Data Science/Computational Social Science Seminar: Philip Resnik
Using LLMs to Analyze What People Mean, Not Just What They Say
Abstract
In computational social science research, we often look at what people say when what we're really interested in is what they think. In linguistics we distinguish between the conventional meaning (semantics) of an utterance and its interpretation (contextualized meaning), which takes into account presuppositions and other unstated aspects of the thought behind the content. Similarly in political communication we talk about framing, distinguishing between what is said and the implications behind how it is said. In this talk I’ll discuss recent research that bridges this gap by using large language models to make plausible inferences about the unstated content behind what people say—identifying, for example, that when someone says “Build the wall”, part of their contextualized meaning is “Illegal immigration over the southern border is a problem”. Surfacing this kind of latent content as natural language, rather than some other representation, makes it possible to apply the full arsenal of natural language processing techniques to what people mean rather than just what they say. We have found in several studies that this approach is valuable in making sense of public opinion, analyzing legislator voting behavior, and in going beyond the pro/con/neutral of conventional stance detection to extract both implicit and explicit attitudes and place them on a continuous scale, enabling more effective analysis of issue attitudes in online communities.
Paper:
Alexander Hoyle, Rupak Sarkar, Pranav Goel, and Philip Resnik. 2023. Natural Language Decompositions of Implicit Content Enable Better Text Representations. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 13188–13214, Singapore. Association for Computational Linguistics.
Speaker bio
Philip Resnik is an MPower Professor at University of Maryland with joint appointments in the Department of Linguistics and the Institute for Advanced Computer Studies. He earned his bachelor’s in computer science at Harvard and his PhD in computer and information science at the University of Pennsylvania, and does research in computational linguistics. Prior to joining UMD, he was an associate scientist at BBN, a graduate summer intern at IBM T.J. Watson Research Center (subsequently awarded an IBM Graduate Fellowship) while at UPenn, and a research scientist at Sun Microsystems Laboratories. In 2020 he was designated a Fellow of the Association for Computational Linguistics. Philip's most recent research has focused in two areas. One is the computational cognitive neuroscience of language, where he has been using computational modeling in connection with brain imaging to look at the role of context and predictive processing during online language comprehension. The other is computational social science, with an emphasis on connecting the signal available in people's language use with underlying mental state—this has applications in computational political science, particularly in connection with ideology, framing and beliefs, and in mental health, focusing on the ways that linguistic behavior may help to identify and monitor depression, schizophrenia and suicidality. Philip is a scientific advisor for NORC at the University of Chicago (a non-partisan, independent social research organization). In entrepreneurial life he was a technical co-founder of CodeRyte (NLP for electronic health records, acquired by 3M in 2012), and is an advisor to FiscalNote (machine learning and analytics for government relations, went public in 2022), and Trustible (a leading technology provider of responsible AI governance).
About the DS/CSS seminar series
The University of Michigan School of Information’s Data Science/Computational Social Science seminar series brings together a vibrant and diverse community of scholars whose cutting-edge research in information science, computer science and the social sciences aims to broaden our understanding of important social and technological issues.
Organizers for the fall 2024 series are UMSI assistant professors Paramveer Dhillon and Sabina Tomkins.