Program

Chair: Gabriella Lapesa

• Beyond the pale: Assessing prevalence and contents of extremist speech in LLM training data
  Dmitry Nikolaev, Ashley A. Mattheis

• Effects of Answer Format Variation on Gender Bias in Large Language Models
  Ksenia Merzlyakova, Sebastian Padó, Franziska Weeber

• Heroes, Villains, and Presidential Administrations: Evaluating LLM-Based Narrative Policy Frameworks in Costa Rican News
  Valentina Tretti Beckles, Adrian Vergara Heidke

• Making Political Text Scaling Comparable: Infrastructure and Hyperparameter Sensitivity for 17 Algorithms
  Patrick Parschan

• [ONLINE] LLMs as Strategic Actors: Behavioral Alignment, Risk Calibration, and Argumentation Framing in Geopolitical Simulations
  Veronika Solopova, Viktoriia Skorik, Maksym Tereshchenko, Alina Haidun, Ostap Vykhopen

• [ONLINE] Expanding the Utility of Protest Event Datasets with Automated Variable Extraction from Event Summaries
  Dominik Allen, Christian Oswald

• A systematic review of LLM-based survey response simulation
  Hao-Ting Chan, Dennis Assenmacher, Sebastian Stier, Claudia Wagner

• Centrist Sycophancy? Examining Sycophantic Responses to Extreme Positions in Large Language Models
  Philipp Heinrich

• From Bias Mitigation to Bias Negotiation: Governing Identity and Sociocultural Reasoning in Generative AI
  Zackary Dunivin

• Graph-Augmented Digital Twins as Decision-Support Tools for Political Organizations
  Jairo Gudiño, Joshua C. Yang

• Guided self-reflection in qualitative coding: Improving automated text annotation through secondary LLM review
  Zackary Dunivin, Mobina Noori, Seth Frey, Curtis Atkisson

• Human Label Variation in Ideological Coherence Judgements of Political Speech
  Karolina Zaczynska, Julian Schlenker, Ines Rehbein, Simone Paolo Ponzetto

• LLMs in scientific research: Who is using open-weight models?
  Zackary Dunivin

• Methods Hub – a collaborative platform for exploring, learning, and sharing computational methods in the social sciences
  Julia Romberg, Christina Viehmann, Johannes Kiesel, Arnim Bleier, Chung-hong Chan, Raniere Gaia Costa da Silva, Ahrabhi Kathirgamalingam, Taimoor Khan, Stephan Linzbach, Fakhri Momeni, Felix Münch, Ran Yu, Stefan Dietze, Claudia Wagner

• Parties and Prejudice: Evaluating Social Bias in Large Language Models Adapted to German Political Parties
  Kristina Thieme, Maximilian Spliethöver, Henning Wachsmuth

• Promptology: A Large-Scale Systematic Analysis of Prompt Elements and Contextual Factors in Argument Generation
  Maximilian Maurer, Ana Lisboa Cotovio, Matteo Melis, Julia Romberg, Aldo Costa, Gabriella Lapesa

• Systematizing Fact-Checking Labels
  Leonie Uhling, Zehra Melce Hüsünbeyi, Tatjana Scheffler

• The Politics of Feelings in U.S. Congressional Speech
  Segun Aroyehun

• The Total Simulated Survey Error (TS2E) Framework: Evaluating LLM-Generated Survey Responses
  Indira Sen, Georg Ahnert, Leah von der Heyde, Bernd Weiss, Jana Lasser, Markus Strohmaier

Chair: Christopher Klamm

• [ONLINE] Are Large Language Models reliable counter-narrative generators? An Evaluation Across Languages
  Mykhailo Pavliuk, Veronika Solopova, Arthur Hilbert, Max Upravitelev, Vera Schmitt

• Modeling Parliamentary Interaction in the German Bundestag via Interjections and Policy Topics
  Hannah Mathilde Steinbach, Manfred Stede, Christoph Maximilian Abels

• Comparing Architectures for Supervised Political Scaling
  Anna Golub, Sebastian Padó

• Modeling Speech Acts in Political Manifestos: A Bayesian Analysis of Political Communication
  Klaus Schmidt, Andreas Niekler, Manuel Burghardt

• BundesTube: A German Political YouTube Corpus with Zero-Shot Transcript Restoration
  Zaher Alkaei, Arthur Hilbert, Ronja Memminger

Social class is a fundamental dimension of social stratification, permeating most aspects of our lives, affecting how we communicate, how we engage with and experience technology, and how we perceive one another. Yet, it remains understudied in NLP and computational social science. In this talk, I reflect on the challenges of studying the interaction between social class and language technologies, and propose directions for a broader research agenda.

Social class is multidimensional, relational, culturally situated, and constantly evolving. From a computational perspective, this creates challenges in defining and measuring class, as well as in accessing appropriate data, tools, and participants. These challenges are further complicated by cultural differences in how class is understood and expressed, and by the limitations of traditional measures for capturing contemporary forms of social stratification.

Drawing on my work on NLP, AI adoption, and classist discourse, I discuss these challenges and consider how computational social science can move towards better data, measurement, tools, and methods for studying social class.

Amanda Cercas Curry is a Lecturer in the Department of Social Statistics at the University of Manchester, researching safety and ethics in language technologies. Her work asks what happens when AI systems become participants in social and emotional interactions. Her interdisciplinary work approaches language technologies and the social worlds in which they are developed and used as mutually constitutive: AI systems are shaped by society, while also shaping how people understand, relate to, and interact with one another. 

Her research spans two connected areas. The first examines the role of emotion in AI, from how systems are built to understand, model, and respond to human emotions to how people perceive and interact with AI as an emotional or socially responsive technology. The second addresses social class as an overlooked dimension of inequality in NLP, including how socioeconomic status shapes people's interactions with AI and how social inequalities influence the performance and benefits people derive from these systems. Her work has been recognised with the Best Social Impact Paper Award at ACL 2025.

Before joining Manchester, she was a Research Scientist at CENTAI and held a postdoctoral position at MilaNLP, Bocconi University. She completed her PhD in Computer Science at Heriot-Watt University.

• Chair: Dennis Assenmacher

• PhD projects

  • Dog Whistling in Political Communication
    Charlotte de Alwis

  • Large Language Models as Synthetic Survey Respondents: Validity, Imputation, and the Role of Digital Data
    Hao-Ting Chan

  • Analyzing Open-Ended Text Data in Survey Research on Political Discontent and Participation
    Anna Joraschek

  • Investigating the Values of Multilingual Large Language Models
    Léo Labat

  • Human-LLM Alignment in Perspective Modelling
    Ana Lisboa Cotovio

  • Data-Driven Fear Speech: Bridging Theoretical Foundations with Computational Resources and Discourse Analysis
    Vigneshwaran Shankaran

  • Beyond Stated Preferences: Evaluating Explicit and Implicit LLM Alignment
    Franziska Weeber

• Program

  • 10:30-10.50: Lightning Talks

  • 10:50-11.30: Poster Session

  • 11:30-11.50: Discussion Round