Political Narratives in Public Discourse: Theory, Detection, and Impact

Since the narrative turn in social sciences, questions about how narratives are formed, how they impact, and how they shape political and public spaces have become central across various disciplines, including political science, sociology, psychology, cognitive sciences, communication studies, history, and more recently, computational social sciences. The purpose of this workshop is to bring together experts from these and other fields to discuss the latest research and to exchange experiences and knowledge on this important topic.

The workshop will cover various aspects of political narratives, including their theoretical foundations, the latest methods for their detection and analysis, and their impact on society and democratic processes. The role of narratives in public discourse, their transformation due to the impact of social media and AI, their influence on political attitudes, belief networks, cultural memory and collective identity, polarization, trust, and uncertainty, will be discussed.

The workshop will feature several panels with invited speakers, providing an interdisciplinary forum for reflecting on the influence of narratives in the context of the new transformation of the public sphere and their potential role as facilitators of new democratic practices.

Several important questions will be addressed at the workshop, including but not limited to:

Defining and Understanding Political Narratives: How do we pinpoint the role of political narratives within public or political discourse? What theoretical backgrounds can be used to understand the impact of narratives on political processes? How do psychological and emotional factors shape and impact political narratives? And vice versa. How do political narratives shape individual and collective identity, attitudes and beliefs? How are political narratives transmitted and transformed across generations, e.g. narratives of collective memories and political and societal narratives of integration? How do social media and AI algorithms impact the transformation of political and societal narratives?

Computational Techniques and Tools: What techniques and tools are the most effective for identifying political narratives in public and semi-public spaces, such as social media platforms? How can AI assist in analyzing large datasets of political narratives? How can we improve the accuracy and reliability of narrative analysis tools? What are the best practices for combining theoretical and empirical approaches in studying political narratives? How can further research contribute to a better understanding of the functions of political narratives?

Impact on Society and Democratic Processes: How do political narratives contribute to social polarization? How do political narratives impact the level of trust in democratic institutions and in interpersonal communication? How can political narratives create or reduce feelings of uncertainty in society? What role do political narratives play in the processes of deliberation? How do political narratives impact the quality and outcomes of deliberative processes in terms of seeking common ground? How can new narratives facilitate the transformation of existing political frameworks at national and transnational levels?

What are the future directions and challenges in political narrative research?

These questions will guide discussions at the workshop and encourage an in-depth exploration of the role of political narratives in public discourse and their impact on democratic processes.

The workshop is part of the SoMe4Dem project that aims at understanding the causal mechanisms underlying the impact of social media platform practices of digital citizenship.

The workshop starts Monday, November 11, 2024 at 01:30pm and ends Wednesday, November 13, 2024 at 01:00pm

Program

13:30 - 14:00	Registration & Welcome Coffee
	Concepts for defining and understanding narratives
14:00 - 14:30	Eckehard Olbrich (MPI MiS) Welcome and Introduction
14:30 - 15:00	Armin Pournaki (MPI MiS) Narratives and Narratology - an attempt at an overview
15:00 - 15:30	Amy Weissenbach (Columbia University) Pathways to precarity: Modeling narrative accounts of poverty and despair in America, 2006-2016 In June 2006, Warren Buffett announced he would give away the great majority of his wealth in his lifetime. Soon he began to receive letters, most of them appeals for assistance. Buffett gave the letters to his sister, Doris, who proceeded to read and respond to the writers, sometimes offering help in the form of small grants and in-kind gifts. The letters continued to pour in by the thousands over the following decade. In this article we examine 10,058 personal appeals received by the Buffetts between 2006 and 2016, from individuals and households across the United States. Most of them contain narrative accounts of who the writers are, how and why they fell into trouble, and how they think the Buffetts might help get them on track. The letters present a fresh opportunity to address questions of longstanding interest to sociologists, about how pathways to financial insecurity unfold in time, as understood and communicated by those who experience them. We develop methods to identify typical pathways associated with stories focused on different kinds of adversity. To do this, we extract narrative sequences from letters and represent them as network graphs. We then reduce these graphs to relationships among forms of adversity in the author’s story in order to identify ideal typical pathways associated with stories focused on different topic areas (e.g. housing insecurity, abuse, medical issues, educational challenges). In doing so we lay the groundwork for an examination of how pathways to precarity unfold.
15:30 - 16:00	Coffee Break
16:00 - 16:30	David Tuckett (University College London \| European University Institute) Knowledge with and without representation: how and why narratives not facts coordinate collective action In the talk I will discuss the need for the term narrative to be restricted to the meaning it had in the “narrative turn” (rather than the very strange homemade definitions in, for example, economics) and try to talk about the core biological, social and psychological characteristics of narrative and their role in collective action.
16:30 - 17:00	Thierry Poibeau (Lattice, CNRS,ENS-PSL) Computing the Story: Novel Narratives Decoded Seventy years of NLP research has led to today’s hype around large language models, but semantic and syntactic parsing have been integral from the start. Early techniques were inadequate, but LLMs (and improved computing power) have changed the situation. Now, we can parse and analyze massive corpora, like the web, with high accuracy. How can these advancements be applied in sociology? Do we still need traditional parsing, or should we rely on LLMs’ abstract representations, for the analysis of narratives for example? And why do sociologists seem fixated on SVO triplets, a concept linguists rarely use? These are the questions I will explore in my talk—though answers may be limited, it may open up discussions.
17:00 - 18:00	Discussion (Mod.: Tom Willaert) - What are narratives and what is the narrative turn in social sciences? What is specific for political narratives?
19:30 -	Conference Dinner at Chinas Welt (individual payment)

09:00 - 09:30	Welcome Coffee
	Formal approaches to narratives - computational tools and models
09:30 - 10:00	Joshua Introne (Syracuse University) Narrative Structure as a Predictor of Information Diffusion In this talk, I'll present emerging findings from a large-scale online study examining how narrative structure and social influence jointly shape the diffusion of narrative information. Our experimental framework takes the form of a multiplayer game where participants collaborate in a social network to reconstruct a narrative by selectively adopting information items from their neighbors. Drawing on established psychological models of narrative coherence and work in information diffusion, we explore how individuals' prior knowledge influences their adoption of new narrative elements. I will focus on two key preliminary findings. First, our analysis reveals that a model of "narrative intensity"-based on an individual's previously adopted story elements-strongly predicts subsequent adoption decisions. This suggests that people actively seek narrative coherence when constructing stories, even in social contexts. Second, we uncover an unexpected interaction between social influence and individual adoption tendencies: while social pressure appears to motivate hesitant adopters to embrace new information, it paradoxically discourages early adopters from incorporating items with low narrative intensity into their story constructions. These findings contribute to our understanding of how narrative structures shape information diffusion processes, with potential implications for how stories-from news to misinformation-spread through social networks. As this research is ongoing, I look forward to engaging with the audience to deepen our collective understanding of these phenomena and explore their broader implications.
10:00 - 10:30	Tom Willaert (Vrije Universiteit Brussel), Armin Pournaki (MPI MiS) A graph-based approach to extracting narrative signals from text We propose a graph-based formalism and machine-guided method for extracting, representing, and analyzing selected narrative signals from digital textual corpora, based on Abstract Meaning Representation (AMR). The formalism and method introduced here specifically cater to the study of \textit{political} narratives that figure in texts from digital media such as archived political speeches, social media posts, transcripts of parliamentary debates, and political manifestos on party websites. We conceptualize these political narratives as a type of ontological narratives: stories by which actors position themselves as political beings, and which are akin to political ``worldviews'' in which actors present their normative vision of the world, or aspects thereof. We approach the study of such political narratives as a problem of information retrieval: starting from a textual corpus, we first extract a graph-like representation of the meaning of each sentence in the corpus by means of the Abstract Meaning Representation (AMR) formalism. Drawing on transferable concepts from narratology, we then apply a set of heuristics to filter these graphs for representations (subgraphs) of 1) actors, 2) the events in which these actors figure, and 3) traces of the perspectivization of these events. We approach these references to actors, events and instances of perspectivization as core narrative signals that initiate a further analysis by alluding to larger political narratives. By means of a case study of State of the European Union addresses (2010-2023), we demonstrate how the formalism can be used to inductively surface signals of political narratives from public discourse.
10:30 - 11:00	Jean-Philippe Cointet (SciencesPo médialab Paris) The Late Replacement
11:00 - 11:30	Coffee Break
	Narratives and belief systems
11:30 - 12:00	Lara Verheyen (Vrije Universiteit Brussel) The Candide Model: How Narratives Emerge Where Observations Meet Beliefs In this talk, I will discuss the Candide model as a computational architecture for modelling human-like, narrative-based language understanding. The model starts from the idea that narratives emerge through the process of interpreting novel linguistic observations, such as utterances, paragraphs and texts, with respect to previously acquired knowledge and beliefs. Narratives are personal, as they are rooted in past experiences, and constitute perspectives on the world that might motivate different interpretations of the same observations. Concretely, the Candide model operationalises this idea by dynamically modelling the belief systems and background knowledge of individual agents, updating these as new linguistic observations come in, and exposing them to a logic reasoning engine that reveals the possible sources of divergent interpretations. I will introduce the foundational ideas and a proof-of-concept implementation.
12:00 - 12:30	Jan Babnik (IRRIS Insitute), Jure Koražija (IRRIS Insitute) Analyzing Social Media's Impact on Commemorative Narratives with a Large Language Model Cultural memories are shared memories, narratives, and representations of the past that are collectively held, transmitted, and preserved within a cultural group or society. They help shape a group's identity and provide a sense of continuity and connection across generations. We present a comprehensive method for conceptualizing and measuring the impact of social media on the transformation of cultural memories. The conceptualization is based on changes in patterns of deliberation, while measurements are conducted on textual analysis of social media using machine learning approaches. Our hypothesis is that social media transform discussions on significant historical events by altering deliberative dynamics and amplifying the polarizing function of commemorative narratives, which ultimately undermines the coherence of the public sphere. To investigate how the deliberative patterns are being transformed, we developed a computational approach grounded in the theory of deliberative democracy. The specific theoretical model is Greimas' actantial model which is supported with natural language processing tools to extract relevant information from social media. We first automatically identify narratives associated with specific commemorations. These narratives are then analyzed in terms of legitimization in decision-making patterns specific to different democratic models (illiberal, liberal, epistocratic, participatory, deliberative, radical, and algorithmic) using large language models (LLMs). Next, we apply LLMs to identify tailored deliberative democracy indicators that measure the quality and goals of discussions on commemorations, emphasizing a summative over an additive approach in deliberative democracy studies. The paper motivates the use of various language technologies and tasks, such as LLMs, topic modeling, and sentiment analysis, to automatically extract relevant information throughout the analysis. The approach is demonstrated using the Slovenian commemoration of the Day of Resistance as a case study. The results show the divisive nature of this commemoration and the transformations of cultural memories associated with it in social media relative to more established sources, such as newspapers and books. We underscore the universal applicability of the demonstrated approach, including the potential to develop an automated pipeline for analyzing debates on divisive collective memories in various national and international contexts. Special attention is given to identifying points of conflict and establishing areas of common ground through computational analysis.
12:30 - 13:00	Felix Victor Münch (Leibniz-Institut für Medienforschung \| Hans-Bredow-Institut) Context Matters: Time as a Parameter in Explorable Hierarchical Topic Modeling in Large-Scale Cross-Platform Online Discourses In times of several global crises at once, such as climate change, pandemics, and armed conflicts with global repercussions it seems a daunting task to gain a general overview of the most relevant parts of public discourse. Superficially unrelated discourses influence each other on multiple levels of scopes and timescales. Therefore, this endeavour necessitates approaches that can simplify the inherent complexity of these topics at a top-level while allowing data analysts to choose where to dig deeper. Not only the interactive explorability of topic relations and sub-topic structures is key, but ever-changing topics and crises necessitate the introduction of time as a topic-defining parameter. Therefore, we developed a combination of methods, leaning on the embedding-clustering sequence of BERTopic, but based on semantic similarity networks, augmented with a time-parameter, and hierarchical network community detection methods. In this contribution we describe our method and its validation by means of a large-scale test case, namely the German-language cross-platform discourse around climate change on Facebook, Instagram, Twitter, and Telegram from 2019 to 2023. Furthermore, we present how the method supported analysts in discovering known and novel dis- and misinformation narratives. Lastly, we will discuss how not only time but also other contextual data, such as the behavioural history of content producers, could augment and improve the results of this and similar natural language processing methods in the future.
13:00 - 14:00	Lunch Break
	Applications and case studies
14:00 - 14:30	Carlo Romano Marcello Alessandro Santagiustina (Sciences Po médialab) Explaining one’s identity: ideological and national variation in self-representation in social media This study examines how followers of Members of Parliament (MPs) from eight EU member states express their socio-political identities through social media. In contrast to existing approaches that focus on posts and threads, our method highlights how users' digital self-presentation and group identity cues conveyed through bios reflect broader socio-political participation and discourse within the EU public sphere. We treat these bios not only as proxies for the stances, views, and priorities of politically engaged European citizens but also as expressions of how groups construct and communicate their socio-political identities through narratives, in relation to wider socio-political discourse and participation. Our goal is to explore both commonalities and differences in the topic proportions across EU countries, and in topic contents used by left- and right-wing users to express their identities. To achieve this, we combine a Structural Topic Model (STM) with attitudinal embeddings, drawing our topic seed words from the Manifesto Project's annotated data and ontology. By grounding our topics in this well-established and systematised framework for analysing political discourse, we enhance the scope and interpretability of our results. This methodology allows us to map how discourse in users' bios varies by political orientation and other dimensions of interest, revealing how European group identity narratives are socialised and co-constructed in online environments.
14:30 - 15:00	Marc Keuschnigg (Leipzig University) Measuring concept interpretations with seeded topic models Social scientists are discussing the need for more formal ways to extract meaning from digital text. We focus attention on an old school model---the seeded topic model---that allows domain knowledge to be infused into the computational learning of meaning structures. Seed words help crystallize topics around known concepts, while utilizing topic models’ functionality to identify associations in text based on word co-occurrences. We implement a scalable version of that model to estimate a concept’s shared interpretation (or framing) via its associations with other frequently co-occurring topics. In a case study, we extract longitudinal measures of the understanding of immigration from a corpus of millions of Swedish newspaper articles from the period 1945–2019. We infer turning points that partition the immigration discourse into meaningful eras and locate Sweden’s era of multicultural ideals that coined its tolerant reputation.
15:00 - 15:30	Coffee Break
15:30 - 16:00	Veronika Batzdorfer (Karlsruhe Institute of Technology), Sven Banisch (Karlsruhe Institute for Technology) How Social Feedback, Habits, and Norms Shape Conspiracy Engagement in Online Communities This study examines the motivational mechanisms behind habitual online engagement, specifically focusing on conspiracy theories (CTs). Given the routine nature of much of human behavior, understanding how varying susceptibility to rewards-triggered by contextual cues-affect the proclivity to CT content is crucial. Using a framework from evolutionary psychology, we analyze topic-agnostic CT narratives both at the post level and community norms at the subverse level. We hypothesize that (1) as users become more habitual, they become less sensitive to rewards (e.g. upvotes), and habit strength negatively predicts posting latency; (2) social feedback increases engagement in CT-related spaces; (3) different narrative aspects of a user’s perspective on conspiracies such as individual trait-based (secrecy and pattern) as well as group-identification (action, actor, threat) can be indicative of CTs and enhance content engagement. Our study reveals that user behavior may be influenced by both reinforcement learning (where users adjust their posting based on feedback) and norm learning mechanisms (where they adapt to the norms of multiple communities).
16:00 - 16:30	Marc Tuters (University of Amsterdam) Global Digital Culture War
16:30 - 17:30	Discussion (Mod.: Armin Pournaki) - What are the main challenges for unsupervised narrative detection? How do narratives shape belief systems and how can we model that? For which research questions do we need narratives to address them empirically?
19:30 -	Conference Dinner at Auerbachs Keller (individual payment)

09:00 - 09:30	Welcome Coffee
	Narratives and Democracy
09:30 - 10:00	Marjan Horvat (IRRIS Institute) Political Narratives, Social Media and Deliberative Democracy: Collective Memory Transformation as a Case Study Historically, narrative has always been an important element of political grammar and discourse. Specific interpretations of complex topics shape political and social spaces and connect different layers of society with coherent ideational frameworks. From the perspective of deliberative democracy's regulatory ideals, which emphasize reason and good argumentation in decision-making, the issue of narratives poses a particular challenge. On the one hand, theory acknowledges the importance and role of personal narratives such as storytelling and testimonies, as supplementary forms of argument justification. On the other hand, at the systemic level, the role of narratives is still under-explored, as deliberative democracy is designed as a form of progressive, reason-based Enlightenment intervention in existing decision- and policy-making patterns, effectively bypassing pre-existing embedded political and societal narratives. In this paper, we argue that the scope and impact of this deliberative intervention are limited due to the underexplored impact of epistemic modalities on embedded patterns of decision-making. We support our thesis with a sociological critique of deliberative democracy, highlighting the importance of socio-political context in opinion formation and political decision-making, as well as the role of embedded values, cultural and historical factors, the specific configuration of the social contract and evolution of public deliberation on a macro scale. This perspective significantly alters the potential for enacting deliberative democracy standards in a given society. We posit that deciphering the embedded narrative patterns that constitute socio-political discourse of legitimation is a prerequisite for facilitating norms of deliberative democracy. Computational tools can assist in this task substantially. From this standpoint, we explore the consequences of transforming and creating new narratives on social media, impacting the underlying structural solidarity of the public sphere in a given society. We particularly focus on cases of collective memory narratives, which can serve as integrative or polarizing factors in socio-political substratum. Finally, we present a theoretical framework for conceptualizing the impact of transformed collective memory narratives on the contemporary political and societal landscape through different theory- and narrative-based legitimacy models, with a specific focus on the transformation of existing political registers and narratives under the framework of deliberative decision-making.
10:00 - 10:30	André Bächtiger (Universität Stuttgart) Narratives and Deliberation First generation deliberative theories (Bächtiger et al. 2018) have frequently put a stress on rational argument as a key goal of deliberation. This has raised criticisms from difference democrats and feminists, arguing that rational argument including logical deduction and general principles excludes specific social groups (especially disadvantaged citizens and cultural minorities; Sanders 1997; Young 2001). Instead, a focus on narratives might be better apt to promote inclusionary goals of democratic deliberation (Polletta and Gardner 2018). Second-generation theories of deliberation have corrected this by including narratives in the conceptual apparatus of desirable deliberative standards. Drawing from empirical research in citizen forums and parliamentary committees (Gerber et al. 2018; Bächtiger et al. 2010), I will show that narratives and rational argument do not constitute opposites but fulfil different and complementary functions in public and political discourse. I will end with a systemic perspective on narratives and deliberation, arguing that “rational argument” in democratic societies is better understand as an “emergent” property at the level of public discourse rather than as an immediate outcome of concrete conversations and debates.
10:30 - 11:00	Discussion (Mod.: Eckehard Olbrich) - What roles do narratives play in democratic discourse? How is the relationship between (political) narratives and rational argumentation? What has changed with the rise of social media in this regard?
11:00 - 11:30	Coffee Break
	Future work and projects
11:30 - 12:00	Lisa Oswald (Max Planck Institute for Human Development), Philipp Lorenz-Spreen (MPIB) Why do people share content? Ideas for future work on narratives on the data from the field experiments. Twitter data (Philipp Lorenz-Spreen): What motivates people to share and create content online? And do those motives change for different narratives? In real time, we linked each of N=2,762 individual posts (retweets and newly created content) with the self-reported motives from a sample of N=137 highly active US Twitter users over the course of one week. We also captured their total activity of N=48,419 posts over 10 weeks (March-May 2022). This labeled data provides a unique opportunity to link the motives of the users with the narratives contained in the content itself. Potentially, certain groups of narratives or overarching themes have common motives for the people who contribute to and disseminate them. Reddit data (Lisa Oswald): Counterfactuals are difficult to find when analysing narratives in social media. In the course of an online field experiment on Reddit, we elicited discussions on 20 different political issues and let discussions unfold in parallel across 6 closed, private online communities. In this session, we provide a brief overview over the existing dataset and would like to propose it as basis for future analyses with a focus on narratives.
12:00 - 12:45	Discussion - Future projects and collaborations
12:45 - 13:00	Closing Remarks

Organizers

Jürgen Jost

Max-Planck-Institut für Mathematik in den Naturwissenschaften

Eckehard Olbrich

Max Planck Institute for Mathematics in the Sciences

Marjan Horvat

IRRIS Institute

Tom Willaert

Vrije Universiteit Brussel

Armin Pournaki

Max Planck Institute for Mathematics in the Sciences

Administrative Contact

Katharina Matschke

Max Planck Institute for Mathematics in the Sciences Contact via Mail