I'm Nick, a fifth-year Computer Science PhD candidate in Columbia University's Natural Language Text Processing Lab and a member of the Columbia Core AI Lab (CAIL). My research interests broadly lie at the intersection of Natural Language Processing, linguistics, and the social sciences working toward more nuanced and improved understanding of attitudes in LLMs including the relationship between attitude expression and identity. I am grateful to be advised by Professor Kathleen McKeown, and I am generously supported by the NSF Graduate Research Fellowship, the Columbia Presidential Fellowship, and the Provost Diversity Fellowship.
Much of the text found online used for pre-training language models reflects particular forms of English, namely White Mainstream English, leading to models that are less capable of grappling with other language varieties of English, such as African American Language, or varieties of other languages. My work in this direction focuses on evaluating and mitigating linguistic biases to help create models capable of broadly understanding language across identities.
Modeling Social Psychology and Attitudes
As language models continue to be applied in psychological research and mental health settings, they must be capable of interpreting nuanced expressions of emotion or mental health. My work in this direction draws on theories and findings in social psychology to imbue models with better understanding of human expressions of attitudes and interpersonal behavior through language.
Modeling Political Perspectives and Polarization
With increased interaction on social media and a diverse array of media sources, it is important to ensure that language models fairly learn from and aid constructive public discourse. This research direction aims to develop systems capable of faithfully reasoning about group perspectives, fairly represent perspectives across groups, and develop approaches to better understand nuances of political speech and rhetoric, particularly online.
Recently, I am also increasingly interested in applying mechanistic interpretability approaches to study the above concepts in language models, studying pretraining data, and multi/cross-lingual settings.
Contact & Collaboration
If you are an undergraduate or master's student at Columbia interested in working with me or our lab, I typically look for students through the research fair at the beginning of each semester.
I would also suggest taking Professor McKeown's language generation and summarization seminar this Fall, which involves research-based projects advised by her and students in our lab.
If you are otherwise interested in talking about research or potential collaborations, please reach out to me over email!
Jeffrey A. Fine, D. Hudson Smith, Cierra Oliveira, Nicholas Deas, Spencer Shellnutt, Riley Stotzky, Rachel Clyburn (Journal of Information Technology & Politics)
Nicholas Deas, Robin Kowalski, Sophie Finnell, Emily Radovic, Hailey Carroll, Chelsea Robbins, Andrew Cook, Kenzie Hurley, Natalie Cote, Kelly Evans, Isabella Lorenzo, Kelly Kiser, Gabriela Mochizuki, Meredith Mock, and Lyndsey Brewer (Computers in Human Behavior 2023)
