drawn to the department’s collaborative atmosphere and cross-disciplinary partnerships, particularly with faculty in machine learning, language sciences and the iSchool. She also noted the university’s emphasis on computational support as essential to her work.

“The department has a forward-thinking approach to ensuring researchers have access to the computational resources we need,” she said.

As generative AI continues to evolve, Wiegreffe plans to explore how these systems can be developed more efficiently. She pointed out that many current models rely on computationally intensive and data-inefficient methods. Her goal is to complement engineering advances with a scientific understanding of how language models function internally.

“Developing a science of language modeling can help us build better systems,” she said. “It also gives us the foundation to improve the way humans interact with those systems, especially in high-stakes environments.”

Wiegreffe sees her research as occupying the space between technical development and user experience. By clarifying how models behave and offering strategies to interpret their output, she hopes to support both AI developers and everyday users navigating the growing presence of these technologies.

—Story by Samuel Malede Zewdu, CS Communications