The University of Massachusetts Amherst

A view of two hands, one of a woman and one of a robot, over a keyboard. Credit: Getty Images
Research

Linguistics’ Brian Dillon Receives NSF Grant to Explore AI and Human Language Processing

Image
Brian Dillon
Brian Dillon

Brian Dillon, professor of linguistics in the College of Humanities and Fine Arts, has been awarded a four-year, $432,656 research grant from the National Science Foundation to investigate how artificial intelligence (AI) systems and humans differ in the way they process language.

The project, titled “Collaborative Research: Semantic Focusing: Controlling LM Interpretations for Human-Model Alignment,” is set to begin in fall 2025. Dillon will lead the work at UMass Amherst in collaboration with Tal Linzen, associate professor of linguistics and data science at New York University.

To date, the collaborative project with NYU has received approximately $1.2 million in combined total funding, including the recent award of $432,656 to UMass Amherst.

“This research aims to better understand some of the most important ways in which AI systems and humans differ in how they process language, and the way AI systems encode aspects of the meaning conveyed by simple language,” Dillion said.

“Ultimately, with this research, we’re hoping to bring the models’ language processing more in line with how we think humans process language — that is, attempting to identify the best interpretation of linguistic input as quickly as possible.”

Dillon, a psycholinguist who directs the Computational Sentence Processing Lab in the Integrative Learning Center, studies the interaction between grammar and working memory in language comprehension. His research combines psycholinguistic and computational approaches to investigate sentence processing in adults.

According to Dillon, prediction plays a key role in both AI models and human language processing. While many AI systems learn by predicting the next word in a sentence, research suggests that linguistic structure also strongly influences human reading expectations.

“We’ve shown that readers’ difficulty with unexpected words is often tied to the grammatical structure of the text, not just word probability,” Dillion said. “This supports long-standing psycholinguistic theories and opens up new possibilities for improving AI language models.”

The collaboration with Linzen, who is also a research scientist at Google, is expected to contribute to research that bridges psycholinguistics and language technology. Dillion hopes the findings will contribute to more data-efficient and human-like AI systems, while also understanding how people process and learn language. 

More information on the NSF grant project and Dillion’s research can be found on the College of Humanities and Fine Arts news website.