Ninth International Workshop on Narrative Extraction from Texts
held in conjunction with the 48th European Conference on Information Retrieval
For eight years, Text2Story has gathered researchers on narrative understanding, detecting, representing, and reasoning over stories in text. Recent progress and discussions have been driven by Transformers and Large Language Models (LLMs), which are increasingly central to the field and have reshaped approaches to narrative extraction and interpretation. Despite rapid progress, challenges persist in fine-grained narrative structure, dialogue and multimodal narratives, robustness to domain shift, and evaluation beyond surface accuracy. The ninth edition of the workshop will spotlight the growing convergence of Information Retrieval (IR) with Transformers/LLMs and agentic AI. By centering these IR-aligned problems, Text2Story aims to advance both fundamental insights and practical applications in narrative search, understanding, and generation
Research papers submitted to the workshop should advance the understanding, modeling, and application of narratives across text and multimodal contexts. We encourage work on narrative extraction, representation, analysis, and generation, particularly in connection with Transformers, LLMs, and Agentic AI, as well as research that bridges narrative understanding and information retrieval. We encourage the submission of high-quality and original submissions covering the following topics and contributions focused on low and medium-resource languages.
Overall, the workshop aims to: (1) raise awareness within the Information Retrieval (IR) and NLP communities about the challenges and opportunities in narrative extraction, comprehension, and generation; (2) bridge the gap between academic research, industrial practice, and applied systems involving LLMs and Agentic AI for narrative tasks; (3) foster discussion on new methods, recent advances, and emerging challenges in narrative understanding, reasoning, and evaluation; (4) share experiences from projects, case studies, and applications structured around key research questions in narrative representation and generation; (5) explore the impact of automation, LLMs, and agentic systems on the creation and interpretation of narratives; (6) encourage reflection on both successful and unexpected outcomes in narrative research, contributing to a deeper understanding of limitations and open problems.
We expect contributions from researchers addressing all aspects of narrative extraction, representation, analysis, and generation. This includes the detection and formal modeling of events, participants, and their temporal and causal relations, as well as methods for reasoning and orderinge. Submissions that explore narrative comprehension and interpretation, such as the analysis or evaluation of LLM-generated narratives, are particularly encouraged.
We also invite work on innovative ways to present and interact with narrative information, including automatic timeline construction, multimodal summarization, and narrative visualization. Research tackling misinformation, bias, and verification of extracted facts, along with the development of datasets, annotation schemas, and evaluation methodologies, is highly valued. Contributions addressing low and medium-resource languages, multilingual and cross-lingual settings are especially welcome.
Building on these themes, the field now faces pressing questions that can guide authors in shaping their submissions. How can LLMs and Agentic AI be leveraged for more coherent, grounded, and explainable narrative understanding? What strategies enable the integration of multimodal content—text, image, audio, video—into unified and trustworthy narratives? How can models dynamically adapt to evolving narratives across domains, genres, and languages? What methods best support the evaluation and benchmarking of narrative systems, from extraction to generation and reasoning? How can we ensure transparency, fairness, and robustness in narrative systems that may influence real-world perceptions and decisions? To what extent can collaboration between human experts and automated systems enhance interpretability, cultural awareness, and inclusivity in narrative understanding?
Original and high-quality unpublished contributions to the theory and practical aspects of the narrative extraction task. Full papers should introduce existing approaches, describe the methodology and the experiments conducted in detail. Negative result papers to highlight tested hypotheses that did not get the expected outcome are also welcomed.
Unpublished short papers describing work in progress; position papers introducing a new point of view, a research vision or a reasoned opinion on the workshop topics; and dissemination papers describing project ideas, ongoing research lines, case studies or summarized versions of previously published papers in high-quality conferences/journals that is worthwhile sharing with the Text2Story community, but where novelty is not a fundamental issue.
Unpublished papers presenting research/industrial demos; papers describing important resources (datasets or software packages) to the Text2Story community;
Papers must be submitted electronically in PDF format through Easy Chair . All submissions must be in English and formatted according to the one-column CEUR-ART style with no page numbers. Templates, available in either LaTeX or ODT (LibreOffice) format, can be found in the following zip folder. Do not use Word for the ODT template. CEUR requires the use of the Libertinus font family. Instructions on installing these fonts are found in the ODT template. There is also an Overleaf page for LaTeX users. Please also note that your work must include a Declaration on Generative AI to comply with the CEUR-WS GenAI policy. Both the LaTeX and LibreOffice templates already include this required section.
IMPORTANT: Please include between brackets the type of submission (full; negative results; work in progress; demo and resource; position; dissemination) in the paper title.
Papers submitted to Text2Story 2026 should be original work and different from papers that have been previously published, accepted for publication, or that are under review at other venues. Exceptions to this rule are "dissemination papers". Pre-prints submitted to ArXiv are eligible.
All papers will be refereed through a double-blind peer-review process by at least two members of the programme committee. We plan to publish the workshop proceedings as a CEUR volume, which will ensure they are indexed by DBLP and made available online via open access. Authors are therefore responsible for ensuring that their submissions do not infringe on any existing publication rights.
Abstract:
Extracting event knowledge from unstructured text is a well-known challenge in natural language processing (NLP) and is particularly difficult when dealing with fiction. Subtext, instead of explicit information, and figurative style in fictional narratives complicate event extraction. Recent advances in Large Language Models (LLMs) have improved performance in various NLP tasks. However, their effective ness in extracting events from fiction remains underexplored. In this talk, I present an evaluation of different techniques to extract narrative information from fiction, including the prompting of open-weights LLMs and the semantic parsing via Abstract Meaning Representation (AMR).
This work has been done in the context of the GOLEM project (“Graphs and Ontologies for Literary Evolution Models”).
Bio: Federico Pianzola is Assistant Professor of Computational Humanities at the University of Groningen (Netherlands), where he coordinates the Master’s in Digital Humanities. He is the Principal Investigator of the ERC Starting Grant project GOLEM and previously completed a Marie Sklodowska-Curie Global Fellowship at the University of Milan-Bicocca (Italy) and Sogang University (South Korea). He is member of the scientific advisory board of OPERAS (the Research Infrastructure supporting open scholarly communication in the social sciences and humanities in the European Research Area) and president of the International Society of the Empirical Study of Literature (IGEL). He is also managing editor of the journal Scientific Study of Literature and he is on the editorial board of three other journals: the Journal of Computational Literary Studies, the Journal of Cultural Analytics, and the Korean Journal of Digital Humanities. His most recent book Digital Social Reading: Sharing Fiction in the 21st Century has been published by MIT Press in January 2025.
Abstract:
As large language models and AI agents define the current era of enterprise AI adoption across business, government, and public sectors, organizations face a critical challenge: how to systematically evaluate and monitor these systems at scale while maintaining meaningful human oversight. This keynote addresses the realities encountered in real projects and the urgent need to comprehensively review evaluation methodologies for generative AI and agentic systems, from narrative extraction to generation and reasoning, with particular emphasis on the critical role of human-in-the-loop approaches in this context.
We begin by examining why traditional evaluation approaches and academic benchmarks fall short for generative AI in enterprise settings. While LLM benchmarks may show impressive scores on standardized tests, they often fail to predict real-world performance where outputs are diverse, context-dependent, domain-specific, and subject to business constraints. The talk presents a framework for designing domain-specific evaluation workflows that combine quantitative metrics, human judgment, emerging techniques such as LLM-as-a-Judge approaches and its challenges. We explore practical methods for collecting and analyzing evaluation signals from real-world usage while addressing the most common implementation difficulties.
A central focus is the transition from evaluation-driven research in laboratory settings to evaluation-driven development in industry practice. We discuss how to approach systematic continuous improvement of generative AI systems, including synthetic data generation, domain-specific metrics, data segmentation, and related tasks to enable development lifecycles with iterative improvements.
By the end of this keynote, attendees will gain an industry perspective on how to design, implement, and operationalize evaluation methodologies that enable the development and deployment of trustworthy generative AI and agentic systems.
Bio: Arian Pasquali is an Applied AI Researcher at Orq.ai, an AI Engineering & Evaluation Platform focused on Generative AI and AI agents for enterprises, based in Amsterdam. He specializes in NLP and LLMs, with focus on evaluation and annotation workflows applied to AI agents. During the last years he has been bridging the gap between research and practical applications across government, education, and enterprise domains. His current research interests focus on designing processes for the systematic evaluation of large language models and agentic architectures. He has published research in top-tier conferences on keyword extraction, temporal narrative generation, and interactive systems for information retrieval. Arian is also an active contributor to the academic community, where he has served as a reviewer and chair for major international conferences, including ECIR and ECML-PKDD. His contributions have been recognized with several awards, including the World Summit Award 2019 for Government and Citizen Engagement, as well as paper awards at ECIR.
Text2Story 2026 will be held at the 48th European Conference on Information Retrieval (ECIR'26) in Delft, The Netherlands.
Registration at ECIR 2026 is required to attend the workshop (don't forget to select the Text2Story workshop).
This work is financed by National Funds through the FCT - Fundação para a Ciência e a Tecnologia, I.P. (Portuguese Foundation for Science and Technology) within the project StorySense, with reference 2022.09312.PTDC (DOI 10.54499/2022.09312.PTDC).
