This talk is part of the NLP Seminar Series.

Measuring Depictions and Expressions of Social Groups with NLP

Lucy Li, University of California, Berkeley
Date: 11:00am - 12:00pm, May 11th 2023
Venue: Room 287, Gates Computer Science Building; Zoom (link hidden)


This talk covers my work at the intersection of NLP and computational social science across three domains: machine-generated narratives, online discussion forums, and scientific literature. In the first, I show that GPT-3’s stories replicate gendered narrative patterns of books, and discuss my attempts to nudge it away from stereotypes baked into its default behavior. Second, I'll present a new contextualized word embedding approach for characterizing the depiction of people, and apply this method on a longitudinal, multi-platform case study of online discussions around women. Finally, I'll present an analysis of specialized language across hundreds of scientific disciplines. I find that though jargon is nearly always negatively associated with interdisciplinary impact, fields have varying norms around how much they adjust their vocabularies for broader audiences.


Lucy is a PhD student at UC Berkeley’s School of Information and Berkeley AI Research, supported by an NSF Graduate Research Fellowship. During her PhD, she has interned at Microsoft Research and the Allen Institute for AI, and during the latter, she received an Outstanding Intern of the Year award. Aside from prompt engineering, her hobbies include running out of memory, digitizing datasets by hand, and hosting a podcast during a pandemic. She has a B.S. in Symbolic Systems and M.S. in Computer Science from Stanford.